Merge commit 'origin/gallium-0.2' into gallium-xlib-rework

147 files changed, 4575 insertions, 2553 deletions

diff --git a/src/egl/drivers/glx/egl_glx.c b/src/egl/drivers/glx/egl_glx.c
index c93e568cffd..8ba70ba7a60 100644
--- a/src/egl/drivers/glx/egl_glx.c
+++ b/src/egl/drivers/glx/egl_glx.c
@@ -578,7 +578,7 @@ GLX_eglMakeCurrent(_EGLDriver *drv, EGLDisplay dpy, EGLSurface d,
       return EGL_FALSE;
 
 #ifdef GLX_VERSION_1_3
-   if (!glXMakeContextCurrent(disp->Xdpy, GLX_dsurf->drawable, GLX_rsurf->drawable, GLX_ctx->context))
+   if (!glXMakeContextCurrent(disp->Xdpy, GLX_dsurf ? GLX_dsurf->drawable : 0, GLX_rsurf ? GLX_rsurf->drawable : 0, GLX_ctx ? GLX_ctx->context : NULL))
 #endif
       if (!glXMakeCurrent(disp->Xdpy, GLX_dsurf ? GLX_dsurf->drawable : 0, GLX_ctx ? GLX_ctx->context : NULL))
          return EGL_FALSE;
diff --git a/src/gallium/auxiliary/gallivm/Makefile b/src/gallium/auxiliary/gallivm/Makefile
index c3f7bfba93b..5a96d94ec37 100644
--- a/src/gallium/auxiliary/gallivm/Makefile
+++ b/src/gallium/auxiliary/gallivm/Makefile
@@ -66,12 +66,12 @@ depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(INC_SOURCES)
 
 gallivm_builtins.cpp: llvm_builtins.c
 	clang --emit-llvm < $< |llvm-as|opt -std-compile-opts > temp1.bin
-	(echo "static const unsigned char llvm_builtins_data[] = {"; od -txC temp1.bin | sed -e "s/^[0-9]*//" -e s"/ \([0-9a-f][0-9a-f]\)/0x\1,/g" -e"\$$d" | sed -e"\$$s/,$$/};/") >$@
+	(echo "static const unsigned char llvm_builtins_data[] = {"; od -txC temp1.bin | sed -e "s/^[0-9]*//" -e s"/ \([0-9a-f][0-9a-f]\)/0x\1,/g" -e"\$$d" | sed -e"\$$s/,$$/,0x00};/") >$@
 	rm temp1.bin
 
 gallivmsoabuiltins.cpp: soabuiltins.c
 	clang --emit-llvm < $< |llvm-as|opt -std-compile-opts > temp2.bin
-	(echo "static const unsigned char soabuiltins_data[] = {"; od -txC temp2.bin | sed -e "s/^[0-9]*//" -e s"/ \([0-9a-f][0-9a-f]\)/0x\1,/g" -e"\$$d" | sed -e"\$$s/,$$/};/") >$@
+	(echo "static const unsigned char soabuiltins_data[] = {"; od -txC temp2.bin | sed -e "s/^[0-9]*//" -e s"/ \([0-9a-f][0-9a-f]\)/0x\1,/g" -e"\$$d" | sed -e"\$$s/,$$/,0x00};/") >$@
 	rm temp2.bin
 
 # Emacs tags
diff --git a/src/gallium/auxiliary/gallivm/gallivm.cpp b/src/gallium/auxiliary/gallivm/gallivm.cpp
index 29adeea47de..f4af5cc8ad5 100644
--- a/src/gallium/auxiliary/gallivm/gallivm.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm.cpp
@@ -53,7 +53,7 @@
 #include <llvm/ModuleProvider.h>
 #include <llvm/Pass.h>
 #include <llvm/PassManager.h>
-#include <llvm/ParameterAttributes.h>
+#include <llvm/Attributes.h>
 #include <llvm/Support/PatternMatch.h>
 #include <llvm/ExecutionEngine/JIT.h>
 #include <llvm/ExecutionEngine/Interpreter.h>
diff --git a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
index fcc5c05794a..634bac01507 100644
--- a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
@@ -137,4 +137,4 @@ static const unsigned char llvm_builtins_data[] = {
 0x58,0x85,0x05,0x14,0xbe,0x34,0x45,0xb5,0x21,0x10,0x82,0x23,0x15,0x46,0x30,0x2c,
 0xc8,0x64,0x02,0x06,0xf0,0x3c,0x91,0x73,0x19,0x00,0xe1,0x4b,0x53,0x64,0x0a,0x84,
 0x84,0x34,0x85,0x25,0x0c,0x92,0x20,0x59,0xc1,0x20,0x30,0x8f,0x2d,0x10,0x95,0x84,
-0x34,0x00,0x00,0x00,0x00,0x00,0x00,0x00};
+0x34,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};
diff --git a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
index 93a9748bdb3..1bd00a0c2a6 100644
--- a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
@@ -56,7 +56,7 @@
 #include <llvm/ModuleProvider.h>
 #include <llvm/Pass.h>
 #include <llvm/PassManager.h>
-#include <llvm/ParameterAttributes.h>
+#include <llvm/Attributes.h>
 #include <llvm/Support/PatternMatch.h>
 #include <llvm/ExecutionEngine/JIT.h>
 #include <llvm/ExecutionEngine/Interpreter.h>
diff --git a/src/gallium/auxiliary/gallivm/instructions.cpp b/src/gallium/auxiliary/gallivm/instructions.cpp
index 599975d5add..ee8162efce5 100644
--- a/src/gallium/auxiliary/gallivm/instructions.cpp
+++ b/src/gallium/auxiliary/gallivm/instructions.cpp
@@ -43,7 +43,7 @@
 #include <llvm/Function.h>
 #include <llvm/InstrTypes.h>
 #include <llvm/Instructions.h>
-#include <llvm/ParameterAttributes.h>
+#include <llvm/Attributes.h>
 #include <llvm/Support/MemoryBuffer.h>
 #include <llvm/Bitcode/ReaderWriter.h>
 
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
index d5600fd22da..925e948763e 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.cpp
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
@@ -37,7 +37,7 @@
 #include <llvm/Function.h>
 #include <llvm/Instructions.h>
 #include <llvm/Transforms/Utils/Cloning.h>
-#include <llvm/ParameterAttributes.h>
+#include <llvm/Attributes.h>
 #include <llvm/Support/MemoryBuffer.h>
 #include <llvm/Bitcode/ReaderWriter.h>
 
@@ -206,11 +206,12 @@ llvm::Module * InstructionsSoa::currentModule() const
 
 void InstructionsSoa::createBuiltins()
 {
+   std::string ErrMsg;
    MemoryBuffer *buffer = MemoryBuffer::getMemBuffer(
       (const char*)&soabuiltins_data[0],
-      (const char*)&soabuiltins_data[Elements(soabuiltins_data)]);
-   m_builtins = ParseBitcodeFile(buffer);
-   std::cout<<"Builtins created at "<<m_builtins<<std::endl;
+      (const char*)&soabuiltins_data[Elements(soabuiltins_data) - 1]);
+   m_builtins = ParseBitcodeFile(buffer, &ErrMsg);
+   std::cout<<"Builtins created at "<<m_builtins<<" ("<<ErrMsg<<")"<<std::endl;
    assert(m_builtins);
    createDependencies();
 }
diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
index c11b88af9ec..6b18a68fe66 100644
--- a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
@@ -25,7 +25,7 @@
 #include <llvm/ModuleProvider.h>
 #include <llvm/Pass.h>
 #include <llvm/PassManager.h>
-#include <llvm/ParameterAttributes.h>
+#include <llvm/Attributes.h>
 #include <llvm/Support/PatternMatch.h>
 #include <llvm/ExecutionEngine/JIT.h>
 #include <llvm/ExecutionEngine/Interpreter.h>
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
index 071bc2015c7..53a0e722cff 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -213,8 +213,8 @@ emit_instruction(struct spe_function *p, uint32_t inst_bits)
 
 
 
-static void emit_RR(struct spe_function *p, unsigned op, unsigned rT,
-		    unsigned rA, unsigned rB, const char *name)
+static void emit_RR(struct spe_function *p, unsigned op, int rT,
+		    int rA, int rB, const char *name)
 {
     union spe_inst_RR inst;
     inst.inst.op = op;
@@ -230,8 +230,8 @@ static void emit_RR(struct spe_function *p, unsigned op, unsigned rT,
 }
 
 
-static void emit_RRR(struct spe_function *p, unsigned op, unsigned rT,
-                     unsigned rA, unsigned rB, unsigned rC, const char *name)
+static void emit_RRR(struct spe_function *p, unsigned op, int rT,
+                     int rA, int rB, int rC, const char *name)
 {
     union spe_inst_RRR inst;
     inst.inst.op = op;
@@ -248,8 +248,8 @@ static void emit_RRR(struct spe_function *p, unsigned op, unsigned rT,
 }
 
 
-static void emit_RI7(struct spe_function *p, unsigned op, unsigned rT,
-		     unsigned rA, int imm, const char *name)
+static void emit_RI7(struct spe_function *p, unsigned op, int rT,
+		     int rA, int imm, const char *name)
 {
     union spe_inst_RI7 inst;
     inst.inst.op = op;
@@ -266,8 +266,8 @@ static void emit_RI7(struct spe_function *p, unsigned op, unsigned rT,
 
 
 
-static void emit_RI8(struct spe_function *p, unsigned op, unsigned rT,
-		     unsigned rA, int imm, const char *name)
+static void emit_RI8(struct spe_function *p, unsigned op, int rT,
+		     int rA, int imm, const char *name)
 {
     union spe_inst_RI8 inst;
     inst.inst.op = op;
@@ -284,8 +284,8 @@ static void emit_RI8(struct spe_function *p, unsigned op, unsigned rT,
 
 
 
-static void emit_RI10(struct spe_function *p, unsigned op, unsigned rT,
-		      unsigned rA, int imm, const char *name)
+static void emit_RI10(struct spe_function *p, unsigned op, int rT,
+		      int rA, int imm, const char *name)
 {
     union spe_inst_RI10 inst;
     inst.inst.op = op;
@@ -302,8 +302,8 @@ static void emit_RI10(struct spe_function *p, unsigned op, unsigned rT,
 
 
 /** As above, but do range checking on signed immediate value */
-static void emit_RI10s(struct spe_function *p, unsigned op, unsigned rT,
-                       unsigned rA, int imm, const char *name)
+static void emit_RI10s(struct spe_function *p, unsigned op, int rT,
+                       int rA, int imm, const char *name)
 {
     assert(imm <= 511);
     assert(imm >= -512);
@@ -311,7 +311,7 @@ static void emit_RI10s(struct spe_function *p, unsigned op, unsigned rT,
 }
 
 
-static void emit_RI16(struct spe_function *p, unsigned op, unsigned rT,
+static void emit_RI16(struct spe_function *p, unsigned op, int rT,
 		      int imm, const char *name)
 {
     union spe_inst_RI16 inst;
@@ -326,7 +326,7 @@ static void emit_RI16(struct spe_function *p, unsigned op, unsigned rT,
 }
 
 
-static void emit_RI18(struct spe_function *p, unsigned op, unsigned rT,
+static void emit_RI18(struct spe_function *p, unsigned op, int rT,
 		      int imm, const char *name)
 {
     union spe_inst_RI18 inst;
@@ -348,61 +348,61 @@ void _name (struct spe_function *p) \
 }
 
 #define EMIT_(_name, _op) \
-void _name (struct spe_function *p, unsigned rT) \
+void _name (struct spe_function *p, int rT) \
 { \
    emit_RR(p, _op, rT, 0, 0, __FUNCTION__); \
 }
 
 #define EMIT_R(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA) \
+void _name (struct spe_function *p, int rT, int rA) \
 { \
    emit_RR(p, _op, rT, rA, 0, __FUNCTION__);                 \
 }
 
 #define EMIT_RR(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA, unsigned rB) \
+void _name (struct spe_function *p, int rT, int rA, int rB) \
 { \
    emit_RR(p, _op, rT, rA, rB, __FUNCTION__);                \
 }
 
 #define EMIT_RRR(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA, unsigned rB, unsigned rC) \
+void _name (struct spe_function *p, int rT, int rA, int rB, int rC) \
 { \
    emit_RRR(p, _op, rT, rA, rB, rC, __FUNCTION__);           \
 }
 
 #define EMIT_RI7(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
 { \
    emit_RI7(p, _op, rT, rA, imm, __FUNCTION__);              \
 }
 
 #define EMIT_RI8(_name, _op, bias) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
 { \
    emit_RI8(p, _op, rT, rA, bias - imm, __FUNCTION__);       \
 }
 
 #define EMIT_RI10(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
 { \
    emit_RI10(p, _op, rT, rA, imm, __FUNCTION__);             \
 }
 
 #define EMIT_RI10s(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
 { \
    emit_RI10s(p, _op, rT, rA, imm, __FUNCTION__);             \
 }
 
 #define EMIT_RI16(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, int imm) \
+void _name (struct spe_function *p, int rT, int imm) \
 { \
    emit_RI16(p, _op, rT, imm, __FUNCTION__);                 \
 }
 
 #define EMIT_RI18(_name, _op) \
-void _name (struct spe_function *p, unsigned rT, int imm) \
+void _name (struct spe_function *p, int rT, int imm) \
 { \
    emit_RI18(p, _op, rT, imm, __FUNCTION__);                 \
 }
@@ -424,7 +424,7 @@ void _name (struct spe_function *p, int imm) \
  */
 void spe_init_func(struct spe_function *p, unsigned code_size)
 {
-    unsigned int i;
+    uint i;
 
     if (!code_size)
        code_size = 64;
@@ -503,6 +503,7 @@ int spe_allocate_register(struct spe_function *p, int reg)
  */
 void spe_release_register(struct spe_function *p, int reg)
 {
+   assert(reg >= 0);
    assert(reg < SPE_NUM_REGS);
    assert(p->regs[reg] == 1);
 
@@ -517,7 +518,7 @@ void spe_release_register(struct spe_function *p, int reg)
  */
 void spe_allocate_register_set(struct spe_function *p)
 {
-   unsigned int i;
+   uint i;
 
    /* Keep track of the set count.  If it ever wraps around to 0, 
     * we're in trouble.
@@ -538,7 +539,7 @@ void spe_allocate_register_set(struct spe_function *p)
 
 void spe_release_register_set(struct spe_function *p)
 {
-   unsigned int i;
+   uint i;
 
    /* If the set count drops below zero, we're in trouble. */
    assert(p->set_count > 0);
@@ -599,7 +600,7 @@ spe_comment(struct spe_function *p, int rel_indent, const char *s)
  * Load quad word.
  * NOTE: offset is in bytes and the least significant 4 bits must be zero!
  */
-void spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset)
+void spe_lqd(struct spe_function *p, int rT, int rA, int offset)
 {
    const boolean pSave = p->print;
 
@@ -624,7 +625,7 @@ void spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset)
  * Store quad word.
  * NOTE: offset is in bytes and the least significant 4 bits must be zero!
  */
-void spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset)
+void spe_stqd(struct spe_function *p, int rT, int rA, int offset)
 {
    const boolean pSave = p->print;
 
@@ -653,51 +654,51 @@ void spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset)
  */
 
 /** Branch Indirect to address in rA */
-void spe_bi(struct spe_function *p, unsigned rA, int d, int e)
+void spe_bi(struct spe_function *p, int rA, int d, int e)
 {
    emit_RI7(p, 0x1a8, 0, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Interupt Return */
-void spe_iret(struct spe_function *p, unsigned rA, int d, int e)
+void spe_iret(struct spe_function *p, int rA, int d, int e)
 {
    emit_RI7(p, 0x1aa, 0, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Branch indirect and set link on external data */
-void spe_bisled(struct spe_function *p, unsigned rT, unsigned rA, int d,
+void spe_bisled(struct spe_function *p, int rT, int rA, int d,
 		int e)
 {
    emit_RI7(p, 0x1ab, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Branch indirect and set link.  Save PC in rT, jump to rA. */
-void spe_bisl(struct spe_function *p, unsigned rT, unsigned rA, int d,
+void spe_bisl(struct spe_function *p, int rT, int rA, int d,
 		int e)
 {
    emit_RI7(p, 0x1a9, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Branch indirect if zero word.  If rT.word[0]==0, jump to rA. */
-void spe_biz(struct spe_function *p, unsigned rT, unsigned rA, int d, int e)
+void spe_biz(struct spe_function *p, int rT, int rA, int d, int e)
 {
    emit_RI7(p, 0x128, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Branch indirect if non-zero word.  If rT.word[0]!=0, jump to rA. */
-void spe_binz(struct spe_function *p, unsigned rT, unsigned rA, int d, int e)
+void spe_binz(struct spe_function *p, int rT, int rA, int d, int e)
 {
    emit_RI7(p, 0x129, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Branch indirect if zero halfword.  If rT.halfword[1]==0, jump to rA. */
-void spe_bihz(struct spe_function *p, unsigned rT, unsigned rA, int d, int e)
+void spe_bihz(struct spe_function *p, int rT, int rA, int d, int e)
 {
    emit_RI7(p, 0x12a, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
 
 /** Branch indirect if non-zero halfword.  If rT.halfword[1]!=0, jump to rA. */
-void spe_bihnz(struct spe_function *p, unsigned rT, unsigned rA, int d, int e)
+void spe_bihnz(struct spe_function *p, int rT, int rA, int d, int e)
 {
    emit_RI7(p, 0x12b, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
 }
@@ -733,7 +734,7 @@ EMIT_R   (spe_mtspr, 0x10c);
 
 
 void
-spe_load_float(struct spe_function *p, unsigned rT, float x)
+spe_load_float(struct spe_function *p, int rT, float x)
 {
    if (x == 0.0f) {
       spe_il(p, rT, 0x0);
@@ -760,7 +761,7 @@ spe_load_float(struct spe_function *p, unsigned rT, float x)
 
 
 void
-spe_load_int(struct spe_function *p, unsigned rT, int i)
+spe_load_int(struct spe_function *p, int rT, int i)
 {
    if (-32768 <= i && i <= 32767) {
       spe_il(p, rT, i);
@@ -772,7 +773,7 @@ spe_load_int(struct spe_function *p, unsigned rT, int i)
    }
 }
 
-void spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui)
+void spe_load_uint(struct spe_function *p, int rT, uint ui)
 {
    /* If the whole value is in the lower 18 bits, use ila, which
     * doesn't sign-extend.  Otherwise, if the two halfwords of
@@ -793,7 +794,7 @@ void spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui)
       ((ui & 0x00ff0000) == 0 || (ui & 0x00ff0000) == 0x00ff0000) &&
       ((ui & 0xff000000) == 0 || (ui & 0xff000000) == 0xff000000)
    ) {
-      unsigned int mask = 0;
+      uint mask = 0;
       /* fsmbi duplicates each bit in the given mask eight times,
        * using a 16-bit value to initialize a 16-byte quadword.
        * Each 4-bit nybble of the mask corresponds to a full word
@@ -822,7 +823,7 @@ void spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui)
  * Changes to one should be made in the other.
  */
 void
-spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+spe_and_uint(struct spe_function *p, int rT, int rA, uint ui)
 {
    /* If we can, emit a single instruction, either And Byte Immediate
     * (which uses the same constant across each byte), And Halfword Immediate
@@ -832,7 +833,7 @@ spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
     *
     * Otherwise, we'll need to use a temporary register.
     */
-   unsigned int tmp;
+   uint tmp;
 
    /* If the upper 23 bits are all 0s or all 1s, sign extension
     * will work and we can use And Word Immediate
@@ -863,7 +864,7 @@ spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
    }
 
    /* Otherwise, we'll have to use a temporary register. */
-   unsigned int tmp_reg = spe_allocate_available_register(p);
+   int tmp_reg = spe_allocate_available_register(p);
    spe_load_uint(p, tmp_reg, ui);
    spe_and(p, rT, rA, tmp_reg);
    spe_release_register(p, tmp_reg);
@@ -875,7 +876,7 @@ spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
  * Changes to one should be made in the other.
  */
 void
-spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+spe_xor_uint(struct spe_function *p, int rT, int rA, uint ui)
 {
    /* If we can, emit a single instruction, either Exclusive Or Byte 
     * Immediate (which uses the same constant across each byte), Exclusive 
@@ -885,7 +886,7 @@ spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
     *
     * Otherwise, we'll need to use a temporary register.
     */
-   unsigned int tmp;
+   uint tmp;
 
    /* If the upper 23 bits are all 0s or all 1s, sign extension
     * will work and we can use Exclusive Or Word Immediate
@@ -916,14 +917,14 @@ spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
    }
 
    /* Otherwise, we'll have to use a temporary register. */
-   unsigned int tmp_reg = spe_allocate_available_register(p);
+   int tmp_reg = spe_allocate_available_register(p);
    spe_load_uint(p, tmp_reg, ui);
    spe_xor(p, rT, rA, tmp_reg);
    spe_release_register(p, tmp_reg);
 }
 
 void
-spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+spe_compare_equal_uint(struct spe_function *p, int rT, int rA, uint ui)
 {
    /* If the comparison value is 9 bits or less, it fits inside a
     * Compare Equal Word Immediate instruction.
@@ -933,7 +934,7 @@ spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigne
    }
    /* Otherwise, we're going to have to load a word first. */
    else {
-      unsigned int tmp_reg = spe_allocate_available_register(p);
+      int tmp_reg = spe_allocate_available_register(p);
       spe_load_uint(p, tmp_reg, ui);
       spe_ceq(p, rT, rA, tmp_reg);
       spe_release_register(p, tmp_reg);
@@ -941,7 +942,7 @@ spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigne
 }
 
 void
-spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+spe_compare_greater_uint(struct spe_function *p, int rT, int rA, uint ui)
 {
    /* If the comparison value is 10 bits or less, it fits inside a
     * Compare Logical Greater Than Word Immediate instruction.
@@ -951,7 +952,7 @@ spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsig
    }
    /* Otherwise, we're going to have to load a word first. */
    else {
-      unsigned int tmp_reg = spe_allocate_available_register(p);
+      int tmp_reg = spe_allocate_available_register(p);
       spe_load_uint(p, tmp_reg, ui);
       spe_clgt(p, rT, rA, tmp_reg);
       spe_release_register(p, tmp_reg);
@@ -959,10 +960,10 @@ spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsig
 }
 
 void
-spe_splat(struct spe_function *p, unsigned rT, unsigned rA)
+spe_splat(struct spe_function *p, int rT, int rA)
 {
    /* Use a temporary, just in case rT == rA */
-   unsigned int tmp_reg = spe_allocate_available_register(p);
+   int tmp_reg = spe_allocate_available_register(p);
    /* Duplicate bytes 0, 1, 2, and 3 across the whole register */
    spe_ila(p, tmp_reg, 0x00010203);
    spe_shufb(p, rT, rA, rA, tmp_reg);
@@ -971,14 +972,14 @@ spe_splat(struct spe_function *p, unsigned rT, unsigned rA)
 
 
 void
-spe_complement(struct spe_function *p, unsigned rT, unsigned rA)
+spe_complement(struct spe_function *p, int rT, int rA)
 {
    spe_nor(p, rT, rA, rA);
 }
 
 
 void
-spe_move(struct spe_function *p, unsigned rT, unsigned rA)
+spe_move(struct spe_function *p, int rT, int rA)
 {
    /* Use different instructions depending on the instruction address
     * to take advantage of the dual pipelines.
@@ -991,14 +992,14 @@ spe_move(struct spe_function *p, unsigned rT, unsigned rA)
 
 
 void
-spe_zero(struct spe_function *p, unsigned rT)
+spe_zero(struct spe_function *p, int rT)
 {
    spe_xor(p, rT, rT, rT);
 }
 
 
 void
-spe_splat_word(struct spe_function *p, unsigned rT, unsigned rA, int word)
+spe_splat_word(struct spe_function *p, int rT, int rA, int word)
 {
    assert(word >= 0);
    assert(word <= 3);
@@ -1038,9 +1039,9 @@ spe_splat_word(struct spe_function *p, unsigned rT, unsigned rA, int word)
  * like "x = min(x, a)", we always allocate a new register to be safe.
  */
 void 
-spe_float_min(struct spe_function *p, unsigned rT, unsigned rA, unsigned rB)
+spe_float_min(struct spe_function *p, int rT, int rA, int rB)
 {
-   unsigned int compare_reg = spe_allocate_available_register(p);
+   int compare_reg = spe_allocate_available_register(p);
    spe_fcgt(p, compare_reg, rA, rB);
    spe_selb(p, rT, rA, rB, compare_reg);
    spe_release_register(p, compare_reg);
@@ -1055,9 +1056,9 @@ spe_float_min(struct spe_function *p, unsigned rT, unsigned rA, unsigned rB)
  * so that the larger of the two is selected instead of the smaller.
  */
 void 
-spe_float_max(struct spe_function *p, unsigned rT, unsigned rA, unsigned rB)
+spe_float_max(struct spe_function *p, int rT, int rA, int rB)
 {
-   unsigned int compare_reg = spe_allocate_available_register(p);
+   int compare_reg = spe_allocate_available_register(p);
    spe_fcgt(p, compare_reg, rA, rB);
    spe_selb(p, rT, rB, rA, compare_reg);
    spe_release_register(p, compare_reg);
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
index f9ad2acacdd..65d9c774154 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
@@ -79,9 +79,9 @@ struct spe_function
 };
 
 
-extern void spe_init_func(struct spe_function *p, unsigned code_size);
+extern void spe_init_func(struct spe_function *p, uint code_size);
 extern void spe_release_func(struct spe_function *p);
-extern unsigned spe_code_size(const struct spe_function *p);
+extern uint spe_code_size(const struct spe_function *p);
 
 extern int spe_allocate_available_register(struct spe_function *p);
 extern int spe_allocate_register(struct spe_function *p, int reg);
@@ -89,8 +89,7 @@ extern void spe_release_register(struct spe_function *p, int reg);
 extern void spe_allocate_register_set(struct spe_function *p);
 extern void spe_release_register_set(struct spe_function *p);
 
-extern unsigned
-spe_get_registers_used(const struct spe_function *p, ubyte used[]);
+extern uint spe_get_registers_used(const struct spe_function *p, ubyte used[]);
 
 extern void spe_print_code(struct spe_function *p, boolean enable);
 extern void spe_indent(struct spe_function *p, int spaces);
@@ -103,31 +102,25 @@ extern void spe_comment(struct spe_function *p, int rel_indent, const char *s);
 #define EMIT(_name, _op) \
     extern void _name (struct spe_function *p);
 #define EMIT_(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT);
+    extern void _name (struct spe_function *p, int rT);
 #define EMIT_R(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA);
+    extern void _name (struct spe_function *p, int rT, int rA);
 #define EMIT_RR(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
-                       unsigned rB);
+    extern void _name (struct spe_function *p, int rT, int rA, int rB);
 #define EMIT_RRR(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
-                       unsigned rB, unsigned rC);
+    extern void _name (struct spe_function *p, int rT, int rA, int rB, int rC);
 #define EMIT_RI7(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
-                       int imm);
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
 #define EMIT_RI8(_name, _op, bias) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
-                       int imm);
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
 #define EMIT_RI10(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
-                       int imm);
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
 #define EMIT_RI10s(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
-                       int imm);
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
 #define EMIT_RI16(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, int imm);
+    extern void _name (struct spe_function *p, int rT, int imm);
 #define EMIT_RI18(_name, _op) \
-    extern void _name (struct spe_function *p, unsigned rT, int imm);
+    extern void _name (struct spe_function *p, int rT, int imm);
 #define EMIT_I16(_name, _op) \
     extern void _name (struct spe_function *p, int imm);
 #define UNDEF_EMIT_MACROS
@@ -301,82 +294,82 @@ EMIT_RI16(spe_brhz,      0x044)
 EMIT     (spe_lnop,      0x001)
 
 extern void
-spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset);
+spe_lqd(struct spe_function *p, int rT, int rA, int offset);
 
 extern void
-spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset);
+spe_stqd(struct spe_function *p, int rT, int rA, int offset);
 
-extern void spe_bi(struct spe_function *p, unsigned rA, int d, int e);
-extern void spe_iret(struct spe_function *p, unsigned rA, int d, int e);
-extern void spe_bisled(struct spe_function *p, unsigned rT, unsigned rA,
+extern void spe_bi(struct spe_function *p, int rA, int d, int e);
+extern void spe_iret(struct spe_function *p, int rA, int d, int e);
+extern void spe_bisled(struct spe_function *p, int rT, int rA,
     int d, int e);
-extern void spe_bisl(struct spe_function *p, unsigned rT, unsigned rA,
+extern void spe_bisl(struct spe_function *p, int rT, int rA,
     int d, int e);
-extern void spe_biz(struct spe_function *p, unsigned rT, unsigned rA,
+extern void spe_biz(struct spe_function *p, int rT, int rA,
     int d, int e);
-extern void spe_binz(struct spe_function *p, unsigned rT, unsigned rA,
+extern void spe_binz(struct spe_function *p, int rT, int rA,
     int d, int e);
-extern void spe_bihz(struct spe_function *p, unsigned rT, unsigned rA,
+extern void spe_bihz(struct spe_function *p, int rT, int rA,
     int d, int e);
-extern void spe_bihnz(struct spe_function *p, unsigned rT, unsigned rA,
+extern void spe_bihnz(struct spe_function *p, int rT, int rA,
     int d, int e);
 
 
 /** Load/splat immediate float into rT. */
 extern void
-spe_load_float(struct spe_function *p, unsigned rT, float x);
+spe_load_float(struct spe_function *p, int rT, float x);
 
 /** Load/splat immediate int into rT. */
 extern void
-spe_load_int(struct spe_function *p, unsigned rT, int i);
+spe_load_int(struct spe_function *p, int rT, int i);
 
 /** Load/splat immediate unsigned int into rT. */
 extern void
-spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui);
+spe_load_uint(struct spe_function *p, int rT, uint ui);
 
 /** And immediate value into rT. */
 extern void
-spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+spe_and_uint(struct spe_function *p, int rT, int rA, uint ui);
 
 /** Xor immediate value into rT. */
 extern void
-spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+spe_xor_uint(struct spe_function *p, int rT, int rA, uint ui);
 
 /** Compare equal with immediate value. */
 extern void
-spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+spe_compare_equal_uint(struct spe_function *p, int rT, int rA, uint ui);
 
 /** Compare greater with immediate value. */
 extern void
-spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+spe_compare_greater_uint(struct spe_function *p, int rT, int rA, uint ui);
 
 /** Replicate word 0 of rA across rT. */
 extern void
-spe_splat(struct spe_function *p, unsigned rT, unsigned rA);
+spe_splat(struct spe_function *p, int rT, int rA);
 
 /** rT = complement_all_bits(rA). */
 extern void
-spe_complement(struct spe_function *p, unsigned rT, unsigned rA);
+spe_complement(struct spe_function *p, int rT, int rA);
 
 /** rT = rA. */
 extern void
-spe_move(struct spe_function *p, unsigned rT, unsigned rA);
+spe_move(struct spe_function *p, int rT, int rA);
 
 /** rT = {0,0,0,0}. */
 extern void
-spe_zero(struct spe_function *p, unsigned rT);
+spe_zero(struct spe_function *p, int rT);
 
 /** rT = splat(rA, word) */
 extern void
-spe_splat_word(struct spe_function *p, unsigned rT, unsigned rA, int word);
+spe_splat_word(struct spe_function *p, int rT, int rA, int word);
 
 /** rT = float min(rA, rB) */
 extern void
-spe_float_min(struct spe_function *p, unsigned rT, unsigned rA, unsigned rB);
+spe_float_min(struct spe_function *p, int rT, int rA, int rB);
 
 /** rT = float max(rA, rB) */
 extern void
-spe_float_max(struct spe_function *p, unsigned rT, unsigned rA, unsigned rB);
+spe_float_max(struct spe_function *p, int rT, int rA, int rB);
 
 
 /* Floating-point instructions
diff --git a/src/gallium/auxiliary/util/u_memory.h b/src/gallium/auxiliary/util/u_memory.h
index 626b13af83d..1a6b596421f 100644
--- a/src/gallium/auxiliary/util/u_memory.h
+++ b/src/gallium/auxiliary/util/u_memory.h
@@ -52,7 +52,7 @@ extern "C" {
 #endif
 
 
-#if defined(PIPE_OS_WINDOWS) && defined(DEBUG) 
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) && defined(DEBUG) 
 
 /* memory debugging */
 
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index 98554d7f521..1f6860da119 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -49,6 +49,15 @@
    }
 
 
+
+#define JOIN(x, y) JOIN_AGAIN(x, y)
+#define JOIN_AGAIN(x, y) x ## y
+
+#define STATIC_ASSERT(e) \
+{typedef char JOIN(assertion_failed_at_line_, __LINE__) [(e) ? 1 : -1];}
+
+
+
 /** for sanity checking */
 #define ASSERT_ALIGN16(ptr) \
   ASSERT((((unsigned long) (ptr)) & 0xf) == 0);
@@ -134,6 +143,11 @@ struct cell_fence
    volatile uint status[CELL_MAX_SPUS][4];
 };
 
+#ifdef __SPU__
+typedef vector unsigned int opcode_t;
+#else
+typedef unsigned int opcode_t[4];
+#endif
 
 /**
  * Fence command sent to SPUs.  In response, the SPUs will write
@@ -141,8 +155,9 @@ struct cell_fence
  */
 struct cell_command_fence
 {
-   uint64_t opcode;      /**< CELL_CMD_FENCE */
+   opcode_t opcode;      /**< CELL_CMD_FENCE */
    struct cell_fence *fence;
+   uint32_t pad_[3];
 };
 
 
@@ -163,7 +178,7 @@ struct cell_command_fence
  */
 struct cell_command_fragment_ops
 {
-   uint64_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_OPS */
+   opcode_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_OPS */
 
    /* Fields for the fallback case */
    struct pipe_depth_stencil_alpha_state dsa;
@@ -189,8 +204,9 @@ struct cell_command_fragment_ops
  */
 struct cell_command_fragment_program
 {
-   uint64_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_PROGRAM */
+   opcode_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_PROGRAM */
    uint num_inst;        /**< Number of instructions */
+   uint32_t pad[3];
    unsigned code[SPU_MAX_FRAGMENT_PROGRAM_INSTS];
 };
 
@@ -200,10 +216,11 @@ struct cell_command_fragment_program
  */
 struct cell_command_framebuffer
 {
-   uint64_t opcode;     /**< CELL_CMD_STATE_FRAMEBUFFER */
+   opcode_t opcode;     /**< CELL_CMD_STATE_FRAMEBUFFER */
    int width, height;
    void *color_start, *depth_start;
    enum pipe_format color_format, depth_format;
+   uint32_t pad_[2];
 };
 
 
@@ -212,7 +229,7 @@ struct cell_command_framebuffer
  */
 struct cell_command_rasterizer
 {
-   uint64_t opcode;    /**< CELL_CMD_STATE_RASTERIZER */
+   opcode_t opcode;    /**< CELL_CMD_STATE_RASTERIZER */
    struct pipe_rasterizer_state rasterizer;
 };
 
@@ -222,9 +239,10 @@ struct cell_command_rasterizer
  */
 struct cell_command_clear_surface
 {
-   uint64_t opcode;     /**< CELL_CMD_CLEAR_SURFACE */
+   opcode_t opcode;     /**< CELL_CMD_CLEAR_SURFACE */
    uint surface; /**< Temporary: 0=color, 1=Z */
    uint value;
+   uint32_t pad[2];
 };
 
 
@@ -271,7 +289,7 @@ struct cell_shader_info
 #define SPU_VERTS_PER_BATCH 64
 struct cell_command_vs
 {
-   uint64_t opcode;       /**< CELL_CMD_VS_EXECUTE */
+   opcode_t opcode;       /**< CELL_CMD_VS_EXECUTE */
    uint64_t vOut[SPU_VERTS_PER_BATCH];
    unsigned num_elts;
    unsigned elts[SPU_VERTS_PER_BATCH];
@@ -283,7 +301,7 @@ struct cell_command_vs
 
 struct cell_command_render
 {
-   uint64_t opcode;   /**< CELL_CMD_RENDER */
+   opcode_t opcode;   /**< CELL_CMD_RENDER */
    uint prim_type;    /**< PIPE_PRIM_x */
    uint num_verts;
    uint vertex_size;  /**< bytes per vertex */
@@ -292,27 +310,30 @@ struct cell_command_render
    float xmin, ymin, xmax, ymax;  /* XXX another dummy field */
    uint min_index;
    boolean inline_verts;
+   uint32_t pad_[1];
 };
 
 
 struct cell_command_release_verts
 {
-   uint64_t opcode;         /**< CELL_CMD_RELEASE_VERTS */
+   opcode_t opcode;         /**< CELL_CMD_RELEASE_VERTS */
    uint vertex_buf;    /**< in [0, CELL_NUM_BUFFERS-1] */
+   uint32_t pad_[3];
 };
 
 
 struct cell_command_sampler
 {
-   uint64_t opcode;         /**< CELL_CMD_STATE_SAMPLER */
+   opcode_t opcode;         /**< CELL_CMD_STATE_SAMPLER */
    uint unit;
    struct pipe_sampler_state state;
+   uint32_t pad_[1];
 };
 
 
 struct cell_command_texture
 {
-   uint64_t opcode;     /**< CELL_CMD_STATE_TEXTURE */
+   opcode_t opcode;     /**< CELL_CMD_STATE_TEXTURE */
    uint target;         /**< PIPE_TEXTURE_x */
    uint unit;
    void *start[CELL_MAX_TEXTURE_LEVELS];   /**< Address in main memory */
diff --git a/src/gallium/drivers/cell/ppu/cell_batch.c b/src/gallium/drivers/cell/ppu/cell_batch.c
index 962775cd335..fe144f8b849 100644
--- a/src/gallium/drivers/cell/ppu/cell_batch.c
+++ b/src/gallium/drivers/cell/ppu/cell_batch.c
@@ -108,15 +108,16 @@ emit_fence(struct cell_context *cell)
       fence->status[i][0] = CELL_FENCE_EMITTED;
    }
 
+   STATIC_ASSERT(sizeof(struct cell_command_fence) % 16 == 0);
+   ASSERT(size % 16 == 0);
    ASSERT(size + sizeof(struct cell_command_fence) <= CELL_BUFFER_SIZE);
 
    fence_cmd = (struct cell_command_fence *) (cell->buffer[batch] + size);
-   fence_cmd->opcode = CELL_CMD_FENCE;
+   fence_cmd->opcode[0] = CELL_CMD_FENCE;
    fence_cmd->fence = fence;
 
    /* update batch buffer size */
    cell->buffer_size[batch] = size + sizeof(struct cell_command_fence);
-   assert(sizeof(struct cell_command_fence) % 8 == 0);
 }
 
 
@@ -192,69 +193,18 @@ cell_batch_free_space(const struct cell_context *cell)
 
 
 /**
- * Append data to the current batch buffer.
- * \param data  address of block of bytes to append
- * \param bytes  size of block of bytes
- */
-void
-cell_batch_append(struct cell_context *cell, const void *data, uint bytes)
-{
-   uint size;
-
-   ASSERT(bytes % 8 == 0);
-   ASSERT(bytes <= CELL_BUFFER_SIZE);
-   ASSERT(cell->cur_batch >= 0);
-
-#ifdef ASSERT
-   {
-      uint spu;
-      for (spu = 0; spu < cell->num_spus; spu++) {
-         ASSERT(cell->buffer_status[spu][cell->cur_batch][0]
-                 == CELL_BUFFER_STATUS_USED);
-      }
-   }
-#endif
-
-   size = cell->buffer_size[cell->cur_batch];
-
-   if (bytes > cell_batch_free_space(cell)) {
-      cell_batch_flush(cell);
-      size = 0;
-   }
-
-   ASSERT(size + bytes <= CELL_BUFFER_SIZE);
-
-   memcpy(cell->buffer[cell->cur_batch] + size, data, bytes);
-
-   cell->buffer_size[cell->cur_batch] = size + bytes;
-}
-
-
-/**
  * Allocate space in the current batch buffer for 'bytes' space.
+ * Bytes must be a multiple of 16 bytes.  Allocation will be 16 byte aligned.
  * \return address in batch buffer to put data
  */
 void *
-cell_batch_alloc(struct cell_context *cell, uint bytes)
-{
-   return cell_batch_alloc_aligned(cell, bytes, 1);
-}
-
-
-/**
- * Same as \sa cell_batch_alloc, but return an address at a particular
- * alignment.
- */
-void *
-cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
-                         uint alignment)
+cell_batch_alloc16(struct cell_context *cell, uint bytes)
 {
    void *pos;
-   uint size, padbytes;
+   uint size;
 
-   ASSERT(bytes % 8 == 0);
+   ASSERT(bytes % 16 == 0);
    ASSERT(bytes <= CELL_BUFFER_SIZE);
-   ASSERT(alignment > 0);
    ASSERT(cell->cur_batch >= 0);
 
 #ifdef ASSERT
@@ -269,17 +219,12 @@ cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
 
    size = cell->buffer_size[cell->cur_batch];
 
-   padbytes = (alignment - (size % alignment)) % alignment;
-
-   if (padbytes + bytes > cell_batch_free_space(cell)) {
+   if (bytes > cell_batch_free_space(cell)) {
       cell_batch_flush(cell);
       size = 0;
    }
-   else {
-      size += padbytes;
-   }
 
-   ASSERT(size % alignment == 0);
+   ASSERT(size % 16 == 0);
    ASSERT(size + bytes <= CELL_BUFFER_SIZE);
 
    pos = (void *) (cell->buffer[cell->cur_batch] + size);
diff --git a/src/gallium/drivers/cell/ppu/cell_batch.h b/src/gallium/drivers/cell/ppu/cell_batch.h
index f74dd600791..290136031a1 100644
--- a/src/gallium/drivers/cell/ppu/cell_batch.h
+++ b/src/gallium/drivers/cell/ppu/cell_batch.h
@@ -44,15 +44,8 @@ cell_batch_flush(struct cell_context *cell);
 extern uint
 cell_batch_free_space(const struct cell_context *cell);
 
-extern void
-cell_batch_append(struct cell_context *cell, const void *data, uint bytes);
-
-extern void *
-cell_batch_alloc(struct cell_context *cell, uint bytes);
-
 extern void *
-cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
-                         uint alignment);
+cell_batch_alloc16(struct cell_context *cell, uint bytes);
 
 extern void
 cell_init_batch_buffers(struct cell_context *cell);
diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c
index 037635e4660..c2e276988ca 100644
--- a/src/gallium/drivers/cell/ppu/cell_clear.c
+++ b/src/gallium/drivers/cell/ppu/cell_clear.c
@@ -99,10 +99,11 @@ cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps,
 
    /* Build a CLEAR command and place it in the current batch buffer */
    {
+      STATIC_ASSERT(sizeof(struct cell_command_clear_surface) % 16 == 0);
       struct cell_command_clear_surface *clr
          = (struct cell_command_clear_surface *)
-         cell_batch_alloc(cell, sizeof(*clr));
-      clr->opcode = CELL_CMD_CLEAR_SURFACE;
+         cell_batch_alloc16(cell, sizeof(*clr));
+      clr->opcode[0] = CELL_CMD_CLEAR_SURFACE;
       clr->surface = surfIndex;
       clr->value = clearValue;
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.c b/src/gallium/drivers/cell/ppu/cell_flush.c
index a64967b4b9e..8275c9dc9c7 100644
--- a/src/gallium/drivers/cell/ppu/cell_flush.c
+++ b/src/gallium/drivers/cell/ppu/cell_flush.c
@@ -72,8 +72,9 @@ cell_flush_int(struct cell_context *cell, unsigned flags)
    flushing = TRUE;
 
    if (flags & CELL_FLUSH_WAIT) {
-      uint64_t *cmd = (uint64_t *) cell_batch_alloc(cell, sizeof(uint64_t));
-      *cmd = CELL_CMD_FINISH;
+      STATIC_ASSERT(sizeof(opcode_t) % 16 == 0);
+      opcode_t *cmd = (opcode_t*) cell_batch_alloc16(cell, sizeof(opcode_t));
+      *cmd[0] = CELL_CMD_FINISH;
    }
 
    cell_batch_flush(cell);
@@ -101,11 +102,11 @@ void
 cell_flush_buffer_range(struct cell_context *cell, void *ptr,
 			unsigned size)
 {
-   uint64_t batch[1 + (ROUNDUP8(sizeof(struct cell_buffer_range)) / 8)];
-   struct cell_buffer_range *br = (struct cell_buffer_range *) & batch[1];
-
+   STATIC_ASSERT((sizeof(opcode_t) + sizeof(struct cell_buffer_range)) % 16 == 0);
+   uint32_t *batch = (uint32_t*)cell_batch_alloc16(cell, 
+      sizeof(opcode_t) + sizeof(struct cell_buffer_range));
+   struct cell_buffer_range *br = (struct cell_buffer_range *) &batch[4];
    batch[0] = CELL_CMD_FLUSH_BUFFER_RANGE;
    br->base = (uintptr_t) ptr;
    br->size = size;
-   cell_batch_append(cell, batch, sizeof(batch));
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
index 2c64eb1bccd..0ea8f017ef9 100644
--- a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
@@ -2,6 +2,7 @@
  * 
  * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
+ * Copyright 2009 VMware, Inc.  All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the
@@ -25,11 +26,10 @@
  * 
  **************************************************************************/
 
-
-
 /**
  * Generate SPU per-fragment code (actually per-quad code).
  * \author Brian Paul
+ * \author Bob Ellison
  */
 
 
@@ -55,7 +55,7 @@
  * \param ifbZ_reg    register containing integer frame buffer Z values (in/out)
  * \param zmask_reg   register containing result of Z test/comparison (out)
  *
- * Returns true if the Z-buffer needs to be updated.
+ * Returns TRUE if the Z-buffer needs to be updated.
  */
 static boolean
 gen_depth_test(struct spe_function *f,
@@ -134,10 +134,10 @@ gen_depth_test(struct spe_function *f,
        * framebufferZ = (ztest_passed ? fragmentZ : framebufferZ;
        */
       spe_selb(f, ifbZ_reg, ifbZ_reg, ifragZ_reg, mask_reg);
-      return true;
+      return TRUE;
    }
 
-   return false;
+   return FALSE;
 }
 
 
@@ -237,41 +237,136 @@ gen_alpha_test(const struct pipe_depth_stencil_alpha_state *dsa,
    spe_release_register(f, amask_reg);
 }
 
-/* This pair of functions is used inline to allocate and deallocate
+
+/**
+ * This pair of functions is used inline to allocate and deallocate
  * optional constant registers.  Once a constant is discovered to be 
  * needed, we will likely need it again, so we don't want to deallocate
  * it and have to allocate and load it again unnecessarily.
  */
-static inline void
-setup_optional_register(struct spe_function *f, boolean *is_already_set, unsigned int *r)
+static INLINE void
+setup_optional_register(struct spe_function *f,
+                        int *r)
 {
-   if (*is_already_set) return;
-   *r = spe_allocate_available_register(f);
-   *is_already_set = true;
+   if (*r < 0)
+      *r = spe_allocate_available_register(f);
 }
 
-static inline void
-release_optional_register(struct spe_function *f, boolean *is_already_set, unsigned int r)
+static INLINE void
+release_optional_register(struct spe_function *f,
+                          int r)
 {
-    if (!*is_already_set) return;
-    spe_release_register(f, r);
-    *is_already_set = false;
+   if (r >= 0)
+      spe_release_register(f, r);
 }
 
-static inline void
-setup_const_register(struct spe_function *f, boolean *is_already_set, unsigned int *r, float value)
+static INLINE void
+setup_const_register(struct spe_function *f,
+                     int *r,
+                     float value)
 {
-   if (*is_already_set) return;
-   setup_optional_register(f, is_already_set, r);
+   if (*r >= 0)
+      return;
+   setup_optional_register(f, r);
    spe_load_float(f, *r, value);
 }
 
-static inline void
-release_const_register(struct spe_function *f, boolean *is_already_set, unsigned int r)
+static INLINE void
+release_const_register(struct spe_function *f,
+                       int r)
 {
-    release_optional_register(f, is_already_set, r);
+   release_optional_register(f, r);
 }
 
+
+
+/**
+ * Unpack/convert framebuffer colors from four 32-bit packed colors
+ * (fbRGBA) to four float RGBA vectors (fbR, fbG, fbB, fbA).
+ * Each 8-bit color component is expanded into a float in [0.0, 1.0].
+ */
+static void
+unpack_colors(struct spe_function *f,
+              enum pipe_format color_format,
+              int fbRGBA_reg,
+              int fbR_reg, int fbG_reg, int fbB_reg, int fbA_reg)
+{
+   int mask0_reg = spe_allocate_available_register(f);
+   int mask1_reg = spe_allocate_available_register(f);
+   int mask2_reg = spe_allocate_available_register(f);
+   int mask3_reg = spe_allocate_available_register(f);
+
+   spe_load_int(f, mask0_reg, 0xff);
+   spe_load_int(f, mask1_reg, 0xff00);
+   spe_load_int(f, mask2_reg, 0xff0000);
+   spe_load_int(f, mask3_reg, 0xff000000);
+
+   spe_comment(f, 0, "Unpack framebuffer colors, convert to floats");
+
+   switch (color_format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      /* fbB = fbRGBA & mask */
+      spe_and(f, fbB_reg, fbRGBA_reg, mask0_reg);
+
+      /* fbG = fbRGBA & mask */
+      spe_and(f, fbG_reg, fbRGBA_reg, mask1_reg);
+
+      /* fbR = fbRGBA & mask */
+      spe_and(f, fbR_reg, fbRGBA_reg, mask2_reg);
+
+      /* fbA = fbRGBA & mask */
+      spe_and(f, fbA_reg, fbRGBA_reg, mask3_reg);
+
+      /* fbG = fbG >> 8 */
+      spe_roti(f, fbG_reg, fbG_reg, -8);
+
+      /* fbR = fbR >> 16 */
+      spe_roti(f, fbR_reg, fbR_reg, -16);
+
+      /* fbA = fbA >> 24 */
+      spe_roti(f, fbA_reg, fbA_reg, -24);
+      break;
+
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      /* fbA = fbRGBA & mask */
+      spe_and(f, fbA_reg, fbRGBA_reg, mask0_reg);
+
+      /* fbR = fbRGBA & mask */
+      spe_and(f, fbR_reg, fbRGBA_reg, mask1_reg);
+
+      /* fbG = fbRGBA & mask */
+      spe_and(f, fbG_reg, fbRGBA_reg, mask2_reg);
+
+      /* fbB = fbRGBA & mask */
+      spe_and(f, fbB_reg, fbRGBA_reg, mask3_reg);
+
+      /* fbR = fbR >> 8 */
+      spe_roti(f, fbR_reg, fbR_reg, -8);
+
+      /* fbG = fbG >> 16 */
+      spe_roti(f, fbG_reg, fbG_reg, -16);
+
+      /* fbB = fbB >> 24 */
+      spe_roti(f, fbB_reg, fbB_reg, -24);
+      break;
+
+   default:
+      ASSERT(0);
+   }
+
+   /* convert int[4] in [0,255] to float[4] in [0.0, 1.0] */
+   spe_cuflt(f, fbR_reg, fbR_reg, 8);
+   spe_cuflt(f, fbG_reg, fbG_reg, 8);
+   spe_cuflt(f, fbB_reg, fbB_reg, 8);
+   spe_cuflt(f, fbA_reg, fbA_reg, 8);
+
+   spe_release_register(f, mask0_reg);
+   spe_release_register(f, mask1_reg);
+   spe_release_register(f, mask2_reg);
+   spe_release_register(f, mask3_reg);
+}
+
+
 /**
  * Generate SPE code to implement the given blend mode for a quad of pixels.
  * \param f          SPE function to append instruction onto.
@@ -310,90 +405,14 @@ gen_blend(const struct pipe_blend_state *blend,
     * if we do use them, make sure we only allocate them once by
     * keeping a flag on each one.
     */
-   boolean one_reg_set = false;
-   unsigned int one_reg;
-   boolean constR_reg_set = false, constG_reg_set = false, 
-      constB_reg_set = false, constA_reg_set = false;
-   unsigned int constR_reg, constG_reg, constB_reg, constA_reg;
+   int one_reg = -1;
+   int constR_reg = -1, constG_reg = -1, constB_reg = -1, constA_reg = -1;
 
    ASSERT(blend->blend_enable);
 
-   /* Unpack/convert framebuffer colors from four 32-bit packed colors
-    * (fbRGBA) to four float RGBA vectors (fbR, fbG, fbB, fbA).
-    * Each 8-bit color component is expanded into a float in [0.0, 1.0].
-    */
-   {
-      int mask_reg = spe_allocate_available_register(f);
-
-      /* mask = {0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff} */
-      spe_load_int(f, mask_reg, 0xff);
-
-      /* XXX there may be more clever ways to implement the following code */
-      switch (color_format) {
-      case PIPE_FORMAT_A8R8G8B8_UNORM:
-         /* fbB = fbB & mask */
-         spe_and(f, fbB_reg, fbRGBA_reg, mask_reg);
-         /* mask = mask << 8 */
-         spe_roti(f, mask_reg, mask_reg, 8);
-
-         /* fbG = fbRGBA & mask */
-         spe_and(f, fbG_reg, fbRGBA_reg, mask_reg);
-         /* fbG = fbG >> 8 */
-         spe_roti(f, fbG_reg, fbG_reg, -8);
-         /* mask = mask << 8 */
-         spe_roti(f, mask_reg, mask_reg, 8);
-
-         /* fbR = fbRGBA & mask */
-         spe_and(f, fbR_reg, fbRGBA_reg, mask_reg);
-         /* fbR = fbR >> 16 */
-         spe_roti(f, fbR_reg, fbR_reg, -16);
-         /* mask = mask << 8 */
-         spe_roti(f, mask_reg, mask_reg, 8);
-
-         /* fbA = fbRGBA & mask */
-         spe_and(f, fbA_reg, fbRGBA_reg, mask_reg);
-         /* fbA = fbA >> 24 */
-         spe_roti(f, fbA_reg, fbA_reg, -24);
-         break;
-
-      case PIPE_FORMAT_B8G8R8A8_UNORM:
-         /* fbA = fbA & mask */
-         spe_and(f, fbA_reg, fbRGBA_reg, mask_reg);
-         /* mask = mask << 8 */
-         spe_roti(f, mask_reg, mask_reg, 8);
-
-         /* fbR = fbRGBA & mask */
-         spe_and(f, fbR_reg, fbRGBA_reg, mask_reg);
-         /* fbR = fbR >> 8 */
-         spe_roti(f, fbR_reg, fbR_reg, -8);
-         /* mask = mask << 8 */
-         spe_roti(f, mask_reg, mask_reg, 8);
-
-         /* fbG = fbRGBA & mask */
-         spe_and(f, fbG_reg, fbRGBA_reg, mask_reg);
-         /* fbG = fbG >> 16 */
-         spe_roti(f, fbG_reg, fbG_reg, -16);
-         /* mask = mask << 8 */
-         spe_roti(f, mask_reg, mask_reg, 8);
-
-         /* fbB = fbRGBA & mask */
-         spe_and(f, fbB_reg, fbRGBA_reg, mask_reg);
-         /* fbB = fbB >> 24 */
-         spe_roti(f, fbB_reg, fbB_reg, -24);
-         break;
-
-      default:
-         ASSERT(0);
-      }
-
-      /* convert int[4] in [0,255] to float[4] in [0.0, 1.0] */
-      spe_cuflt(f, fbR_reg, fbR_reg, 8);
-      spe_cuflt(f, fbG_reg, fbG_reg, 8);
-      spe_cuflt(f, fbB_reg, fbB_reg, 8);
-      spe_cuflt(f, fbA_reg, fbA_reg, 8);
-
-      spe_release_register(f, mask_reg);
-   }
+   /* packed RGBA -> float colors */
+   unpack_colors(f, color_format, fbRGBA_reg,
+                 fbR_reg, fbG_reg, fbB_reg, fbA_reg);
 
    /*
     * Compute Src RGB terms.  We're actually looking for the value
@@ -476,9 +495,9 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_CONST_COLOR:
       /* We need the optional constant color registers */
-      setup_const_register(f, &constR_reg_set, &constR_reg, blend_color->color[0]);
-      setup_const_register(f, &constG_reg_set, &constG_reg, blend_color->color[1]);
-      setup_const_register(f, &constB_reg_set, &constB_reg, blend_color->color[2]);
+      setup_const_register(f, &constR_reg, blend_color->color[0]);
+      setup_const_register(f, &constG_reg, blend_color->color[1]);
+      setup_const_register(f, &constB_reg, blend_color->color[2]);
       /* now, factor = (Rc,Gc,Bc), so term = (R*Rc,G*Gc,B*Bc) */
       spe_fm(f, term1R_reg, fragR_reg, constR_reg);
       spe_fm(f, term1G_reg, fragG_reg, constG_reg);
@@ -486,7 +505,7 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_CONST_ALPHA:
       /* we'll need the optional constant alpha register */
-      setup_const_register(f, &constA_reg_set, &constA_reg, blend_color->color[3]);
+      setup_const_register(f, &constA_reg, blend_color->color[3]);
       /* factor = (Ac,Ac,Ac), so term = (R*Ac,G*Ac,B*Ac) */
       spe_fm(f, term1R_reg, fragR_reg, constA_reg);
       spe_fm(f, term1G_reg, fragG_reg, constA_reg);
@@ -494,9 +513,9 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
       /* We need the optional constant color registers */
-      setup_const_register(f, &constR_reg_set, &constR_reg, blend_color->color[0]);
-      setup_const_register(f, &constG_reg_set, &constG_reg, blend_color->color[1]);
-      setup_const_register(f, &constB_reg_set, &constB_reg, blend_color->color[2]);
+      setup_const_register(f, &constR_reg, blend_color->color[0]);
+      setup_const_register(f, &constG_reg, blend_color->color[1]);
+      setup_const_register(f, &constB_reg, blend_color->color[2]);
       /* factor = (1-Rc,1-Gc,1-Bc), so term = (R*(1-Rc),G*(1-Gc),B*(1-Bc)) 
        * or term = (R-R*Rc, G-G*Gc, B-B*Bc)
        * fnms(a,b,c,d) computes a = d - b*c
@@ -507,9 +526,9 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
       /* We need the optional constant color registers */
-      setup_const_register(f, &constR_reg_set, &constR_reg, blend_color->color[0]);
-      setup_const_register(f, &constG_reg_set, &constG_reg, blend_color->color[1]);
-      setup_const_register(f, &constB_reg_set, &constB_reg, blend_color->color[2]);
+      setup_const_register(f, &constR_reg, blend_color->color[0]);
+      setup_const_register(f, &constG_reg, blend_color->color[1]);
+      setup_const_register(f, &constB_reg, blend_color->color[2]);
       /* factor = (1-Ac,1-Ac,1-Ac), so term = (R*(1-Ac),G*(1-Ac),B*(1-Ac))
        * or term = (R-R*Ac,G-G*Ac,B-B*Ac)
        * fnms(a,b,c,d) computes a = d - b*c
@@ -520,7 +539,7 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
       /* We'll need the optional {1,1,1,1} register */
-      setup_const_register(f, &one_reg_set, &one_reg, 1.0f);
+      setup_const_register(f, &one_reg, 1.0f);
       /* factor = (min(A,1-Afb),min(A,1-Afb),min(A,1-Afb)), so 
        * term = (R*min(A,1-Afb), G*min(A,1-Afb), B*min(A,1-Afb))
        * We could expand the term (as a*min(b,c) == min(a*b,a*c)
@@ -598,7 +617,7 @@ gen_blend(const struct pipe_blend_state *blend,
    case PIPE_BLENDFACTOR_CONST_ALPHA: /* fall through */
    case PIPE_BLENDFACTOR_CONST_COLOR:
       /* We need the optional constA_reg register */
-      setup_const_register(f, &constA_reg_set, &constA_reg, blend_color->color[3]);
+      setup_const_register(f, &constA_reg, blend_color->color[3]);
       /* factor = Ac, so term = A*Ac */
       spe_fm(f, term1A_reg, fragA_reg, constA_reg);
       break;
@@ -606,7 +625,7 @@ gen_blend(const struct pipe_blend_state *blend,
    case PIPE_BLENDFACTOR_INV_CONST_ALPHA: /* fall through */
    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
       /* We need the optional constA_reg register */
-      setup_const_register(f, &constA_reg_set, &constA_reg, blend_color->color[3]);
+      setup_const_register(f, &constA_reg, blend_color->color[3]);
       /* factor = 1-Ac, so term = A*(1-Ac) = A-A*Ac */
       /* fnms(a,b,c,d) computes a = d - b*c */
       spe_fnms(f, term1A_reg, fragA_reg, constA_reg, fragA_reg);
@@ -703,9 +722,9 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_CONST_COLOR:
       /* We need the optional constant color registers */
-      setup_const_register(f, &constR_reg_set, &constR_reg, blend_color->color[0]);
-      setup_const_register(f, &constG_reg_set, &constG_reg, blend_color->color[1]);
-      setup_const_register(f, &constB_reg_set, &constB_reg, blend_color->color[2]);
+      setup_const_register(f, &constR_reg, blend_color->color[0]);
+      setup_const_register(f, &constG_reg, blend_color->color[1]);
+      setup_const_register(f, &constB_reg, blend_color->color[2]);
       /* now, factor = (Rc,Gc,Bc), so term = (Rfb*Rc,Gfb*Gc,Bfb*Bc) */
       spe_fm(f, term2R_reg, fbR_reg, constR_reg);
       spe_fm(f, term2G_reg, fbG_reg, constG_reg);
@@ -713,7 +732,7 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_CONST_ALPHA:
       /* we'll need the optional constant alpha register */
-      setup_const_register(f, &constA_reg_set, &constA_reg, blend_color->color[3]);
+      setup_const_register(f, &constA_reg, blend_color->color[3]);
       /* factor = (Ac,Ac,Ac), so term = (Rfb*Ac,Gfb*Ac,Bfb*Ac) */
       spe_fm(f, term2R_reg, fbR_reg, constA_reg);
       spe_fm(f, term2G_reg, fbG_reg, constA_reg);
@@ -721,9 +740,9 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
       /* We need the optional constant color registers */
-      setup_const_register(f, &constR_reg_set, &constR_reg, blend_color->color[0]);
-      setup_const_register(f, &constG_reg_set, &constG_reg, blend_color->color[1]);
-      setup_const_register(f, &constB_reg_set, &constB_reg, blend_color->color[2]);
+      setup_const_register(f, &constR_reg, blend_color->color[0]);
+      setup_const_register(f, &constG_reg, blend_color->color[1]);
+      setup_const_register(f, &constB_reg, blend_color->color[2]);
       /* factor = (1-Rc,1-Gc,1-Bc), so term = (Rfb*(1-Rc),Gfb*(1-Gc),Bfb*(1-Bc)) 
        * or term = (Rfb-Rfb*Rc, Gfb-Gfb*Gc, Bfb-Bfb*Bc)
        * fnms(a,b,c,d) computes a = d - b*c
@@ -734,9 +753,9 @@ gen_blend(const struct pipe_blend_state *blend,
       break;
    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
       /* We need the optional constant color registers */
-      setup_const_register(f, &constR_reg_set, &constR_reg, blend_color->color[0]);
-      setup_const_register(f, &constG_reg_set, &constG_reg, blend_color->color[1]);
-      setup_const_register(f, &constB_reg_set, &constB_reg, blend_color->color[2]);
+      setup_const_register(f, &constR_reg, blend_color->color[0]);
+      setup_const_register(f, &constG_reg, blend_color->color[1]);
+      setup_const_register(f, &constB_reg, blend_color->color[2]);
       /* factor = (1-Ac,1-Ac,1-Ac), so term = (Rfb*(1-Ac),Gfb*(1-Ac),Bfb*(1-Ac))
        * or term = (Rfb-Rfb*Ac,Gfb-Gfb*Ac,Bfb-Bfb*Ac)
        * fnms(a,b,c,d) computes a = d - b*c
@@ -806,7 +825,7 @@ gen_blend(const struct pipe_blend_state *blend,
    case PIPE_BLENDFACTOR_CONST_ALPHA: /* fall through */
    case PIPE_BLENDFACTOR_CONST_COLOR:
       /* We need the optional constA_reg register */
-      setup_const_register(f, &constA_reg_set, &constA_reg, blend_color->color[3]);
+      setup_const_register(f, &constA_reg, blend_color->color[3]);
       /* factor = Ac, so term = Afb*Ac */
       spe_fm(f, term2A_reg, fbA_reg, constA_reg);
       break;
@@ -814,7 +833,7 @@ gen_blend(const struct pipe_blend_state *blend,
    case PIPE_BLENDFACTOR_INV_CONST_ALPHA: /* fall through */
    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
       /* We need the optional constA_reg register */
-      setup_const_register(f, &constA_reg_set, &constA_reg, blend_color->color[3]);
+      setup_const_register(f, &constA_reg, blend_color->color[3]);
       /* factor = 1-Ac, so term = Afb*(1-Ac) = Afb-Afb*Ac */
       /* fnms(a,b,c,d) computes a = d - b*c */
       spe_fnms(f, term2A_reg, fbA_reg, constA_reg, fbA_reg);
@@ -910,11 +929,11 @@ gen_blend(const struct pipe_blend_state *blend,
    spe_release_register(f, tmp_reg);
 
    /* Free any optional registers that actually got used */
-   release_const_register(f, &one_reg_set, one_reg);
-   release_const_register(f, &constR_reg_set, constR_reg);
-   release_const_register(f, &constG_reg_set, constG_reg);
-   release_const_register(f, &constB_reg_set, constB_reg);
-   release_const_register(f, &constA_reg_set, constA_reg);
+   release_const_register(f, one_reg);
+   release_const_register(f, constR_reg);
+   release_const_register(f, constG_reg);
+   release_const_register(f, constB_reg);
+   release_const_register(f, constA_reg);
 }
 
 
@@ -1055,6 +1074,7 @@ gen_pack_colors(struct spe_function *f,
    spe_release_register(f, ba_reg);
 }
 
+
 static void
 gen_colormask(struct spe_function *f,
               uint colormask,
@@ -1067,10 +1087,10 @@ gen_colormask(struct spe_function *f,
     * are packed according to the given color format, not
     * necessarily RGBA...
     */
-   unsigned int r_mask;
-   unsigned int g_mask;
-   unsigned int b_mask;
-   unsigned int a_mask;
+   uint r_mask;
+   uint g_mask;
+   uint b_mask;
+   uint a_mask;
 
    /* Calculate exactly where the bits for any particular color
     * end up, so we can mask them correctly.
@@ -1111,11 +1131,13 @@ gen_colormask(struct spe_function *f,
       a_mask = 0;
    }
 
-   /* Get a temporary register to hold the mask that will be applied to the fragment */
+   /* Get a temporary register to hold the mask that will be applied
+    * to the fragment
+    */
    int colormask_reg = spe_allocate_available_register(f);
 
-   /* The actual mask we're going to use is an OR of the remaining R, G, B, and A
-    * masks.  Load the result value into our temporary register.
+   /* The actual mask we're going to use is an OR of the remaining R, G, B,
+    * and A masks.  Load the result value into our temporary register.
     */
    spe_load_uint(f, colormask_reg, r_mask | g_mask | b_mask | a_mask);
 
@@ -1135,7 +1157,9 @@ gen_colormask(struct spe_function *f,
     spe_release_register(f, colormask_reg);
 }
 
-/* This function is annoyingly similar to gen_depth_test(), above, except
+
+/**
+ * This function is annoyingly similar to gen_depth_test(), above, except
  * that instead of comparing two varying values (i.e. fragment and buffer),
  * we're comparing a varying value with a static value.  As such, we have
  * access to the Compare Immediate instructions where we don't in 
@@ -1146,16 +1170,20 @@ gen_colormask(struct spe_function *f,
  *
  * The return value in the stencil_pass_reg is a bitmask of valid
  * fragments that also passed the stencil test.  The bitmask of valid
- * fragments that failed would be found in (fragment_mask_reg & ~stencil_pass_reg).
+ * fragments that failed would be found in
+ * (fragment_mask_reg & ~stencil_pass_reg).
  */
 static void
-gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state, 
-                 unsigned int stencil_max_value,
-                 unsigned int fragment_mask_reg, unsigned int fbS_reg, 
-                 unsigned int stencil_pass_reg)
+gen_stencil_test(struct spe_function *f,
+                 const struct pipe_stencil_state *state, 
+                 uint stencil_max_value,
+                 int fragment_mask_reg,
+                 int fbS_reg, 
+                 int stencil_pass_reg)
 {
-   /* Generate code that puts the set of passing fragments into the stencil_pass_reg
-    * register, taking into account whether each fragment was active to begin with.
+   /* Generate code that puts the set of passing fragments into the
+    * stencil_pass_reg register, taking into account whether each fragment
+    * was active to begin with.
     */
    switch (state->func) {
    case PIPE_FUNC_EQUAL:
@@ -1166,9 +1194,10 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
       }
       else {
          /* stencil_pass = fragment_mask & ((s&mask) == (reference&mask)) */
-         unsigned int tmp_masked_stencil = spe_allocate_available_register(f);
+         uint tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
-         spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil, state->value_mask & state->ref_value);
+         spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil,
+                                state->value_mask & state->ref_value);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1182,9 +1211,10 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
       }
       else {
          /* stencil_pass = fragment_mask & ~((s&mask) == (reference&mask)) */
-         unsigned int tmp_masked_stencil = spe_allocate_available_register(f);
+         int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
-         spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil, state->value_mask & state->ref_value);
+         spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil,
+                                state->value_mask & state->ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1198,9 +1228,10 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
       }
       else {
          /* stencil_pass = fragment_mask & ((reference&mask) < (s & mask)) */
-         unsigned int tmp_masked_stencil = spe_allocate_available_register(f);
+         int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
-         spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil, state->value_mask & state->ref_value);
+         spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil,
+                                  state->value_mask & state->ref_value);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1214,7 +1245,7 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
           * comparing directly.  Compare Logical Greater Than Word (clgt) 
           * treats its operands as unsigned - no sign extension.
           */
-         unsigned int tmp_reg = spe_allocate_available_register(f);
+         int tmp_reg = spe_allocate_available_register(f);
          spe_load_uint(f, tmp_reg, state->ref_value);
          spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
@@ -1222,8 +1253,8 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
       }
       else {
          /* stencil_pass = fragment_mask & ((reference&mask) > (s&mask)) */
-         unsigned int tmp_reg = spe_allocate_available_register(f);
-         unsigned int tmp_masked_stencil = spe_allocate_available_register(f);
+         int tmp_reg = spe_allocate_available_register(f);
+         int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_load_uint(f, tmp_reg, state->value_mask & state->ref_value);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
          spe_clgt(f, stencil_pass_reg, tmp_reg, tmp_masked_stencil);
@@ -1237,14 +1268,16 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
       if (state->value_mask == stencil_max_value) {
          /* stencil_pass = fragment_mask & (reference >= s) 
           *              = fragment_mask & ~(s > reference) */
-         spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+         spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg,
+                                  state->ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
       }
       else {
          /* stencil_pass = fragment_mask & ~((s&mask) > (reference&mask)) */
-         unsigned int tmp_masked_stencil = spe_allocate_available_register(f);
+         int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
-         spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil, state->value_mask & state->ref_value);
+         spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil,
+                                  state->value_mask & state->ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1255,7 +1288,7 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
          /* stencil_pass = fragment_mask & (reference <= s) ]
           *               = fragment_mask & ~(reference > s) */
          /* As above, we have to do this by loading a register */
-         unsigned int tmp_reg = spe_allocate_available_register(f);
+         int tmp_reg = spe_allocate_available_register(f);
          spe_load_uint(f, tmp_reg, state->ref_value);
          spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
@@ -1263,8 +1296,8 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
       }
       else {
          /* stencil_pass = fragment_mask & ~((reference&mask) > (s&mask)) */
-         unsigned int tmp_reg = spe_allocate_available_register(f);
-         unsigned int tmp_masked_stencil = spe_allocate_available_register(f);
+         int tmp_reg = spe_allocate_available_register(f);
+         int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_load_uint(f, tmp_reg, state->ref_value & state->value_mask);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
          spe_clgt(f, stencil_pass_reg, tmp_reg, tmp_masked_stencil);
@@ -1290,7 +1323,9 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
     */
 }
 
-/* This function generates code that calculates a set of new stencil values
+
+/**
+ * This function generates code that calculates a set of new stencil values
  * given the earlier values and the operation to apply.  It does not
  * apply any tests.  It is intended to be called up to 3 times
  * (for the stencil fail operation, for the stencil pass-z fail operation,
@@ -1302,9 +1337,12 @@ gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state,
  * in the stencil buffer - in other words, it should be usable as a mask.
  */
 static void
-gen_stencil_values(struct spe_function *f, unsigned int stencil_op,
-                   unsigned int stencil_ref_value, unsigned int stencil_max_value,
-                   unsigned int fbS_reg, unsigned int newS_reg)
+gen_stencil_values(struct spe_function *f,
+                   uint stencil_op,
+                   uint stencil_ref_value,
+                   uint stencil_max_value,
+                   int fbS_reg,
+                   int newS_reg)
 {
    /* The code below assumes that newS_reg and fbS_reg are not the same
     * register; if they can be, the calculations below will have to use
@@ -1346,7 +1384,7 @@ gen_stencil_values(struct spe_function *f, unsigned int stencil_op,
 
    case PIPE_STENCIL_OP_INCR: {
       /* newS = (s == max ? max : s + 1) */
-      unsigned int equals_reg = spe_allocate_available_register(f);
+      int equals_reg = spe_allocate_available_register(f);
 
       spe_compare_equal_uint(f, equals_reg, fbS_reg, stencil_max_value);
       /* Add Word Immediate computes rT = rA + 10-bit signed immediate */
@@ -1359,7 +1397,7 @@ gen_stencil_values(struct spe_function *f, unsigned int stencil_op,
    }
    case PIPE_STENCIL_OP_DECR: {
       /* newS = (s == 0 ? 0 : s - 1) */
-      unsigned int equals_reg = spe_allocate_available_register(f);
+      int equals_reg = spe_allocate_available_register(f);
 
       spe_compare_equal_uint(f, equals_reg, fbS_reg, 0);
       /* Add Word Immediate with a (-1) value works */
@@ -1397,7 +1435,8 @@ gen_stencil_values(struct spe_function *f, unsigned int stencil_op,
 }
 
 
-/* This function generates code to get all the necessary possible
+/**
+ * This function generates code to get all the necessary possible
  * stencil values.  For each of the output registers (fail_reg,
  * zfail_reg, and zpass_reg), it either allocates a new register
  * and calculates a new set of values based on the stencil operation,
@@ -1412,13 +1451,15 @@ gen_stencil_values(struct spe_function *f, unsigned int stencil_op,
  * and released by the corresponding spe_release_register_set() call.
  */
 static void
-gen_get_stencil_values(struct spe_function *f, const struct pipe_stencil_state *stencil,
-                       const unsigned int depth_enabled,
-                       unsigned int fbS_reg, 
-                       unsigned int *fail_reg, unsigned int *zfail_reg, 
-                       unsigned int *zpass_reg)
+gen_get_stencil_values(struct spe_function *f,
+                       const struct pipe_stencil_state *stencil,
+                       const uint depth_enabled,
+                       int fbS_reg, 
+                       int *fail_reg,
+                       int *zfail_reg, 
+                       int *zpass_reg)
 {
-   unsigned zfail_op;
+   uint zfail_op;
 
    /* Stenciling had better be enabled here */
    ASSERT(stencil->enabled);
@@ -1480,7 +1521,8 @@ gen_get_stencil_values(struct spe_function *f, const struct pipe_stencil_state *
    }
 }
 
-/* Note that fbZ_reg may *not* be set on entry, if in fact
+/**
+ * Note that fbZ_reg may *not* be set on entry, if in fact
  * the depth test is not enabled.  This function must not use
  * the register if depth is not enabled.
  */
@@ -1494,7 +1536,7 @@ gen_stencil_depth_test(struct spe_function *f,
    /* True if we've generated code that could require writeback to the
     * depth and/or stencil buffers
     */
-   boolean modified_buffers = false;
+   boolean modified_buffers = FALSE;
 
    boolean need_to_calculate_stencil_values;
    boolean need_to_writemask_stencil_values;
@@ -1504,11 +1546,11 @@ gen_stencil_depth_test(struct spe_function *f,
    /* Registers.  We may or may not actually allocate these, depending
     * on whether the state values indicate that we need them.
     */
-   unsigned int stencil_pass_reg, stencil_fail_reg;
-   unsigned int stencil_fail_values, stencil_pass_depth_fail_values, stencil_pass_depth_pass_values;
-   unsigned int stencil_writemask_reg;
-   unsigned int zmask_reg;
-   unsigned int newS_reg;
+   int stencil_pass_reg, stencil_fail_reg;
+   int stencil_fail_values, stencil_pass_depth_fail_values, stencil_pass_depth_pass_values;
+   int stencil_writemask_reg;
+   int zmask_reg;
+   int newS_reg;
 
    /* Stenciling is quite complex: up to six different configurable stencil 
     * operations/calculations can be required (three each for front-facing
@@ -1555,27 +1597,27 @@ gen_stencil_depth_test(struct spe_function *f,
    if (stencil->fail_op == PIPE_STENCIL_OP_KEEP &&
        stencil->zfail_op == PIPE_STENCIL_OP_KEEP &&
        stencil->zpass_op == PIPE_STENCIL_OP_KEEP) {
-       need_to_calculate_stencil_values = false;
-       need_to_writemask_stencil_values = false;
+       need_to_calculate_stencil_values = FALSE;
+       need_to_writemask_stencil_values = FALSE;
     }
     else if (stencil->write_mask == 0x0) {
       /* All changes are writemasked out, so no need to calculate
        * what those changes might be, and no need to write anything back.
        */
-      need_to_calculate_stencil_values = false;
-      need_to_writemask_stencil_values = false;
+      need_to_calculate_stencil_values = FALSE;
+      need_to_writemask_stencil_values = FALSE;
    }
    else if (stencil->write_mask == 0xff) {
       /* Still trivial, but a little less so.  We need to write the stencil
        * values, but we don't need to mask them.
        */
-      need_to_calculate_stencil_values = true;
-      need_to_writemask_stencil_values = false;
+      need_to_calculate_stencil_values = TRUE;
+      need_to_writemask_stencil_values = FALSE;
    }
    else {
       /* The general case: calculate, mask, and write */
-      need_to_calculate_stencil_values = true;
-      need_to_writemask_stencil_values = true;
+      need_to_calculate_stencil_values = TRUE;
+      need_to_writemask_stencil_values = TRUE;
 
       /* While we're here, generate code that calculates what the
        * writemask should be.  If backface stenciling is enabled,
@@ -1633,7 +1675,9 @@ gen_stencil_depth_test(struct spe_function *f,
        * This function will allocate a variant number of registers that
        * will be released as part of the register set.
        */
-      spe_comment(f, 0, facing == CELL_FACING_FRONT ? "Computing front-facing stencil values" : "Computing back-facing stencil values");
+      spe_comment(f, 0, facing == CELL_FACING_FRONT
+                  ? "Computing front-facing stencil values"
+                  : "Computing back-facing stencil values");
       gen_get_stencil_values(f, stencil, dsa->depth.enabled, fbS_reg, 
          &stencil_fail_values, &stencil_pass_depth_fail_values, 
          &stencil_pass_depth_pass_values);
@@ -1652,7 +1696,8 @@ gen_stencil_depth_test(struct spe_function *f,
    if (dsa->depth.enabled) {
       spe_comment(f, 0, "Running stencil depth test");
       zmask_reg = spe_allocate_available_register(f);
-      modified_buffers |= gen_depth_test(f, dsa, mask_reg, fragZ_reg, fbZ_reg, zmask_reg);
+      modified_buffers |= gen_depth_test(f, dsa, mask_reg, fragZ_reg,
+                                         fbZ_reg, zmask_reg);
    }
 
    if (need_to_calculate_stencil_values) {
@@ -1675,7 +1720,7 @@ gen_stencil_depth_test(struct spe_function *f,
       if (stencil_fail_values != fbS_reg) {
          spe_comment(f, 0, "Loading stencil fail values");
          spe_selb(f, newS_reg, newS_reg, stencil_fail_values, stencil_fail_reg);
-         modified_buffers = true;
+         modified_buffers = TRUE;
       }
 
       /* Same for the stencil pass/depth fail values.  If this calculation
@@ -1689,14 +1734,17 @@ gen_stencil_depth_test(struct spe_function *f,
           * depth passing mask.  Note that zmask_reg *must* have been
           * set above if we're here.
           */
-         unsigned int stencil_pass_depth_fail_mask = spe_allocate_available_register(f);
+         uint stencil_pass_depth_fail_mask =
+            spe_allocate_available_register(f);
+
          spe_comment(f, 0, "Loading stencil pass/depth fail values");
          spe_andc(f, stencil_pass_depth_fail_mask, stencil_pass_reg, zmask_reg);
 
-         spe_selb(f, newS_reg, newS_reg, stencil_pass_depth_fail_values, stencil_pass_depth_fail_mask);
+         spe_selb(f, newS_reg, newS_reg, stencil_pass_depth_fail_values,
+                  stencil_pass_depth_fail_mask);
 
          spe_release_register(f, stencil_pass_depth_fail_mask);
-         modified_buffers = true;
+         modified_buffers = TRUE;
       }
 
       /* Same for the stencil pass/depth pass mask.  Note that we
@@ -1707,7 +1755,7 @@ gen_stencil_depth_test(struct spe_function *f,
        */
       if (stencil_pass_depth_pass_values != fbS_reg) {
          if (dsa->depth.enabled) {
-            unsigned int stencil_pass_depth_pass_mask = spe_allocate_available_register(f);
+            uint stencil_pass_depth_pass_mask = spe_allocate_available_register(f);
             /* We'll need a separate register */
             spe_comment(f, 0, "Loading stencil pass/depth pass values");
             spe_and(f, stencil_pass_depth_pass_mask, stencil_pass_reg, zmask_reg);
@@ -1719,7 +1767,7 @@ gen_stencil_depth_test(struct spe_function *f,
             spe_comment(f, 0, "Loading stencil pass values");
             spe_selb(f, newS_reg, newS_reg, stencil_pass_depth_pass_values, stencil_pass_reg);
          }
-         modified_buffers = true;
+         modified_buffers = TRUE;
       }
 
       /* Almost done.  If we need to writemask, do it now, leaving the
@@ -1749,7 +1797,7 @@ gen_stencil_depth_test(struct spe_function *f,
    spe_comment(f, 0, "Releasing stencil register set");
    spe_release_register_set(f);
 
-   /* Return true if we could have modified the stencil and/or
+   /* Return TRUE if we could have modified the stencil and/or
     * depth buffers.
     */
    return modified_buffers;
@@ -1757,6 +1805,200 @@ gen_stencil_depth_test(struct spe_function *f,
 
 
 /**
+ * Generate depth and/or stencil test code.
+ * \param cell  context
+ * \param dsa  depth/stencil/alpha state
+ * \param f  spe function to emit
+ * \param facing  either CELL_FACING_FRONT or CELL_FACING_BACK
+ * \param mask_reg  register containing the pixel alive/dead mask
+ * \param depth_tile_reg  register containing address of z/stencil tile
+ * \param quad_offset_reg  offset to quad from start of tile
+ * \param fragZ_reg  register containg fragment Z values
+ */
+static void
+gen_depth_stencil(struct cell_context *cell,
+                  const struct pipe_depth_stencil_alpha_state *dsa,
+                  struct spe_function *f,
+                  uint facing,
+                  int mask_reg,
+                  int depth_tile_reg,
+                  int quad_offset_reg,
+                  int fragZ_reg)
+
+{
+   const enum pipe_format zs_format = cell->framebuffer.zsbuf->format;
+   boolean write_depth_stencil;
+
+   /* framebuffer's combined z/stencil values register */
+   int fbZS_reg = spe_allocate_available_register(f);
+
+   /* Framebufer Z values register */
+   int fbZ_reg = spe_allocate_available_register(f);
+
+   /* Framebuffer stencil values register (may not be used) */
+   int fbS_reg = spe_allocate_available_register(f);
+
+   /* 24-bit mask register (may not be used) */
+   int zmask_reg = spe_allocate_available_register(f);
+
+   /**
+    * The following code:
+    * 1. fetch quad of packed Z/S values from the framebuffer tile.
+    * 2. extract the separate the Z and S values from packed values
+    * 3. convert fragment Z values from float in [0,1] to 32/24/16-bit ints
+    *
+    * The instructions for doing this are interleaved for better performance.
+    */
+   spe_comment(f, 0, "Fetch Z/stencil quad from tile");
+
+   switch(zs_format) {
+   case PIPE_FORMAT_S8Z24_UNORM: /* fall through */
+   case PIPE_FORMAT_X8Z24_UNORM:
+      /* prepare mask to extract Z vals from ZS vals */
+      spe_load_uint(f, zmask_reg, 0x00ffffff);
+
+      /* convert fragment Z from [0,1] to 32-bit ints */
+      spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
+
+      /* Load: fbZS_reg = memory[depth_tile_reg + offset_reg] */
+      spe_lqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg);
+
+      /* right shift 32-bit fragment Z to 24 bits */
+      spe_rotmi(f, fragZ_reg, fragZ_reg, -8);
+
+      /* extract 24-bit Z values from ZS values by masking */
+      spe_and(f, fbZ_reg, fbZS_reg, zmask_reg);
+
+      /* extract 8-bit stencil values by shifting */
+      spe_rotmi(f, fbS_reg, fbZS_reg, -24);
+      break;
+
+   case PIPE_FORMAT_Z24S8_UNORM: /* fall through */
+   case PIPE_FORMAT_Z24X8_UNORM:
+      /* convert fragment Z from [0,1] to 32-bit ints */
+      spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
+
+      /* Load: fbZS_reg = memory[depth_tile_reg + offset_reg] */
+      spe_lqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg);
+
+      /* right shift 32-bit fragment Z to 24 bits */
+      spe_rotmi(f, fragZ_reg, fragZ_reg, -8);
+
+      /* extract 24-bit Z values from ZS values by shifting */
+      spe_rotmi(f, fbZ_reg, fbZS_reg, -8);
+
+      /* extract 8-bit stencil values by masking */
+      spe_and_uint(f, fbS_reg, fbZS_reg, 0x000000ff);
+      break;
+
+   case PIPE_FORMAT_Z32_UNORM:
+      /* Load: fbZ_reg = memory[depth_tile_reg + offset_reg] */
+      spe_lqx(f, fbZ_reg, depth_tile_reg, quad_offset_reg);
+
+      /* convert fragment Z from [0,1] to 32-bit ints */
+      spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
+
+      /* No stencil, so can't do anything there */
+      break;
+
+   case PIPE_FORMAT_Z16_UNORM:
+      /* XXX This code for 16bpp Z is broken! */
+
+      /* Load: fbZS_reg = memory[depth_tile_reg + offset_reg] */
+      spe_lqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg);
+
+      /* Copy over 4 32-bit values */
+      spe_move(f, fbZ_reg, fbZS_reg);
+
+      /* convert Z from [0,1] to 16-bit ints */
+      spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
+      spe_rotmi(f, fragZ_reg, fragZ_reg, -16);
+      /* No stencil */
+      break;
+
+   default:
+      ASSERT(0); /* invalid format */
+   }
+
+   /* If stencil is enabled, use the stencil-specific code
+    * generator to generate both the stencil and depth (if needed)
+    * tests.  Otherwise, if only depth is enabled, generate
+    * a quick depth test.  The test generators themselves will
+    * report back whether the depth/stencil buffer has to be
+    * written back.
+    */
+   if (dsa->stencil[0].enabled) {
+      /* This will perform the stencil and depth tests, and update
+       * the mask_reg, fbZ_reg, and fbS_reg as required by the
+       * tests.
+       */
+      ASSERT(fbS_reg >= 0);
+      spe_comment(f, 0, "Perform stencil test");
+
+      /* Note that fbZ_reg may not be set on entry, if stenciling
+       * is enabled but there's no Z-buffer.  The 
+       * gen_stencil_depth_test() function must ignore the
+       * fbZ_reg register if depth is not enabled.
+       */
+      write_depth_stencil = gen_stencil_depth_test(f, dsa, facing,
+                                                   mask_reg, fragZ_reg,
+                                                   fbZ_reg, fbS_reg);
+   }
+   else if (dsa->depth.enabled) {
+      int zmask_reg = spe_allocate_available_register(f);
+      ASSERT(fbZ_reg >= 0);
+      spe_comment(f, 0, "Perform depth test");
+      write_depth_stencil = gen_depth_test(f, dsa, mask_reg, fragZ_reg,
+                                           fbZ_reg, zmask_reg);
+      spe_release_register(f, zmask_reg);
+   }
+   else {
+      write_depth_stencil = FALSE;
+   }
+
+   if (write_depth_stencil) {
+      /* Merge latest Z and Stencil values into fbZS_reg.
+       * fbZ_reg has four Z vals in bits [23..0] or bits [15..0].
+       * fbS_reg has four 8-bit Z values in bits [7..0].
+       */
+      spe_comment(f, 0, "Store quad's depth/stencil values in tile");
+      if (zs_format == PIPE_FORMAT_S8Z24_UNORM ||
+          zs_format == PIPE_FORMAT_X8Z24_UNORM) {
+         spe_shli(f, fbS_reg, fbS_reg, 24); /* fbS = fbS << 24 */
+         spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
+      }
+      else if (zs_format == PIPE_FORMAT_Z24S8_UNORM ||
+               zs_format == PIPE_FORMAT_Z24X8_UNORM) {
+         spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */
+         spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
+      }
+      else if (zs_format == PIPE_FORMAT_Z32_UNORM) {
+         spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */
+      }
+      else if (zs_format == PIPE_FORMAT_Z16_UNORM) {
+         spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */
+      }
+      else if (zs_format == PIPE_FORMAT_S8_UNORM) {
+         ASSERT(0);   /* XXX to do */
+      }
+      else {
+         ASSERT(0); /* bad zs_format */
+      }
+
+      /* Store: memory[depth_tile_reg + quad_offset_reg] = fbZS */
+      spe_stqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg);
+   }
+
+   /* Don't need these any more */
+   spe_release_register(f, fbZS_reg);
+   spe_release_register(f, fbZ_reg);
+   spe_release_register(f, fbS_reg);
+   spe_release_register(f, zmask_reg);
+}
+
+
+
+/**
  * Generate SPE code to implement the fragment operations (alpha test,
  * depth test, stencil test, blending, colormask, and final
  * framebuffer write) as specified by the current context state.
@@ -1782,7 +2024,9 @@ gen_stencil_depth_test(struct spe_function *f,
  *              the fragment ops appended.
  */
 void
-cell_gen_fragment_function(struct cell_context *cell, const uint facing, struct spe_function *f)
+cell_gen_fragment_function(struct cell_context *cell,
+                           const uint facing,
+                           struct spe_function *f)
 {
    const struct pipe_depth_stencil_alpha_state *dsa = cell->depth_stencil;
    const struct pipe_blend_state *blend = cell->blend;
@@ -1809,12 +2053,13 @@ cell_gen_fragment_function(struct cell_context *cell, const uint facing, struct
    int quad_offset_reg;
 
    int fbRGBA_reg;  /**< framebuffer's RGBA colors for quad */
-   int fbZS_reg;    /**< framebuffer's combined z/stencil values for quad */
 
    if (cell->debug_flags & CELL_DEBUG_ASM) {
-      spe_print_code(f, true);
+      spe_print_code(f, TRUE);
       spe_indent(f, 8);
-      spe_comment(f, -4, facing == CELL_FACING_FRONT ? "Begin front-facing per-fragment ops": "Begin back-facing per-fragment ops");
+      spe_comment(f, -4, facing == CELL_FACING_FRONT
+                  ? "Begin front-facing per-fragment ops"
+                  : "Begin back-facing per-fragment ops");
    }
 
    spe_allocate_register(f, x_reg);
@@ -1830,7 +2075,6 @@ cell_gen_fragment_function(struct cell_context *cell, const uint facing, struct
 
    quad_offset_reg = spe_allocate_available_register(f);
    fbRGBA_reg = spe_allocate_available_register(f);
-   fbZS_reg = spe_allocate_available_register(f);
 
    /* compute offset of quad from start of tile, in bytes */
    {
@@ -1855,177 +2099,14 @@ cell_gen_fragment_function(struct cell_context *cell, const uint facing, struct
       gen_alpha_test(dsa, f, mask_reg, fragA_reg);
    }
 
-   /* If we need the stencil buffers (because one- or two-sided stencil is
-    * enabled) or the depth buffer (because the depth test is enabled),
-    * go grab them.  Note that if either one- or two-sided stencil is
-    * enabled, dsa->stencil[0].enabled will be true.
-    */
+   /* generate depth and/or stencil test code */
    if (dsa->depth.enabled || dsa->stencil[0].enabled) {
-      const enum pipe_format zs_format = cell->framebuffer.zsbuf->format;
-      boolean write_depth_stencil;
-
-      /* We may or may not need to allocate a register for Z or stencil values */
-      boolean fbS_reg_set = false, fbZ_reg_set = false;
-      unsigned int fbS_reg, fbZ_reg = 0;
-
-      spe_comment(f, 0, "Fetching Z/stencil quad from tile");
-
-      /* fetch quad of depth/stencil values from tile at (x,y) */
-      /* Load: fbZS_reg = memory[depth_tile_reg + offset_reg] */
-      /* XXX Not sure this is allowed if we've only got a 16-bit Z buffer... */
-      spe_lqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg);
-
-      /* From the Z/stencil buffer format, pull out the bits we need for
-       * Z and/or stencil.  We'll also convert the incoming fragment Z
-       * value in fragZ_reg from a floating point value in [0.0..1.0] to
-       * an unsigned integer value with the appropriate resolution.
-       * Note that even if depth or stencil is *not* enabled, if it's
-       * present in the buffer, we pull it out and put it back later;
-       * otherwise, we can inadvertently destroy the contents of
-       * buffers we're not supposed to touch (e.g., if the user is
-       * clearing the depth buffer but not the stencil buffer, a
-       * quad of constant depth is drawn over the surface; the stencil
-       * buffer must be maintained).
-       */
-      switch(zs_format) {
-
-         case PIPE_FORMAT_S8Z24_UNORM: /* fall through */
-         case PIPE_FORMAT_X8Z24_UNORM:
-            /* Pull out both Z and stencil */
-            setup_optional_register(f, &fbZ_reg_set, &fbZ_reg);
-            setup_optional_register(f, &fbS_reg_set, &fbS_reg);
-
-            /* four 24-bit Z values in the low-order bits */
-            spe_and_uint(f, fbZ_reg, fbZS_reg, 0x00ffffff);
-
-            /* Incoming fragZ_reg value is a float in 0.0...1.0; convert
-             * to a 24-bit unsigned integer
-             */
-            spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
-            spe_rotmi(f, fragZ_reg, fragZ_reg, -8);
-
-            /* four 8-bit stencil values in the high-order bits */
-            spe_rotmi(f, fbS_reg, fbZS_reg, -24);
-         break;
-
-         case PIPE_FORMAT_Z24S8_UNORM: /* fall through */
-         case PIPE_FORMAT_Z24X8_UNORM:
-            setup_optional_register(f, &fbZ_reg_set, &fbZ_reg);
-            setup_optional_register(f, &fbS_reg_set, &fbS_reg);
-
-            /* shift by 8 to get the upper 24-bit values */
-            spe_rotmi(f, fbS_reg, fbZS_reg, -8);
-
-            /* Incoming fragZ_reg value is a float in 0.0...1.0; convert
-             * to a 24-bit unsigned integer
-             */
-            spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
-            spe_rotmi(f, fragZ_reg, fragZ_reg, -8);
-
-            /* 8-bit stencil in the low-order bits - mask them out */
-            spe_and_uint(f, fbS_reg, fbZS_reg, 0x000000ff);
-         break;
-
-         case PIPE_FORMAT_Z32_UNORM:
-            setup_optional_register(f, &fbZ_reg_set, &fbZ_reg);
-            /* Copy over 4 32-bit values */
-            spe_move(f, fbZ_reg, fbZS_reg);
-
-            /* Incoming fragZ_reg value is a float in 0.0...1.0; convert
-             * to a 32-bit unsigned integer
-             */
-            spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
-            /* No stencil, so can't do anything there */
-         break;
-
-         case PIPE_FORMAT_Z16_UNORM:
-            /* XXX Not sure this is correct, but it was here before, so we're
-             * going with it for now
-             */
-            setup_optional_register(f, &fbZ_reg_set, &fbZ_reg);
-            /* Copy over 4 32-bit values */
-            spe_move(f, fbZ_reg, fbZS_reg);
-
-            /* Incoming fragZ_reg value is a float in 0.0...1.0; convert
-             * to a 16-bit unsigned integer
-             */
-            spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
-            spe_rotmi(f, fragZ_reg, fragZ_reg, -16);
-            /* No stencil */
-
-         default:
-            ASSERT(0); /* invalid format */
-      }
-
-      /* If stencil is enabled, use the stencil-specific code
-       * generator to generate both the stencil and depth (if needed)
-       * tests.  Otherwise, if only depth is enabled, generate
-       * a quick depth test.  The test generators themselves will
-       * report back whether the depth/stencil buffer has to be
-       * written back.
-       */
-      if (dsa->stencil[0].enabled) {
-         /* This will perform the stencil and depth tests, and update
-          * the mask_reg, fbZ_reg, and fbS_reg as required by the
-          * tests.
-          */
-         ASSERT(fbS_reg_set);
-         spe_comment(f, 0, "Perform stencil test");
-
-         /* Note that fbZ_reg may not be set on entry, if stenciling
-          * is enabled but there's no Z-buffer.  The 
-          * gen_stencil_depth_test() function must ignore the
-          * fbZ_reg register if depth is not enabled.
-          */
-         write_depth_stencil = gen_stencil_depth_test(f, dsa, facing, mask_reg, fragZ_reg, fbZ_reg, fbS_reg);
-      }
-      else if (dsa->depth.enabled) {
-         int zmask_reg = spe_allocate_available_register(f);
-         ASSERT(fbZ_reg_set);
-         spe_comment(f, 0, "Perform depth test");
-         write_depth_stencil = gen_depth_test(f, dsa, mask_reg, fragZ_reg, fbZ_reg, zmask_reg);
-         spe_release_register(f, zmask_reg);
-      }
-      else {
-         write_depth_stencil = false;
-      }
-
-      if (write_depth_stencil) {
-         /* Merge latest Z and Stencil values into fbZS_reg.
-          * fbZ_reg has four Z vals in bits [23..0] or bits [15..0].
-          * fbS_reg has four 8-bit Z values in bits [7..0].
-          */
-         spe_comment(f, 0, "Store quad's depth/stencil values in tile");
-         if (zs_format == PIPE_FORMAT_S8Z24_UNORM ||
-             zs_format == PIPE_FORMAT_X8Z24_UNORM) {
-            spe_shli(f, fbS_reg, fbS_reg, 24); /* fbS = fbS << 24 */
-            spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
-         }
-         else if (zs_format == PIPE_FORMAT_Z24S8_UNORM ||
-                  zs_format == PIPE_FORMAT_Z24X8_UNORM) {
-            spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */
-            spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
-         }
-         else if (zs_format == PIPE_FORMAT_Z32_UNORM) {
-            spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */
-         }
-         else if (zs_format == PIPE_FORMAT_Z16_UNORM) {
-            spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */
-         }
-         else if (zs_format == PIPE_FORMAT_S8_UNORM) {
-            ASSERT(0);   /* XXX to do */
-         }
-         else {
-            ASSERT(0); /* bad zs_format */
-         }
-
-         /* Store: memory[depth_tile_reg + quad_offset_reg] = fbZS */
-         spe_stqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg);
-      }
-
-      /* Don't need these any more */
-      release_optional_register(f, &fbZ_reg_set, fbZ_reg);
-      release_optional_register(f, &fbS_reg_set, fbS_reg);
+      gen_depth_stencil(cell, dsa, f,
+                        facing,
+                        mask_reg,
+                        depth_tile_reg,
+                        quad_offset_reg,
+                        fragZ_reg);
    }
 
    /* Get framebuffer quad/colors.  We'll need these for blending,
@@ -2089,7 +2170,6 @@ cell_gen_fragment_function(struct cell_context *cell, const uint facing, struct
    spe_bi(f, SPE_REG_RA, 0, 0);  /* return from function call */
 
    spe_release_register(f, fbRGBA_reg);
-   spe_release_register(f, fbZS_reg);
    spe_release_register(f, quad_offset_reg);
 
    if (cell->debug_flags & CELL_DEBUG_ASM) {
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 0a0af81f53f..39b85faeb86 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -133,7 +133,7 @@ lookup_fragment_ops(struct cell_context *cell)
        */
       ops = CALLOC_VARIANT_LENGTH_STRUCT(cell_command_fragment_ops, total_code_size);
       /* populate the new cell_command_fragment_ops object */
-      ops->opcode = CELL_CMD_STATE_FRAGMENT_OPS;
+      ops->opcode[0] = CELL_CMD_STATE_FRAGMENT_OPS;
       ops->total_code_size = total_code_size;
       ops->front_code_index = 0;
       memcpy(ops->code, spe_code_front.store, front_code_size);
@@ -178,10 +178,10 @@ static void
 emit_state_cmd(struct cell_context *cell, uint cmd,
                const void *state, uint state_size)
 {
-   uint64_t *dst = (uint64_t *) 
-       cell_batch_alloc(cell, ROUNDUP8(sizeof(uint64_t) + state_size));
+   uint32_t *dst = (uint32_t *) 
+       cell_batch_alloc16(cell, ROUNDUP16(sizeof(opcode_t) + state_size));
    *dst = cmd;
-   memcpy(dst + 1, state, state_size);
+   memcpy(dst + 4, state, state_size);
 }
 
 
@@ -195,9 +195,10 @@ cell_emit_state(struct cell_context *cell)
    if (cell->dirty & CELL_NEW_FRAMEBUFFER) {
       struct pipe_surface *cbuf = cell->framebuffer.cbufs[0];
       struct pipe_surface *zbuf = cell->framebuffer.zsbuf;
+      STATIC_ASSERT(sizeof(struct cell_command_framebuffer) % 16 == 0);
       struct cell_command_framebuffer *fb
-         = cell_batch_alloc(cell, sizeof(*fb));
-      fb->opcode = CELL_CMD_STATE_FRAMEBUFFER;
+         = cell_batch_alloc16(cell, sizeof(*fb));
+      fb->opcode[0] = CELL_CMD_STATE_FRAMEBUFFER;
       fb->color_start = cell->cbuf_map[0];
       fb->color_format = cbuf->format;
       fb->depth_start = cell->zsbuf_map;
@@ -211,17 +212,19 @@ cell_emit_state(struct cell_context *cell)
    }
 
    if (cell->dirty & (CELL_NEW_RASTERIZER)) {
+      STATIC_ASSERT(sizeof(struct cell_command_rasterizer) % 16 == 0);
       struct cell_command_rasterizer *rast =
-         cell_batch_alloc(cell, sizeof(*rast));
-      rast->opcode = CELL_CMD_STATE_RASTERIZER;
+         cell_batch_alloc16(cell, sizeof(*rast));
+      rast->opcode[0] = CELL_CMD_STATE_RASTERIZER;
       rast->rasterizer = *cell->rasterizer;
    }
 
    if (cell->dirty & (CELL_NEW_FS)) {
       /* Send new fragment program to SPUs */
+      STATIC_ASSERT(sizeof(struct cell_command_fragment_program) % 16 == 0);
       struct cell_command_fragment_program *fp
-            = cell_batch_alloc(cell, sizeof(*fp));
-      fp->opcode = CELL_CMD_STATE_FRAGMENT_PROGRAM;
+            = cell_batch_alloc16(cell, sizeof(*fp));
+      fp->opcode[0] = CELL_CMD_STATE_FRAGMENT_PROGRAM;
       fp->num_inst = cell->fs->code.num_inst;
       memcpy(&fp->code, cell->fs->code.store,
              SPU_MAX_FRAGMENT_PROGRAM_INSTS * SPE_INST_SIZE);
@@ -238,14 +241,14 @@ cell_emit_state(struct cell_context *cell)
       const uint shader = PIPE_SHADER_FRAGMENT;
       const uint num_const = cell->constants[shader].size / sizeof(float);
       uint i, j;
-      float *buf = cell_batch_alloc(cell, 16 + num_const * sizeof(float));
-      uint64_t *ibuf = (uint64_t *) buf;
+      float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float)));
+      uint32_t *ibuf = (uint32_t *) buf;
       const float *constants = pipe_buffer_map(cell->pipe.screen,
                                                cell->constants[shader].buffer,
                                                PIPE_BUFFER_USAGE_CPU_READ);
       ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS;
-      ibuf[1] = num_const;
-      j = 4;
+      ibuf[4] = num_const;
+      j = 8;
       for (i = 0; i < num_const; i++) {
          buf[j++] = constants[i];
       }
@@ -258,7 +261,7 @@ cell_emit_state(struct cell_context *cell)
       struct cell_command_fragment_ops *fops, *fops_cmd;
       /* Note that cell_command_fragment_ops is a variant-sized record */
       fops = lookup_fragment_ops(cell);
-      fops_cmd = cell_batch_alloc(cell, sizeof(*fops_cmd) + fops->total_code_size);
+      fops_cmd = cell_batch_alloc16(cell, ROUNDUP16(sizeof(*fops_cmd) + fops->total_code_size));
       memcpy(fops_cmd, fops, sizeof(*fops) + fops->total_code_size);
    }
 
@@ -267,9 +270,10 @@ cell_emit_state(struct cell_context *cell)
       for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
          if (cell->dirty_samplers & (1 << i)) {
             if (cell->sampler[i]) {
+               STATIC_ASSERT(sizeof(struct cell_command_sampler) % 16 == 0);
                struct cell_command_sampler *sampler
-                  = cell_batch_alloc(cell, sizeof(*sampler));
-               sampler->opcode = CELL_CMD_STATE_SAMPLER;
+                  = cell_batch_alloc16(cell, sizeof(*sampler));
+               sampler->opcode[0] = CELL_CMD_STATE_SAMPLER;
                sampler->unit = i;
                sampler->state = *cell->sampler[i];
             }
@@ -282,9 +286,10 @@ cell_emit_state(struct cell_context *cell)
       uint i;
       for (i = 0;i < CELL_MAX_SAMPLERS; i++) {
          if (cell->dirty_textures & (1 << i)) {
+            STATIC_ASSERT(sizeof(struct cell_command_texture) % 16 == 0);
             struct cell_command_texture *texture
-               =  cell_batch_alloc(cell, sizeof(*texture));
-            texture->opcode = CELL_CMD_STATE_TEXTURE;
+               =  (struct cell_command_texture *)cell_batch_alloc16(cell, sizeof(*texture));
+            texture->opcode[0] = CELL_CMD_STATE_TEXTURE;
             texture->unit = i;
             if (cell->texture[i]) {
                uint level;
diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.c b/src/gallium/drivers/cell/ppu/cell_vbuf.c
index 65ba51b6bb2..ab54e796895 100644
--- a/src/gallium/drivers/cell/ppu/cell_vbuf.c
+++ b/src/gallium/drivers/cell/ppu/cell_vbuf.c
@@ -116,10 +116,11 @@ cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
 
    /* Tell SPUs they can release the vert buf */
    if (cvbr->vertex_buf != ~0U) {
+      STATIC_ASSERT(sizeof(struct cell_command_release_verts) % 16 == 0);
       struct cell_command_release_verts *release
          = (struct cell_command_release_verts *)
-         cell_batch_alloc(cell, sizeof(struct cell_command_release_verts));
-      release->opcode = CELL_CMD_RELEASE_VERTS;
+         cell_batch_alloc16(cell, sizeof(struct cell_command_release_verts));
+      release->opcode[0] = CELL_CMD_RELEASE_VERTS;
       release->vertex_buf = cvbr->vertex_buf;
    }
 
@@ -210,15 +211,16 @@ cell_vbuf_draw(struct vbuf_render *vbr,
 
    /* build/insert batch RENDER command */
    {
-      const uint index_bytes = ROUNDUP8(nr_indices * 2);
-      const uint vertex_bytes = nr_vertices * 4 * cell->vertex_info.size;
+      const uint index_bytes = ROUNDUP16(nr_indices * 2);
+      const uint vertex_bytes = ROUNDUP16(nr_vertices * 4 * cell->vertex_info.size);
+      STATIC_ASSERT(sizeof(struct cell_command_render) % 16 == 0);
       const uint batch_size = sizeof(struct cell_command_render) + index_bytes;
 
       struct cell_command_render *render
          = (struct cell_command_render *)
-         cell_batch_alloc(cell, batch_size);
+         cell_batch_alloc16(cell, batch_size);
 
-      render->opcode = CELL_CMD_RENDER;
+      render->opcode[0] = CELL_CMD_RENDER;
       render->prim_type = cvbr->prim;
 
       render->num_indexes = nr_indices;
@@ -236,7 +238,7 @@ cell_vbuf_draw(struct vbuf_render *vbr,
           min_index == 0 &&
           vertex_bytes + 16 <= cell_batch_free_space(cell)) {
          /* vertex data inlined, after indices, at 16-byte boundary */
-         void *dst = cell_batch_alloc_aligned(cell, vertex_bytes, 16);
+         void *dst = cell_batch_alloc16(cell, vertex_bytes);
          memcpy(dst, vertices, vertex_bytes);
          render->inline_verts = TRUE;
          render->vertex_buf = ~0;
diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c
index 8500d19754e..5c0179d9546 100644
--- a/src/gallium/drivers/cell/spu/spu_command.c
+++ b/src/gallium/drivers/cell/spu/spu_command.c
@@ -292,10 +292,10 @@ cmd_state_fragment_program(const struct cell_command_fragment_program *fp)
 
 
 static uint
-cmd_state_fs_constants(const uint64_t *buffer, uint pos)
+cmd_state_fs_constants(const qword *buffer, uint pos)
 {
-   const uint num_const = buffer[pos + 1];
-   const float *constants = (const float *) &buffer[pos + 2];
+   const uint num_const = spu_extract((vector unsigned int)buffer[pos+1], 0);
+   const float *constants = (const float *) &buffer[pos+2];
    uint i;
 
    D_PRINTF(CELL_DEBUG_CMD, "CMD_STATE_FS_CONSTANTS (%u)\n", num_const);
@@ -306,8 +306,8 @@ cmd_state_fs_constants(const uint64_t *buffer, uint pos)
       spu.constants[i] = spu_splats(constants[i]);
    }
 
-   /* return new buffer pos (in 8-byte words) */
-   return pos + 2 + num_const / 2;
+   /* return new buffer pos (in 16-byte words) */
+   return pos + 2 + (ROUNDUP16(num_const * sizeof(float)) / 16);
 }
 
 
@@ -547,8 +547,8 @@ cmd_batch(uint opcode)
 {
    const uint buf = (opcode >> 8) & 0xff;
    uint size = (opcode >> 16);
-   uint64_t buffer[CELL_BUFFER_SIZE / 8] ALIGN16_ATTRIB;
-   const unsigned usize = size / sizeof(buffer[0]);
+   qword buffer[CELL_BUFFER_SIZE / 16] ALIGN16_ATTRIB;
+   const unsigned usize = ROUNDUP16(size) / sizeof(buffer[0]);
    uint pos;
 
    D_PRINTF(CELL_DEBUG_CMD, "BATCH buffer %u, len %u, from %p\n",
@@ -578,7 +578,7 @@ cmd_batch(uint opcode)
     * Loop over commands in the batch buffer
     */
    for (pos = 0; pos < usize; /* no incr */) {
-      switch (buffer[pos]) {
+      switch (si_to_uint(buffer[pos])) {
       /*
        * rendering commands
        */
@@ -587,7 +587,7 @@ cmd_batch(uint opcode)
             struct cell_command_clear_surface *clr
                = (struct cell_command_clear_surface *) &buffer[pos];
             cmd_clear_surface(clr);
-            pos += sizeof(*clr) / 8;
+            pos += sizeof(*clr) / 16;
          }
          break;
       case CELL_CMD_RENDER:
@@ -596,7 +596,7 @@ cmd_batch(uint opcode)
                = (struct cell_command_render *) &buffer[pos];
             uint pos_incr;
             cmd_render(render, &pos_incr);
-            pos += pos_incr;
+            pos += ((pos_incr+1)&~1) / 2; // should 'fix' cmd_render return
          }
          break;
       /*
@@ -607,7 +607,7 @@ cmd_batch(uint opcode)
             struct cell_command_framebuffer *fb
                = (struct cell_command_framebuffer *) &buffer[pos];
             cmd_state_framebuffer(fb);
-            pos += sizeof(*fb) / 8;
+            pos += sizeof(*fb) / 16;
          }
          break;
       case CELL_CMD_STATE_FRAGMENT_OPS:
@@ -616,7 +616,7 @@ cmd_batch(uint opcode)
                = (struct cell_command_fragment_ops *) &buffer[pos];
             cmd_state_fragment_ops(fops);
             /* This is a variant-sized command */
-            pos += (sizeof(*fops) + fops->total_code_size)/ 8;
+            pos += ROUNDUP16(sizeof(*fops) + fops->total_code_size) / 16;
          }
          break;
       case CELL_CMD_STATE_FRAGMENT_PROGRAM:
@@ -624,7 +624,7 @@ cmd_batch(uint opcode)
             struct cell_command_fragment_program *fp
                = (struct cell_command_fragment_program *) &buffer[pos];
             cmd_state_fragment_program(fp);
-            pos += sizeof(*fp) / 8;
+            pos += sizeof(*fp) / 16;
          }
          break;
       case CELL_CMD_STATE_FS_CONSTANTS:
@@ -635,7 +635,7 @@ cmd_batch(uint opcode)
             struct cell_command_rasterizer *rast =
                (struct cell_command_rasterizer *) &buffer[pos];
             spu.rasterizer = rast->rasterizer;
-            pos += sizeof(*rast) / 8;
+            pos += sizeof(*rast) / 16;
          }
          break;
       case CELL_CMD_STATE_SAMPLER:
@@ -643,7 +643,7 @@ cmd_batch(uint opcode)
             struct cell_command_sampler *sampler
                = (struct cell_command_sampler *) &buffer[pos];
             cmd_state_sampler(sampler);
-            pos += sizeof(*sampler) / 8;
+            pos += sizeof(*sampler) / 16;
          }
          break;
       case CELL_CMD_STATE_TEXTURE:
@@ -651,37 +651,37 @@ cmd_batch(uint opcode)
             struct cell_command_texture *texture
                = (struct cell_command_texture *) &buffer[pos];
             cmd_state_texture(texture);
-            pos += sizeof(*texture) / 8;
+            pos += sizeof(*texture) / 16;
          }
          break;
       case CELL_CMD_STATE_VERTEX_INFO:
          cmd_state_vertex_info((struct vertex_info *) &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct vertex_info)) / 8);
+         pos += 1 + ROUNDUP16(sizeof(struct vertex_info)) / 16;
          break;
       case CELL_CMD_STATE_VIEWPORT:
          (void) memcpy(& draw.viewport, &buffer[pos+1],
                        sizeof(struct pipe_viewport_state));
-         pos += (1 + ROUNDUP8(sizeof(struct pipe_viewport_state)) / 8);
+         pos += 1 + ROUNDUP16(sizeof(struct pipe_viewport_state)) / 16;
          break;
       case CELL_CMD_STATE_UNIFORMS:
-         draw.constants = (const float (*)[4]) (uintptr_t) buffer[pos + 1];
+         draw.constants = (const float (*)[4]) (uintptr_t)spu_extract((vector unsigned int)buffer[pos+1],0);
          pos += 2;
          break;
       case CELL_CMD_STATE_VS_ARRAY_INFO:
          cmd_state_vs_array_info((struct cell_array_info *) &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_array_info)) / 8);
+         pos += 1 + ROUNDUP16(sizeof(struct cell_array_info)) / 16;
          break;
       case CELL_CMD_STATE_BIND_VS:
 #if 0
          spu_bind_vertex_shader(&draw,
                                 (struct cell_shader_info *) &buffer[pos+1]);
 #endif
-         pos += (1 + ROUNDUP8(sizeof(struct cell_shader_info)) / 8);
+         pos += 1 + ROUNDUP16(sizeof(struct cell_shader_info)) / 16;
          break;
       case CELL_CMD_STATE_ATTRIB_FETCH:
          cmd_state_attrib_fetch((struct cell_attribute_fetch_code *)
                                 &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_attribute_fetch_code)) / 8);
+         pos += 1 + ROUNDUP16(sizeof(struct cell_attribute_fetch_code)) / 16;
          break;
       /*
        * misc commands
@@ -695,7 +695,7 @@ cmd_batch(uint opcode)
             struct cell_command_fence *fence_cmd =
                (struct cell_command_fence *) &buffer[pos];
             cmd_fence(fence_cmd);
-            pos += sizeof(*fence_cmd) / 8;
+            pos += sizeof(*fence_cmd) / 16;
          }
          break;
       case CELL_CMD_RELEASE_VERTS:
@@ -703,7 +703,7 @@ cmd_batch(uint opcode)
             struct cell_command_release_verts *release
                = (struct cell_command_release_verts *) &buffer[pos];
             cmd_release_verts(release);
-            pos += sizeof(*release) / 8;
+            pos += sizeof(*release) / 16;
          }
          break;
       case CELL_CMD_FLUSH_BUFFER_RANGE: {
@@ -711,11 +711,11 @@ cmd_batch(uint opcode)
 	     &buffer[pos+1];
 
 	 spu_dcache_mark_dirty((unsigned) br->base, br->size);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_buffer_range)) / 8);
+         pos += 1 + ROUNDUP16(sizeof(struct cell_buffer_range)) / 16;
 	 break;
       }
       default:
-         printf("SPU %u: bad opcode: 0x%llx\n", spu.init.id, buffer[pos]);
+         printf("SPU %u: bad opcode: 0x%x\n", spu.init.id, si_to_uint(buffer[pos]));
          ASSERT(0);
          break;
       }
diff --git a/src/gallium/drivers/cell/spu/spu_shuffle.h b/src/gallium/drivers/cell/spu/spu_shuffle.h
index 7cbdb814d28..74f2a0b6d2e 100644
--- a/src/gallium/drivers/cell/spu/spu_shuffle.h
+++ b/src/gallium/drivers/cell/spu/spu_shuffle.h
@@ -171,7 +171,7 @@
    SHUFFLE_PATTERN_16_##M##__, \
    SHUFFLE_PATTERN_16_##N##__, \
    SHUFFLE_PATTERN_16_##O##__, \
-   SHUFFLE_PATTERN_16_##P
+   SHUFFLE_PATTERN_16_##P##__
 
 #define SHUFFLE16(A, B, C, D, E, F, G, H, I, J, K, L, M, N, O, P) \
    ((const vector unsigned char){ \
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 322be1252e9..0d9fcb99970 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -57,7 +57,7 @@ struct vertex_header {
 
 /* XXX fix this */
 #undef CEILF
-#define CEILF(X) ((float) (int) ((X) + 0.99999))
+#define CEILF(X) ((float) (int) ((X) + 0.99999f))
 
 
 #define QUAD_TOP_LEFT     0
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index 0958bba334a..061a4c064b6 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -67,11 +67,18 @@ struct nv50_rasterizer_stateobj {
 	struct nouveau_stateobj *so;
 };
 
+struct nv50_miptree_level {
+	struct pipe_buffer **image;
+	int *image_offset;
+	unsigned image_dirty_cpu[512/32];
+	unsigned image_dirty_gpu[512/32];
+};
+
 struct nv50_miptree {
 	struct pipe_texture base;
 	struct pipe_buffer *buffer;
 
-	int *image_offset;
+	struct nv50_miptree_level level[PIPE_MAX_TEXTURE_LEVELS];
 	int image_nr;
 	int total_size;
 };
@@ -84,7 +91,6 @@ nv50_miptree(struct pipe_texture *pt)
 
 struct nv50_surface {
 	struct pipe_surface base;
-	struct pipe_buffer *untiled;
 };
 
 static INLINE struct nv50_surface *
@@ -186,4 +192,8 @@ extern boolean nv50_state_validate(struct nv50_context *nv50);
 /* nv50_tex.c */
 extern void nv50_tex_validate(struct nv50_context *);
 
+/* nv50_miptree.c */
+extern void nv50_miptree_sync(struct pipe_screen *, struct nv50_miptree *,
+			      unsigned level, unsigned image);
+
 #endif
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 24973712324..63a23d06b89 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -27,14 +27,16 @@
 #include "nv50_context.h"
 
 static struct pipe_texture *
-nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
+nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 {
 	struct pipe_winsys *ws = pscreen->winsys;
 	struct nv50_miptree *mt = CALLOC_STRUCT(nv50_miptree);
-	unsigned usage, width = pt->width[0], height = pt->height[0];
-	int i;
+	struct pipe_texture *pt = &mt->base;
+	unsigned usage, width = tmp->width[0], height = tmp->height[0];
+	unsigned depth = tmp->depth[0];
+	int i, l;
 
-	mt->base = *pt;
+	mt->base = *tmp;
 	mt->base.refcount = 1;
 	mt->base.screen = pscreen;
 
@@ -59,17 +61,38 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 		mt->image_nr = 1;
 		break;
 	}
-	mt->image_offset = CALLOC(mt->image_nr, sizeof(int));
+
+	for (l = 0; l <= pt->last_level; l++) {
+		struct nv50_miptree_level *lvl = &mt->level[l];
+
+		pt->width[l] = width;
+		pt->height[l] = height;
+		pt->depth[l] = depth;
+		pt->nblocksx[l] = pf_get_nblocksx(&pt->block, width);
+		pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height);
+
+		lvl->image_offset = CALLOC(mt->image_nr, sizeof(int));
+		lvl->image = CALLOC(mt->image_nr, sizeof(struct pipe_buffer *));
+
+		width = MAX2(1, width >> 1);
+		height = MAX2(1, height >> 1);
+		depth = MAX2(1, depth >> 1);
+	}
 
 	for (i = 0; i < mt->image_nr; i++) {
-		int image_size;
+		for (l = 0; l <= pt->last_level; l++) {
+			struct nv50_miptree_level *lvl = &mt->level[l];
+			int size;
+
+			size  = align(pt->width[l], 8) * pt->block.size;
+			size  = align(size, 64);
+			size *= align(pt->height[l], 8) * pt->block.size;
 
-		image_size  = align(width, 8) * pt->block.size;
-		image_size  = align(image_size, 64);
-		image_size *= align(height, 8) * pt->block.size;
+			lvl->image[i] = ws->buffer_create(ws, 256, 0, size);
+			lvl->image_offset[i] = mt->total_size;
 
-		mt->image_offset[i] = mt->total_size;
-		mt->total_size += image_size;
+			mt->total_size += size;
+		}
 	}
 
 	mt->buffer = ws->buffer_create(ws, 256, usage, mt->total_size);
@@ -81,6 +104,24 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	return &mt->base;
 }
 
+static INLINE void
+mark_dirty(uint32_t *flags, unsigned image)
+{
+	flags[image / 32] |= (1 << (image % 32));
+}
+
+static INLINE void
+mark_clean(uint32_t *flags, unsigned image)
+{
+	flags[image / 32] &= ~(1 << (image % 32));
+}
+
+static INLINE int
+is_dirty(uint32_t *flags, unsigned image)
+{
+	return !!(flags[image / 32] & (1 << (image % 32)));
+}
+
 static void
 nv50_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 {
@@ -96,12 +137,86 @@ nv50_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 	}
 }
 
+void
+nv50_miptree_sync(struct pipe_screen *pscreen, struct nv50_miptree *mt,
+		  unsigned level, unsigned image)
+{
+	struct nouveau_winsys *nvws = nv50_screen(pscreen)->nvws;
+	struct nv50_miptree_level *lvl = &mt->level[level];
+	struct pipe_surface *dst, *src;
+	unsigned face = 0, zslice = 0;
+
+	if (!is_dirty(lvl->image_dirty_cpu, image))
+		return;
+
+	if (mt->base.target == PIPE_TEXTURE_CUBE)
+		face = image;
+	else
+	if (mt->base.target == PIPE_TEXTURE_3D)
+		zslice = image;
+
+	/* Mark as clean already - so we don't continually call this function
+	 * trying to get a GPU_WRITE pipe_surface!
+	 */
+	mark_clean(lvl->image_dirty_cpu, image);
+
+	/* Pretend we're doing CPU access so we get the backing pipe_surface
+	 * and not a view into the larger miptree.
+	 */
+	src = pscreen->get_tex_surface(pscreen, &mt->base, face, level, zslice,
+				       PIPE_BUFFER_USAGE_CPU_READ);
+
+	/* Pretend we're only reading with the GPU so surface doesn't get marked
+	 * as dirtied by the GPU.
+	 */
+	dst = pscreen->get_tex_surface(pscreen, &mt->base, face, level, zslice,
+				       PIPE_BUFFER_USAGE_GPU_READ);
+
+	nvws->surface_copy(nvws, dst, 0, 0, src, 0, 0, dst->width, dst->height);
+
+	pscreen->tex_surface_release(pscreen, &dst);
+	pscreen->tex_surface_release(pscreen, &src);
+}
+
+/* The reverse of the above */
+void
+nv50_miptree_sync_cpu(struct pipe_screen *pscreen, struct nv50_miptree *mt,
+		      unsigned level, unsigned image)
+{
+	struct nouveau_winsys *nvws = nv50_screen(pscreen)->nvws;
+	struct nv50_miptree_level *lvl = &mt->level[level];
+	struct pipe_surface *dst, *src;
+	unsigned face = 0, zslice = 0;
+
+	if (!is_dirty(lvl->image_dirty_gpu, image))
+		return;
+
+	if (mt->base.target == PIPE_TEXTURE_CUBE)
+		face = image;
+	else
+	if (mt->base.target == PIPE_TEXTURE_3D)
+		zslice = image;
+
+	mark_clean(lvl->image_dirty_gpu, image);
+
+	src = pscreen->get_tex_surface(pscreen, &mt->base, face, level, zslice,
+				       PIPE_BUFFER_USAGE_GPU_READ);
+	dst = pscreen->get_tex_surface(pscreen, &mt->base, face, level, zslice,
+				       PIPE_BUFFER_USAGE_CPU_READ);
+
+	nvws->surface_copy(nvws, dst, 0, 0, src, 0, 0, dst->width, dst->height);
+
+	pscreen->tex_surface_release(pscreen, &dst);
+	pscreen->tex_surface_release(pscreen, &src);
+}
+
 static struct pipe_surface *
 nv50_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 			 unsigned face, unsigned level, unsigned zslice,
 			 unsigned flags)
 {
 	struct nv50_miptree *mt = nv50_miptree(pt);
+	struct nv50_miptree_level *lvl = &mt->level[level];
 	struct nv50_surface *s;
 	struct pipe_surface *ps;
 	int img;
@@ -128,12 +243,29 @@ nv50_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 	ps->nblocksx = pt->nblocksx[level];
 	ps->nblocksy = pt->nblocksy[level];
 	ps->stride = ps->width * ps->block.size;
-	ps->offset = mt->image_offset[img];
 	ps->usage = flags;
 	ps->status = PIPE_SURFACE_STATUS_DEFINED;
 
-	pipe_texture_reference(&ps->texture, pt);
-	pipe_buffer_reference(pscreen, &ps->buffer, mt->buffer);
+	if (flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) {
+		assert(!(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE));
+		nv50_miptree_sync_cpu(pscreen, mt, level, img);
+
+		ps->offset = 0;
+		pipe_texture_reference(&ps->texture, pt);
+		pipe_buffer_reference(pscreen, &ps->buffer, lvl->image[img]);
+
+		if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
+			mark_dirty(lvl->image_dirty_cpu, img);
+	} else {
+		nv50_miptree_sync(pscreen, mt, level, img);
+
+		ps->offset = lvl->image_offset[img];
+		pipe_texture_reference(&ps->texture, pt);
+		pipe_buffer_reference(pscreen, &ps->buffer, mt->buffer);
+
+		if (flags & PIPE_BUFFER_USAGE_GPU_WRITE)
+			mark_dirty(lvl->image_dirty_gpu, img);
+	}
 
 	return ps;
 }
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index d66e1d0949d..7686f746eb2 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -32,7 +32,7 @@
 #include "nv50_context.h"
 
 #define NV50_SU_MAX_TEMP 64
-#define NV50_PROGRAM_DUMP
+//#define NV50_PROGRAM_DUMP
 
 /* ARL - gallium craps itself on progs/vp/arl.txt
  *
@@ -841,6 +841,28 @@ emit_neg(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 	emit(pc, e);
 }
 
+static void
+emit_kil(struct nv50_pc *pc, struct nv50_reg *src)
+{
+	struct nv50_program_exec *e;
+	const int r_pred = 1;
+
+	/* Sets predicate reg ? */
+	e = exec(pc);
+	e->inst[0] = 0xa00001fd;
+	e->inst[1] = 0xc4014788;
+	set_src_0(pc, src, e);
+	set_pred_wr(pc, 1, r_pred, e);
+	emit(pc, e);
+
+	/* This is probably KILP */
+	e = exec(pc);
+	e->inst[0] = 0x000001fe;
+	set_long(pc, e);
+	set_pred(pc, 1 /* LT? */, r_pred, e);
+	emit(pc, e);
+}
+
 static struct nv50_reg *
 tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst)
 {
@@ -1069,6 +1091,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		}
 		free_temp(pc, temp);
 		break;
+	case TGSI_OPCODE_KIL:
+		emit_kil(pc, src[0][0]);
+		emit_kil(pc, src[0][1]);
+		emit_kil(pc, src[0][2]);
+		emit_kil(pc, src[0][3]);
+		break;
 	case TGSI_OPCODE_LIT:
 		emit_lit(pc, &dst[0], mask, &src[0][0]);
 		break;
@@ -1602,13 +1630,19 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 	if (!upload)
 		return;
 
+#ifdef NV50_PROGRAM_DUMP
 	NOUVEAU_ERR("-------\n");
 	up = ptr = MALLOC(p->exec_size * 4);
 	for (e = p->exec_head; e; e = e->next) {
 		NOUVEAU_ERR("0x%08x\n", e->inst[0]);
 		if (is_long(e))
 			NOUVEAU_ERR("0x%08x\n", e->inst[1]);
+	}
 
+#endif
+
+	up = ptr = MALLOC(p->exec_size * 4);
+	for (e = p->exec_head; e; e = e->next) {
 		*(ptr++) = e->inst[0];
 		if (is_long(e))
 			*(ptr++) = e->inst[1];
@@ -1705,7 +1739,7 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
 		  NOUVEAU_BO_LOW, 0, 0);
 	so_method(so, tesla, 0x1904, 4);
-	so_data  (so, 0x01040404); /* p: 0x01000404 */
+	so_data  (so, 0x00040404); /* p: 0x01000404 */
 	so_data  (so, 0x00000004);
 	so_data  (so, 0x00000000);
 	so_data  (so, 0x00000000);
diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c
index 777e77906d5..b923c820eba 100644
--- a/src/gallium/drivers/nv50/nv50_query.c
+++ b/src/gallium/drivers/nv50/nv50_query.c
@@ -21,41 +21,101 @@
  */
 
 #include "pipe/p_context.h"
+#include "pipe/p_inlines.h"
 
 #include "nv50_context.h"
 
+struct nv50_query {
+	struct pipe_buffer *buffer;
+	unsigned type;
+	boolean ready;
+	uint64_t result;
+};
+
+static INLINE struct nv50_query *
+nv50_query(struct pipe_query *pipe)
+{
+	return (struct nv50_query *)pipe;
+}
+
 static struct pipe_query *
 nv50_query_create(struct pipe_context *pipe, unsigned type)
 {
-	NOUVEAU_ERR("unimplemented\n");
-	return NULL;
+	struct pipe_winsys *ws = pipe->winsys;
+	struct nv50_query *q = CALLOC_STRUCT(nv50_query);
+
+	assert (q->type == PIPE_QUERY_OCCLUSION_COUNTER);
+	q->type = type;
+
+	q->buffer = ws->buffer_create(ws, 256, 0, 16);
+	if (!q->buffer) {
+		FREE(q);
+		return NULL;
+	}
+
+	return (struct pipe_query *)q;
 }
 
 static void
-nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *q)
+nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
 {
-	NOUVEAU_ERR("unimplemented\n");
+	struct nv50_query *q = nv50_query(pq);
+
+	if (q) {
+		pipe_buffer_reference(pipe, &q->buffer, NULL);
+		FREE(q);
+	}
 }
 
 static void
-nv50_query_begin(struct pipe_context *pipe, struct pipe_query *q)
+nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 {
-	NOUVEAU_ERR("unimplemented\n");
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nv50_query *q = nv50_query(pq);
+
+	BEGIN_RING(tesla, 0x1530, 1);
+	OUT_RING  (1);
+	BEGIN_RING(tesla, 0x1514, 1);
+	OUT_RING  (1);
+
+	q->ready = FALSE;
 }
 
 static void
-nv50_query_end(struct pipe_context *pipe, struct pipe_query *q)
+nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 {
-	NOUVEAU_ERR("unimplemented\n");
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nv50_query *q = nv50_query(pq);
+
+	BEGIN_RING(tesla, 0x1b00, 4);
+	OUT_RELOCh(q->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCl(q->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RING  (0x00000000);
+	OUT_RING  (0x0100f002);
+	FIRE_RING (NULL);
 }
 
 static boolean
-nv50_query_result(struct pipe_context *pipe, struct pipe_query *q,
+nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 		  boolean wait, uint64_t *result)
 {
-	NOUVEAU_ERR("unimplemented\n");
-	*result = 0xdeadcafe;
-	return TRUE;
+	struct pipe_winsys *ws = pipe->winsys;
+	struct nv50_query *q = nv50_query(pq);
+
+	/*XXX: Want to be able to return FALSE here instead of blocking
+	 *     until the result is available..
+	 */
+
+	if (!q->ready) {
+		uint32_t *map = ws->buffer_map(ws, q->buffer,
+					       PIPE_BUFFER_USAGE_CPU_READ);
+		q->result = map[1];
+		q->ready = TRUE;
+		ws->buffer_unmap(ws, q->buffer);
+	}
+
+	*result = q->result;
+	return q->ready;
 }
 
 void
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 52f6a406882..ef46233f839 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -57,6 +57,10 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen,
 		case PIPE_FORMAT_A8_UNORM:
 		case PIPE_FORMAT_I8_UNORM:
 		case PIPE_FORMAT_A8L8_UNORM:
+		case PIPE_FORMAT_DXT1_RGB:
+		case PIPE_FORMAT_DXT1_RGBA:
+		case PIPE_FORMAT_DXT3_RGBA:
+		case PIPE_FORMAT_DXT5_RGBA:
 			return TRUE;
 		default:
 			break;
@@ -90,23 +94,23 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 	case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
 		return 32;
 	case PIPE_CAP_NPOT_TEXTURES:
-		return 0;
+		return 1;
 	case PIPE_CAP_TWO_SIDED_STENCIL:
 		return 1;
 	case PIPE_CAP_GLSL:
 		return 0;
 	case PIPE_CAP_S3TC:
-		return 0;
+		return 1;
 	case PIPE_CAP_ANISOTROPIC_FILTER:
-		return 0;
+		return 1;
 	case PIPE_CAP_POINT_SPRITE:
 		return 0;
 	case PIPE_CAP_MAX_RENDER_TARGETS:
 		return 8;
 	case PIPE_CAP_OCCLUSION_QUERY:
-		return 0;
+		return 1;
 	case PIPE_CAP_TEXTURE_SHADOW_MAP:
-		return 0;
+		return 1;
 	case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
 		return 13;
 	case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 95f9d408b5e..38c1d938b8e 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -175,6 +175,32 @@ nv50_sampler_state_create(struct pipe_context *pipe,
 		break;
 	}
 
+	if (cso->max_anisotropy >= 16.0)
+		tsc[0] |= (7 << 20);
+	else
+	if (cso->max_anisotropy >= 12.0)
+		tsc[0] |= (6 << 20);
+	else
+	if (cso->max_anisotropy >= 10.0)
+		tsc[0] |= (5 << 20);
+	else
+	if (cso->max_anisotropy >= 8.0)
+		tsc[0] |= (4 << 20);
+	else
+	if (cso->max_anisotropy >= 6.0)
+		tsc[0] |= (3 << 20);
+	else
+	if (cso->max_anisotropy >= 4.0)
+		tsc[0] |= (2 << 20);
+	else
+	if (cso->max_anisotropy >= 2.0)
+		tsc[0] |= (1 << 20);
+
+	if (cso->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+		tsc[0] |= (1 << 8);
+		tsc[0] |= (nvgl_comparison_op(cso->compare_func) & 0x7);
+	}
+
 	return (void *)tsc;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index 5bf97d3a6bf..3f45a2fe186 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -63,48 +63,17 @@ static void *
 nv50_surface_map(struct pipe_screen *screen, struct pipe_surface *ps,
 		 unsigned flags )
 {
-	struct nouveau_winsys *nvws = nv50_screen(screen)->nvws;
 	struct pipe_winsys *ws = screen->winsys;
-	struct nv50_surface *s = nv50_surface(ps);
-	struct nv50_surface m = *s;
-	void *map;
 
-	if (!s->untiled) {
-		s->untiled = ws->buffer_create(ws, 0, 0, ps->buffer->size);
-
-		m.base.buffer = s->untiled;
-		nvws->surface_copy(nvws, &m.base, 0, 0, &s->base, 0, 0,
-					 ps->width, ps->height);
-	}
-
-	/* Map original tiled surface to disallow it being validated while
-	 * untiled mirror is mapped.
-	 */
-	ws->buffer_map(ws, ps->buffer, flags);
-
-	map = ws->buffer_map(ws, s->untiled, flags);
-	if (!map)
-		return NULL;
-
-	return map;
+	return ws->buffer_map(ws, ps->buffer, flags);
 }
 
 static void
 nv50_surface_unmap(struct pipe_screen *pscreen, struct pipe_surface *ps)
 {
-	struct nouveau_winsys *nvws = nv50_screen(pscreen)->nvws;
 	struct pipe_winsys *ws = pscreen->winsys;
-	struct nv50_surface *s = nv50_surface(ps);
-	struct nv50_surface m = *s;
 
-	ws->buffer_unmap(ws, s->untiled);
 	ws->buffer_unmap(ws, ps->buffer);
-
-	m.base.buffer = s->untiled;
-	nvws->surface_copy(nvws, &s->base, 0, 0, &m.base, 0, 0,
-				 ps->width, ps->height);
-
-	pipe_buffer_reference(pscreen, &s->untiled, NULL);
 }
 
 void
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index fde3c97c059..239407c92bb 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -85,6 +85,34 @@ nv50_tex_construct(struct nouveau_stateobj *so, struct nv50_miptree *mt)
 			    NV50TIC_0_0_MAPB_C0 | NV50TIC_0_0_TYPEB_UNORM |
 			    NV50TIC_0_0_FMT_8_8);
 		break;
+	case PIPE_FORMAT_DXT1_RGB:
+		so_data(so, NV50TIC_0_0_MAPA_ONE | NV50TIC_0_0_TYPEA_UNORM |
+			    NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM |
+			    NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM |
+			    NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM |
+			    NV50TIC_0_0_FMT_DXT1);
+		break;
+	case PIPE_FORMAT_DXT1_RGBA:
+		so_data(so, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM |
+			    NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM |
+			    NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM |
+			    NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM |
+			    NV50TIC_0_0_FMT_DXT1);
+		break;
+	case PIPE_FORMAT_DXT3_RGBA:
+		so_data(so, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM |
+			    NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM |
+			    NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM |
+			    NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM |
+			    NV50TIC_0_0_FMT_DXT3);
+		break;
+	case PIPE_FORMAT_DXT5_RGBA:
+		so_data(so, NV50TIC_0_0_MAPA_C3 | NV50TIC_0_0_TYPEA_UNORM |
+			    NV50TIC_0_0_MAPR_C0 | NV50TIC_0_0_TYPER_UNORM |
+			    NV50TIC_0_0_MAPG_C1 | NV50TIC_0_0_TYPEG_UNORM |
+			    NV50TIC_0_0_MAPB_C2 | NV50TIC_0_0_TYPEB_UNORM |
+			    NV50TIC_0_0_FMT_DXT5);
+		break;
 	default:
 		return 1;
 	}
@@ -105,14 +133,23 @@ nv50_tex_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nouveau_stateobj *so;
-	int i;
+	int unit, level, image;
 
 	so = so_new(nv50->miptree_nr * 8 + 3, nv50->miptree_nr * 2);
 	so_method(so, tesla, 0x0f00, 1);
 	so_data  (so, NV50_CB_TIC);
 	so_method(so, tesla, 0x40000f04, nv50->miptree_nr * 8);
-	for (i = 0; i < nv50->miptree_nr; i++) {
-		if (nv50_tex_construct(so, nv50->miptree[i])) {
+	for (unit = 0; unit < nv50->miptree_nr; unit++) {
+		struct nv50_miptree *mt = nv50->miptree[unit];
+
+		for (level = 0; level <= mt->base.last_level; level++) {
+			for (image = 0; image < mt->image_nr; image++) {
+				nv50_miptree_sync(&nv50->screen->pipe, mt,
+						  level, image);
+			}
+		}
+
+		if (nv50_tex_construct(so, mt)) {
 			NOUVEAU_ERR("failed tex validate\n");
 			so_ref(NULL, &so);
 			return;
diff --git a/src/gallium/drivers/nv50/nv50_texture.h b/src/gallium/drivers/nv50/nv50_texture.h
index 6861d67b4d5..aca622c73b1 100644
--- a/src/gallium/drivers/nv50/nv50_texture.h
+++ b/src/gallium/drivers/nv50/nv50_texture.h
@@ -50,6 +50,9 @@
 #define NV50TIC_0_0_FMT_5_6_5                                     0x00000015
 #define NV50TIC_0_0_FMT_8_8                                       0x00000018
 #define NV50TIC_0_0_FMT_8                                         0x0000001d
+#define NV50TIC_0_0_FMT_DXT1                                      0x00000024
+#define NV50TIC_0_0_FMT_DXT3                                      0x00000025
+#define NV50TIC_0_0_FMT_DXT5                                      0x00000026
 
 #define NV50TIC_0_1_OFFSET_LOW_MASK                               0xffffffff
 #define NV50TIC_0_1_OFFSET_LOW_SHIFT                                       0
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index 584336682e4..435dc9777da 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -60,6 +60,10 @@ nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 	OUT_RING  (0);
 	BEGIN_RING(tesla, 0x142c, 1);
 	OUT_RING  (0);
+	BEGIN_RING(tesla, 0x1440, 1);
+	OUT_RING  (0);
+	BEGIN_RING(tesla, 0x1334, 1);
+	OUT_RING  (0);
 
 	BEGIN_RING(tesla, NV50TCL_VERTEX_BEGIN, 1);
 	OUT_RING  (nv50_prim(mode));
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 5c6a92b53bc..4f0b301f317 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -204,6 +204,7 @@ enum pipe_texture_target {
 #define PIPE_BUFFER_USAGE_VERTEX    (1 << 5)
 #define PIPE_BUFFER_USAGE_INDEX     (1 << 6)
 #define PIPE_BUFFER_USAGE_CONSTANT  (1 << 7)
+#define PIPE_BUFFER_USAGE_DISCARD   (1 << 8)
 /** Pipe driver custom usage flags should be greater or equal to this value */
 #define PIPE_BUFFER_USAGE_CUSTOM    (1 << 16)
 
diff --git a/src/gallium/include/state_tracker/drm_api.h b/src/gallium/include/state_tracker/drm_api.h
new file mode 100644
index 00000000000..54480fa0477
--- /dev/null
+++ b/src/gallium/include/state_tracker/drm_api.h
@@ -0,0 +1,33 @@
+
+#ifndef _DRM_API_H_
+#define _DRM_API_H_
+
+struct pipe_screen;
+struct pipe_winsys;
+struct pipe_context;
+
+struct drm_api
+{
+	/**
+	 * Special buffer function
+	 */
+	/*@{*/
+	struct pipe_screen*  (*create_screen)(int drmFB, int pciID);
+	struct pipe_context* (*create_context)(struct pipe_screen *screen);
+	/*@}*/
+
+	/**
+	 * Special buffer function
+	 */
+	/*@{*/
+	struct pipe_buffer* (*buffer_from_handle)(struct pipe_winsys *winsys, const char *name, unsigned handle);
+	unsigned (*handle_from_buffer)(struct pipe_winsys *winsys, struct pipe_buffer *buffer);
+	/*@}*/
+};
+
+/**
+ * A driver needs to export this symbol
+ */
+extern struct drm_api drm_api_hocks;
+
+#endif
diff --git a/src/gallium/state_trackers/egl/Makefile b/src/gallium/state_trackers/egl/Makefile
new file mode 100644
index 00000000000..17d1a7a8e40
--- /dev/null
+++ b/src/gallium/state_trackers/egl/Makefile
@@ -0,0 +1,29 @@
+TARGET     = libegldrm.a
+CFILES     = $(wildcard ./*.c)
+OBJECTS    = $(patsubst ./%.c,./%.o,$(CFILES))
+GALLIUMDIR = ../..
+TOP        = ../../../..
+
+include ${TOP}/configs/current
+
+CFLAGS += -g -Wall -Werror=implicit-function-declaration -fPIC \
+          $(shell pkg-config --cflags pixman-1 xorg-server) \
+          -I${GALLIUMDIR}/include \
+          -I${GALLIUMDIR}/auxiliary \
+          -I${TOP}/src/mesa/drivers/dri/common \
+          -I${TOP}/src/mesa \
+          -I$(TOP)/include \
+          -I$(TOP)/src/egl/main \
+          ${LIBDRM_CFLAGS}
+
+#############################################
+
+.PHONY	= all clean
+
+all: ${TARGET}
+
+${TARGET}: ${OBJECTS}
+	ar rcs $@ $^
+
+clean:
+	rm -rf ${OBJECTS} ${TARGET}
diff --git a/src/gallium/state_trackers/egl/egl_context.c b/src/gallium/state_trackers/egl/egl_context.c
new file mode 100644
index 00000000000..217fe00338e
--- /dev/null
+++ b/src/gallium/state_trackers/egl/egl_context.c
@@ -0,0 +1,194 @@
+
+#include "utils.h"
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include "egl_tracker.h"
+
+#include "egllog.h"
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_winsys.h"
+
+#include "state_tracker/st_public.h"
+#include "state_tracker/drm_api.h"
+
+#include "GL/internal/glcore.h"
+
+#define need_GL_ARB_multisample
+#define need_GL_ARB_point_parameters
+#define need_GL_ARB_texture_compression
+#define need_GL_ARB_vertex_buffer_object
+#define need_GL_ARB_vertex_program
+#define need_GL_ARB_window_pos
+#define need_GL_EXT_blend_color
+#define need_GL_EXT_blend_equation_separate
+#define need_GL_EXT_blend_func_separate
+#define need_GL_EXT_blend_minmax
+#define need_GL_EXT_cull_vertex
+#define need_GL_EXT_fog_coord
+#define need_GL_EXT_framebuffer_object
+#define need_GL_EXT_multi_draw_arrays
+#define need_GL_EXT_secondary_color
+#define need_GL_NV_vertex_program
+#include "extension_helper.h"
+
+/**
+ * TODO HACK! FUGLY!
+ * Copied for intel extentions.
+ */
+const struct dri_extension card_extensions[] = {
+	{"GL_ARB_multisample", GL_ARB_multisample_functions},
+	{"GL_ARB_multitexture", NULL},
+	{"GL_ARB_point_parameters", GL_ARB_point_parameters_functions},
+	{"GL_ARB_texture_border_clamp", NULL},
+	{"GL_ARB_texture_compression", GL_ARB_texture_compression_functions},
+	{"GL_ARB_texture_cube_map", NULL},
+	{"GL_ARB_texture_env_add", NULL},
+	{"GL_ARB_texture_env_combine", NULL},
+	{"GL_ARB_texture_env_dot3", NULL},
+	{"GL_ARB_texture_mirrored_repeat", NULL},
+	{"GL_ARB_texture_rectangle", NULL},
+	{"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions},
+	{"GL_ARB_pixel_buffer_object", NULL},
+	{"GL_ARB_vertex_program", GL_ARB_vertex_program_functions},
+	{"GL_ARB_window_pos", GL_ARB_window_pos_functions},
+	{"GL_EXT_blend_color", GL_EXT_blend_color_functions},
+	{"GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions},
+	{"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions},
+	{"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions},
+	{"GL_EXT_blend_subtract", NULL},
+	{"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions},
+	{"GL_EXT_fog_coord", GL_EXT_fog_coord_functions},
+	{"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions},
+	{"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions},
+	{"GL_EXT_packed_depth_stencil", NULL},
+	{"GL_EXT_pixel_buffer_object", NULL},
+	{"GL_EXT_secondary_color", GL_EXT_secondary_color_functions},
+	{"GL_EXT_stencil_wrap", NULL},
+	{"GL_EXT_texture_edge_clamp", NULL},
+	{"GL_EXT_texture_env_combine", NULL},
+	{"GL_EXT_texture_env_dot3", NULL},
+	{"GL_EXT_texture_filter_anisotropic", NULL},
+	{"GL_EXT_texture_lod_bias", NULL},
+	{"GL_3DFX_texture_compression_FXT1", NULL},
+	{"GL_APPLE_client_storage", NULL},
+	{"GL_MESA_pack_invert", NULL},
+	{"GL_MESA_ycbcr_texture", NULL},
+	{"GL_NV_blend_square", NULL},
+	{"GL_NV_vertex_program", GL_NV_vertex_program_functions},
+	{"GL_NV_vertex_program1_1", NULL},
+	{"GL_SGIS_generate_mipmap", NULL },
+	{NULL, NULL}
+};
+
+EGLContext
+drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext share_list, const EGLint *attrib_list)
+{
+	struct drm_device *dev = (struct drm_device *)drv;
+	struct drm_context *ctx;
+	struct drm_context *share = NULL;
+	struct st_context *st_share = NULL;
+	_EGLConfig *conf;
+	int i;
+	__GLcontextModes *visual;
+
+	conf = _eglLookupConfig(drv, dpy, config);
+	if (!conf) {
+		_eglError(EGL_BAD_CONFIG, "eglCreateContext");
+		return EGL_NO_CONTEXT;
+	}
+
+	for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) {
+		switch (attrib_list[i]) {
+			/* no attribs defined for now */
+			default:
+				_eglError(EGL_BAD_ATTRIBUTE, "eglCreateContext");
+				return EGL_NO_CONTEXT;
+		}
+	}
+
+	ctx = (struct drm_context *) calloc(1, sizeof(struct drm_context));
+	if (!ctx)
+		goto err_c;
+
+	_eglInitContext(drv, dpy, &ctx->base, config, attrib_list);
+
+	ctx->pipe = drm_api_hocks.create_context(dev->screen);
+	if (!ctx->pipe)
+		goto err_pipe;
+
+	if (share)
+		st_share = share->st;
+
+	visual = drm_visual_from_config(conf);
+	ctx->st = st_create_context(ctx->pipe, visual, st_share);
+	drm_visual_modes_destroy(visual);
+
+	if (!ctx->st)
+		goto err_gl;
+
+	/* generate handle and insert into hash table */
+	_eglSaveContext(&ctx->base);
+	assert(_eglGetContextHandle(&ctx->base));
+
+	return _eglGetContextHandle(&ctx->base);
+
+err_gl:
+	ctx->pipe->destroy(ctx->pipe);
+err_pipe:
+	free(ctx);
+err_c:
+	return EGL_NO_CONTEXT;
+}
+
+EGLBoolean
+drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context)
+{
+	struct drm_context *c = lookup_drm_context(context);
+	_eglRemoveContext(&c->base);
+	if (c->base.IsBound) {
+		c->base.DeletePending = EGL_TRUE;
+	} else {
+		st_destroy_context(c->st);
+		c->pipe->destroy(c->pipe);
+		free(c);
+	}
+	return EGL_TRUE;
+}
+
+EGLBoolean
+drm_make_current(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext context)
+{
+	struct drm_surface *readSurf = lookup_drm_surface(read);
+	struct drm_surface *drawSurf = lookup_drm_surface(draw);
+	struct drm_context *ctx = lookup_drm_context(context);
+	EGLBoolean b;
+
+	b = _eglMakeCurrent(drv, dpy, draw, read, context);
+	if (!b)
+		return EGL_FALSE;
+
+	if (ctx) {
+		if (!drawSurf || !readSurf)
+			return EGL_FALSE;
+
+		drawSurf->user = ctx;
+		readSurf->user = ctx;
+
+		st_make_current(ctx->st, drawSurf->stfb, readSurf->stfb);
+
+		/* st_resize_framebuffer needs a bound context to work */
+		st_resize_framebuffer(drawSurf->stfb, drawSurf->w, drawSurf->h);
+		st_resize_framebuffer(readSurf->stfb, readSurf->w, readSurf->h);
+	} else {
+		drawSurf->user = NULL;
+		readSurf->user = NULL;
+
+		st_make_current(NULL, NULL, NULL);
+	}
+
+	return EGL_TRUE;
+}
diff --git a/src/gallium/state_trackers/egl/egl_surface.c b/src/gallium/state_trackers/egl/egl_surface.c
new file mode 100644
index 00000000000..71292bf2a91
--- /dev/null
+++ b/src/gallium/state_trackers/egl/egl_surface.c
@@ -0,0 +1,418 @@
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include "egl_tracker.h"
+
+#include "egllog.h"
+
+#include "pipe/p_inlines.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_context.h"
+
+#include "state_tracker/drm_api.h"
+
+/*
+ * Util functions
+ */
+
+static struct drm_mode_modeinfo *
+drm_find_mode(drmModeConnectorPtr connector, _EGLMode *mode)
+{
+	int i;
+	struct drm_mode_modeinfo *m = NULL;
+
+	for (i = 0; i < connector->count_modes; i++) {
+		m = &connector->modes[i];
+		if (m->hdisplay == mode->Width && m->vdisplay == mode->Height && m->vrefresh == mode->RefreshRate)
+			break;
+		m = &connector->modes[0]; /* if we can't find one, return first */
+	}
+
+	return m;
+}
+
+static struct st_framebuffer *
+drm_create_framebuffer(const __GLcontextModes *visual,
+                       unsigned width,
+                       unsigned height,
+                       void *priv)
+{
+	enum pipe_format colorFormat, depthFormat, stencilFormat;
+
+	if (visual->redBits == 5)
+		colorFormat = PIPE_FORMAT_R5G6B5_UNORM;
+	else
+		colorFormat = PIPE_FORMAT_A8R8G8B8_UNORM;
+
+	if (visual->depthBits == 16)
+		depthFormat = PIPE_FORMAT_Z16_UNORM;
+	else if (visual->depthBits == 24)
+		depthFormat = PIPE_FORMAT_S8Z24_UNORM;
+	else
+		depthFormat = PIPE_FORMAT_NONE;
+
+	if (visual->stencilBits == 8)
+		stencilFormat = PIPE_FORMAT_S8Z24_UNORM;
+	else
+		stencilFormat = PIPE_FORMAT_NONE;
+
+	return st_create_framebuffer(visual,
+	                             colorFormat,
+	                             depthFormat,
+	                             stencilFormat,
+	                             width,
+	                             height,
+	                             priv);
+}
+
+static void
+drm_create_texture(_EGLDriver *drv,
+                   struct drm_screen *scrn,
+                   unsigned w, unsigned h)
+{
+	struct drm_device *dev = (struct drm_device *)drv;
+	struct pipe_screen *screen = dev->screen;
+	struct pipe_surface *surface;
+	struct pipe_texture *texture;
+	struct pipe_texture templat;
+	struct pipe_buffer *buf;
+	unsigned stride = 1024;
+	unsigned pitch = 0;
+	unsigned size = 0;
+	void *ptr;
+
+	/* ugly */
+	if (stride < w)
+		stride = 2048;
+
+	pitch = stride * 4;
+	size = h * 2 * pitch;
+
+	buf = pipe_buffer_create(screen,
+	                         0, /* alignment */
+	                         PIPE_BUFFER_USAGE_GPU_READ_WRITE |
+	                         PIPE_BUFFER_USAGE_CPU_READ_WRITE,
+	                         size);
+
+	if (!buf)
+		goto err_buf;
+
+#if DEBUG
+	ptr = pipe_buffer_map(screen, buf, PIPE_BUFFER_USAGE_CPU_WRITE);
+	memset(ptr, 0xFF, size);
+	pipe_buffer_unmap(screen, buf);
+#else
+	(void)ptr;
+#endif
+
+	memset(&templat, 0, sizeof(templat));
+	templat.tex_usage |= PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+	templat.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET;
+	templat.target = PIPE_TEXTURE_2D;
+	templat.last_level = 0;
+	templat.depth[0] = 1;
+	templat.format = PIPE_FORMAT_A8R8G8B8_UNORM;
+	templat.width[0] = w;
+	templat.height[0] = h;
+	pf_get_block(templat.format, &templat.block);
+
+	texture = screen->texture_blanket(dev->screen,
+	                                  &templat,
+	                                  &pitch,
+	                                  buf);
+	if (!texture)
+		goto err_tex;
+
+	surface = screen->get_tex_surface(screen,
+	                                  texture,
+	                                  0,
+	                                  0,
+	                                  0,
+	                                  PIPE_BUFFER_USAGE_GPU_WRITE);
+
+	if (!surface)
+		goto err_surf;
+
+
+	scrn->tex = texture;
+	scrn->surface = surface;
+	scrn->buffer = buf;
+	scrn->front.width = w;
+	scrn->front.height = h;
+	scrn->front.pitch = pitch;
+	scrn->front.handle = drm_api_hocks.handle_from_buffer(dev->winsys, scrn->buffer);
+	if (0)
+		goto err_handle;
+
+	return;
+
+err_handle:
+	pipe_surface_reference(&surface, NULL);
+err_surf:
+	pipe_texture_reference(&texture, NULL);
+err_tex:
+	pipe_buffer_reference(screen, &buf, NULL);
+err_buf:
+	return;
+}
+
+/*
+ * Exported functions
+ */
+
+void
+drm_takedown_shown_screen(_EGLDriver *drv, struct drm_screen *screen)
+{
+	struct drm_device *dev = (struct drm_device *)drv;
+
+	screen->surf = NULL;
+
+	drmModeSetCrtc(
+		dev->drmFD,
+		screen->crtcID,
+		0, // FD
+		0, 0,
+		NULL, 0, // List of output ids
+		NULL);
+
+	drmModeRmFB(dev->drmFD, screen->fbID);
+	drmModeFreeFB(screen->fb);
+	screen->fb = NULL;
+
+	pipe_surface_reference(&screen->surface, NULL);
+	pipe_texture_reference(&screen->tex, NULL);
+	pipe_buffer_reference(dev->screen, &screen->buffer, NULL);
+
+	screen->shown = 0;
+}
+
+EGLSurface
+drm_create_window_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativeWindowType window, const EGLint *attrib_list)
+{
+	return EGL_NO_SURFACE;
+}
+
+
+EGLSurface
+drm_create_pixmap_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativePixmapType pixmap, const EGLint *attrib_list)
+{
+	return EGL_NO_SURFACE;
+}
+
+
+EGLSurface
+drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
+                           const EGLint *attrib_list)
+{
+	int i;
+	int width = -1;
+	int height = -1;
+	struct drm_surface *surf = NULL;
+	__GLcontextModes *visual;
+	_EGLConfig *conf;
+
+	conf = _eglLookupConfig(drv, dpy, config);
+	if (!conf) {
+		_eglError(EGL_BAD_CONFIG, "eglCreatePbufferSurface");
+		return EGL_NO_CONTEXT;
+	}
+
+	for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) {
+		switch (attrib_list[i]) {
+			case EGL_WIDTH:
+				width = attrib_list[++i];
+				break;
+			case EGL_HEIGHT:
+				height = attrib_list[++i];
+				break;
+			default:
+				_eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface");
+				return EGL_NO_SURFACE;
+		}
+	}
+
+	if (width < 1 || height < 1) {
+		_eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface");
+		return EGL_NO_SURFACE;
+	}
+
+	surf = (struct drm_surface *) calloc(1, sizeof(struct drm_surface));
+	if (!surf)
+		goto err;
+
+	if (!_eglInitSurface(drv, dpy, &surf->base, EGL_PBUFFER_BIT, config, attrib_list))
+		goto err_surf;
+
+	surf->w = width;
+	surf->h = height;
+
+	visual = drm_visual_from_config(conf);
+	surf->stfb = drm_create_framebuffer(visual,
+	                                    width,
+	                                    height,
+	                                    (void*)surf);
+	drm_visual_modes_destroy(visual);
+
+	_eglSaveSurface(&surf->base);
+	return surf->base.Handle;
+
+err_surf:
+	free(surf);
+err:
+	return EGL_NO_SURFACE;
+}
+
+EGLSurface
+drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg,
+                               const EGLint *attrib_list)
+{
+	EGLSurface surf = drm_create_pbuffer_surface(drv, dpy, cfg, attrib_list);
+
+	return surf;
+}
+
+EGLBoolean
+drm_show_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy,
+                             EGLScreenMESA screen,
+                             EGLSurface surface, EGLModeMESA m)
+{
+	struct drm_device *dev = (struct drm_device *)drv;
+	struct drm_surface *surf = lookup_drm_surface(surface);
+	struct drm_screen *scrn = lookup_drm_screen(dpy, screen);
+	_EGLMode *mode = _eglLookupMode(dpy, m);
+	int ret;
+	unsigned int i, k;
+
+	if (scrn->shown)
+		drm_takedown_shown_screen(drv, scrn);
+
+
+	drm_create_texture(drv, scrn, mode->Width, mode->Height);
+	if (!scrn->buffer)
+		return EGL_FALSE;
+
+	ret = drmModeAddFB(dev->drmFD,
+	                   scrn->front.width, scrn->front.height,
+	                   32, 32, scrn->front.pitch,
+	                   scrn->front.handle,
+	                   &scrn->fbID);
+
+	if (ret)
+		goto err_bo;
+
+	scrn->fb = drmModeGetFB(dev->drmFD, scrn->fbID);
+	if (!scrn->fb)
+		goto err_bo;
+
+	/* find a fitting crtc */
+	{
+		drmModeConnector *con = scrn->connector;
+
+		scrn->mode = drm_find_mode(con, mode);
+		if (!scrn->mode)
+			goto err_fb;
+
+		for (k = 0; k < con->count_encoders; k++) {
+			drmModeEncoder *enc = drmModeGetEncoder(dev->drmFD, con->encoders[k]);
+			for (i = 0; i < dev->res->count_crtcs; i++) {
+				if (enc->possible_crtcs & (1<<i)) {
+					/* save the ID */
+					scrn->crtcID = dev->res->crtcs[i];
+
+					/* skip the rest */
+					i = dev->res->count_crtcs;
+					k = dev->res->count_encoders;
+				}
+			}
+			drmModeFreeEncoder(enc);
+		}
+	}
+
+	ret = drmModeSetCrtc(dev->drmFD,
+	                     scrn->crtcID,
+	                     scrn->fbID,
+	                     0, 0,
+	                     &scrn->connectorID, 1,
+	                     scrn->mode);
+
+	if (ret)
+		goto err_crtc;
+
+	surf->screen = scrn;
+
+	scrn->surf = surf;
+	scrn->shown = 1;
+
+	return EGL_TRUE;
+
+err_crtc:
+	scrn->crtcID = 0;
+
+err_fb:
+	drmModeRmFB(dev->drmFD, scrn->fbID);
+	drmModeFreeFB(scrn->fb);
+	scrn->fb = NULL;
+
+err_bo:
+	pipe_surface_reference(&scrn->surface, NULL);
+	pipe_texture_reference(&scrn->tex, NULL);
+	pipe_buffer_reference(dev->screen, &scrn->buffer, NULL);
+
+	return EGL_FALSE;
+}
+
+EGLBoolean
+drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface)
+{
+	struct drm_surface *surf = lookup_drm_surface(surface);
+	_eglRemoveSurface(&surf->base);
+	if (surf->base.IsBound) {
+		surf->base.DeletePending = EGL_TRUE;
+	} else {
+		if (surf->screen)
+			drm_takedown_shown_screen(drv, surf->screen);
+		st_unreference_framebuffer(surf->stfb);
+		free(surf);
+	}
+	return EGL_TRUE;
+}
+
+EGLBoolean
+drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw)
+{
+	struct drm_surface *surf = lookup_drm_surface(draw);
+	struct pipe_surface *back_surf;
+
+	if (!surf)
+		return EGL_FALSE;
+
+	/* error checking */
+	if (!_eglSwapBuffers(drv, dpy, draw))
+		return EGL_FALSE;
+
+	back_surf = st_get_framebuffer_surface(surf->stfb,
+	                                       ST_SURFACE_BACK_LEFT);
+
+	if (back_surf) {
+
+		st_notify_swapbuffers(surf->stfb);
+
+		if (surf->screen) {
+			surf->user->pipe->flush(surf->user->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL);
+			surf->user->pipe->surface_copy(surf->user->pipe,
+				0,
+				surf->screen->surface,
+				0, 0,
+				back_surf,
+				0, 0,
+				surf->w, surf->h);
+			surf->user->pipe->flush(surf->user->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL);
+			/* TODO stuff here */
+		}
+
+		st_notify_swapbuffers_complete(surf->stfb);
+	}
+
+	return EGL_TRUE;
+}
diff --git a/src/gallium/state_trackers/egl/egl_tracker.c b/src/gallium/state_trackers/egl/egl_tracker.c
new file mode 100644
index 00000000000..3ca5acb68bf
--- /dev/null
+++ b/src/gallium/state_trackers/egl/egl_tracker.c
@@ -0,0 +1,217 @@
+
+#include "utils.h"
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include "egl_tracker.h"
+
+#include "egllog.h"
+#include "state_tracker/drm_api.h"
+
+#include "pipe/p_screen.h"
+#include "pipe/p_winsys.h"
+
+/** HACK */
+void* driDriverAPI;
+extern const struct dri_extension card_extensions[];
+
+
+/*
+ * Exported functions
+ */
+
+/**
+ * The bootstrap function.  Return a new drm_driver object and
+ * plug in API functions.
+ */
+_EGLDriver *
+_eglMain(_EGLDisplay *dpy, const char *args)
+{
+	struct drm_device *drm;
+
+	drm = (struct drm_device *) calloc(1, sizeof(struct drm_device));
+	if (!drm) {
+		return NULL;
+	}
+
+	/* First fill in the dispatch table with defaults */
+	_eglInitDriverFallbacks(&drm->base);
+	/* then plug in our Drm-specific functions */
+	drm->base.API.Initialize = drm_initialize;
+	drm->base.API.Terminate = drm_terminate;
+	drm->base.API.CreateContext = drm_create_context;
+	drm->base.API.MakeCurrent = drm_make_current;
+	drm->base.API.CreateWindowSurface = drm_create_window_surface;
+	drm->base.API.CreatePixmapSurface = drm_create_pixmap_surface;
+	drm->base.API.CreatePbufferSurface = drm_create_pbuffer_surface;
+	drm->base.API.DestroySurface = drm_destroy_surface;
+	drm->base.API.DestroyContext = drm_destroy_context;
+	drm->base.API.CreateScreenSurfaceMESA = drm_create_screen_surface_mesa;
+	drm->base.API.ShowScreenSurfaceMESA = drm_show_screen_surface_mesa;
+	drm->base.API.SwapBuffers = drm_swap_buffers;
+
+	drm->base.ClientAPIsMask = EGL_OPENGL_BIT /*| EGL_OPENGL_ES_BIT*/;
+	drm->base.Name = "DRM/Gallium/Win";
+
+	/* enable supported extensions */
+	drm->base.Extensions.MESA_screen_surface = EGL_TRUE;
+	drm->base.Extensions.MESA_copy_context = EGL_TRUE;
+
+	return &drm->base;
+}
+
+static void
+drm_get_device_id(struct drm_device *device)
+{
+	char path[512];
+	FILE *file;
+
+	/* TODO get the real minor */
+	int minor = 0;
+
+	snprintf(path, sizeof(path), "/sys/class/drm/card%d/device/device", minor);
+	file = fopen(path, "r");
+	if (!file) {
+		_eglLog(_EGL_WARNING, "Could not retrive device ID\n");
+		return;
+	}
+
+	fgets(path, sizeof( path ), file);
+	sscanf(path, "%x", &device->deviceID);
+	fclose(file);
+}
+
+static void
+drm_update_res(struct drm_device *dev)
+{
+	drmModeFreeResources(dev->res);
+	dev->res = drmModeGetResources(dev->drmFD);
+}
+
+static void
+drm_add_modes_from_connector(_EGLScreen *screen, drmModeConnectorPtr connector)
+{
+	struct drm_mode_modeinfo *m;
+	int i;
+
+	for (i = 0; i < connector->count_modes; i++) {
+		m = &connector->modes[i];
+		_eglAddNewMode(screen, m->hdisplay, m->vdisplay, m->vrefresh, m->name);
+	}
+}
+
+EGLBoolean
+drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor)
+{
+	_EGLDisplay *disp = _eglLookupDisplay(dpy);
+	struct drm_device *dev = (struct drm_device *)drv;
+	struct drm_screen *screen = NULL;
+	drmModeConnectorPtr connector = NULL;
+	drmModeResPtr res = NULL;
+	unsigned count_connectors = 0;
+	int num_screens = 0;
+	EGLint i;
+	int fd;
+
+	fd = drmOpen("i915", NULL);
+	if (fd < 0)
+		goto err_fd;
+
+	dev->drmFD = fd;
+	drm_get_device_id(dev);
+
+	dev->screen = drm_api_hocks.create_screen(dev->drmFD, dev->deviceID);
+	if (!dev->screen)
+		goto err_screen;
+	dev->winsys = dev->screen->winsys;
+
+	/* TODO HACK */
+	driInitExtensions(NULL, card_extensions, GL_FALSE);
+
+	drm_update_res(dev);
+	res = dev->res;
+	if (res)
+		count_connectors = res->count_connectors;
+	else
+		_eglLog(_EGL_WARNING, "Could not retrive kms information\n");
+
+	for(i = 0; i < count_connectors && i < MAX_SCREENS; i++) {
+		connector = drmModeGetConnector(fd, res->connectors[i]);
+
+		if (!connector)
+			continue;
+
+		if (connector->connection != DRM_MODE_CONNECTED) {
+			drmModeFreeConnector(connector);
+			continue;
+		}
+
+		screen = malloc(sizeof(struct drm_screen));
+		memset(screen, 0, sizeof(*screen));
+		screen->connector = connector;
+		screen->connectorID = connector->connector_id;
+		_eglInitScreen(&screen->base);
+		_eglAddScreen(disp, &screen->base);
+		drm_add_modes_from_connector(&screen->base, connector);
+		dev->screens[num_screens++] = screen;
+	}
+	dev->count_screens = num_screens;
+
+	/* for now we only have one config */
+	_EGLConfig *config = calloc(1, sizeof(*config));
+	memset(config, 1, sizeof(*config));
+	_eglInitConfig(config, 1);
+	_eglSetConfigAttrib(config, EGL_RED_SIZE, 8);
+	_eglSetConfigAttrib(config, EGL_GREEN_SIZE, 8);
+	_eglSetConfigAttrib(config, EGL_BLUE_SIZE, 8);
+	_eglSetConfigAttrib(config, EGL_ALPHA_SIZE, 8);
+	_eglSetConfigAttrib(config, EGL_BUFFER_SIZE, 32);
+	_eglSetConfigAttrib(config, EGL_DEPTH_SIZE, 24);
+	_eglSetConfigAttrib(config, EGL_STENCIL_SIZE, 8);
+	_eglSetConfigAttrib(config, EGL_SURFACE_TYPE, EGL_PBUFFER_BIT);
+	_eglAddConfig(disp, config);
+
+	drv->Initialized = EGL_TRUE;
+
+	*major = 1;
+	*minor = 4;
+
+	return EGL_TRUE;
+
+err_screen:
+	drmClose(fd);
+err_fd:
+	return EGL_FALSE;
+}
+
+EGLBoolean
+drm_terminate(_EGLDriver *drv, EGLDisplay dpy)
+{
+	struct drm_device *dev = (struct drm_device *)drv;
+	struct drm_screen *screen;
+	int i = 0;
+
+	drmFreeVersion(dev->version);
+
+	for (i = 0; i < dev->count_screens; i++) {
+		screen = dev->screens[i];
+
+		if (screen->shown)
+			drm_takedown_shown_screen(drv, screen);
+
+		drmModeFreeConnector(screen->connector);
+		_eglDestroyScreen(&screen->base);
+		dev->screens[i] = NULL;
+	}
+
+	dev->screen->destroy(dev->screen);
+	dev->winsys = NULL;
+
+	drmClose(dev->drmFD);
+
+	_eglCleanupDisplay(_eglLookupDisplay(dpy));
+	free(dev);
+
+	return EGL_TRUE;
+}
diff --git a/src/gallium/state_trackers/egl/egl_tracker.h b/src/gallium/state_trackers/egl/egl_tracker.h
new file mode 100644
index 00000000000..0b4dd9797d5
--- /dev/null
+++ b/src/gallium/state_trackers/egl/egl_tracker.h
@@ -0,0 +1,191 @@
+
+#ifndef _EGL_TRACKER_H_
+#define _EGL_TRACKER_H_
+
+#include <stdint.h>
+
+#include "eglconfig.h"
+#include "eglcontext.h"
+#include "egldisplay.h"
+#include "egldriver.h"
+#include "eglglobals.h"
+#include "eglmode.h"
+#include "eglscreen.h"
+#include "eglsurface.h"
+
+#include "xf86drm.h"
+#include "xf86drmMode.h"
+
+#include "pipe/p_compiler.h"
+
+#include "state_tracker/st_public.h"
+
+#define MAX_SCREENS 16
+
+struct pipe_winsys;
+struct pipe_screen;
+struct pipe_context;
+struct state_tracker;
+
+struct drm_screen;
+struct drm_context;
+
+struct drm_device
+{
+	_EGLDriver base; /* base class/object */
+
+	/*
+	 * pipe
+	 */
+
+	struct pipe_winsys *winsys;
+	struct pipe_screen *screen;
+
+	/*
+	 * drm
+	 */
+
+	int drmFD;
+	drmVersionPtr version;
+	int deviceID;
+
+	drmModeResPtr res;
+
+	struct drm_screen *screens[MAX_SCREENS];
+	size_t count_screens;
+};
+
+struct drm_surface
+{
+	_EGLSurface base; /* base class/object */
+
+	/*
+	 * pipe
+	 */
+
+
+	struct st_framebuffer *stfb;
+
+	/*
+	 * drm
+	 */
+
+	struct drm_context *user;
+	struct drm_screen *screen;
+
+	int w;
+	int h;
+};
+
+struct drm_context
+{
+	_EGLContext base; /* base class/object */
+
+	/* pipe */
+
+	struct pipe_context *pipe;
+	struct st_context *st;
+};
+
+struct drm_screen
+{
+	_EGLScreen base;
+
+	/*
+	 * pipe
+	 */
+
+	struct pipe_buffer *buffer;
+	struct pipe_texture *tex;
+	struct pipe_surface *surface;
+
+	/*
+	 * drm
+	 */
+
+	struct {
+		unsigned height;
+		unsigned width;
+		unsigned pitch;
+		unsigned handle;
+	} front;
+
+	/* currently only support one connector */
+	drmModeConnectorPtr connector;
+	uint32_t connectorID;
+
+	/* Has this screen been shown */
+	int shown;
+
+	/* Surface that is currently attached to this screen */
+	struct drm_surface *surf;
+
+	/* framebuffer */
+	drmModeFBPtr fb;
+	uint32_t fbID;
+
+	/* crtc and mode used */
+	/*drmModeCrtcPtr crtc;*/
+	uint32_t crtcID;
+
+	struct drm_mode_modeinfo *mode;
+};
+
+
+static INLINE struct drm_context *
+lookup_drm_context(EGLContext context)
+{
+	_EGLContext *c = _eglLookupContext(context);
+	return (struct drm_context *) c;
+}
+
+
+static INLINE struct drm_surface *
+lookup_drm_surface(EGLSurface surface)
+{
+	_EGLSurface *s = _eglLookupSurface(surface);
+	return (struct drm_surface *) s;
+}
+
+static INLINE struct drm_screen *
+lookup_drm_screen(EGLDisplay dpy, EGLScreenMESA screen)
+{
+	_EGLScreen *s = _eglLookupScreen(dpy, screen);
+	return (struct drm_screen *) s;
+}
+
+/**
+ * egl_visual.h
+ */
+/*@{*/
+void drm_visual_modes_destroy(__GLcontextModes *modes);
+__GLcontextModes* drm_visual_modes_create(unsigned count, size_t minimum_size);
+__GLcontextModes* drm_visual_from_config(_EGLConfig *conf);
+/*@}*/
+
+/**
+ * egl_surface.h
+ */
+/*@{*/
+void drm_takedown_shown_screen(_EGLDriver *drv, struct drm_screen *screen);
+/*@}*/
+
+/**
+ * All function exported to the egl side.
+ */
+/*@{*/
+EGLBoolean drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor);
+EGLBoolean drm_terminate(_EGLDriver *drv, EGLDisplay dpy);
+EGLContext drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext share_list, const EGLint *attrib_list);
+EGLBoolean drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context);
+EGLSurface drm_create_window_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativeWindowType window, const EGLint *attrib_list);
+EGLSurface drm_create_pixmap_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativePixmapType pixmap, const EGLint *attrib_list);
+EGLSurface drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, const EGLint *attrib_list);
+EGLSurface drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg, const EGLint *attrib_list);
+EGLBoolean drm_show_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLScreenMESA screen, EGLSurface surface, EGLModeMESA m);
+EGLBoolean drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface);
+EGLBoolean drm_make_current(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext context);
+EGLBoolean drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw);
+/*@}*/
+
+#endif
diff --git a/src/gallium/state_trackers/egl/egl_visual.c b/src/gallium/state_trackers/egl/egl_visual.c
new file mode 100644
index 00000000000..e59f893851e
--- /dev/null
+++ b/src/gallium/state_trackers/egl/egl_visual.c
@@ -0,0 +1,85 @@
+
+#include "egl_tracker.h"
+
+#include "egllog.h"
+
+void
+drm_visual_modes_destroy(__GLcontextModes *modes)
+{
+   _eglLog(_EGL_DEBUG, "%s", __FUNCTION__);
+
+   while (modes) {
+      __GLcontextModes * const next = modes->next;
+      free(modes);
+      modes = next;
+   }
+}
+
+__GLcontextModes *
+drm_visual_modes_create(unsigned count, size_t minimum_size)
+{
+	/* This code copied from libGLX, and modified */
+	const size_t size = (minimum_size > sizeof(__GLcontextModes))
+		? minimum_size : sizeof(__GLcontextModes);
+	__GLcontextModes * head = NULL;
+	__GLcontextModes ** next;
+	unsigned   i;
+
+	_eglLog(_EGL_DEBUG, "%s %d %d", __FUNCTION__, count, minimum_size);
+
+	next = & head;
+	for (i = 0 ; i < count ; i++) {
+		*next = (__GLcontextModes *) calloc(1, size);
+		if (*next == NULL) {
+			drm_visual_modes_destroy(head);
+			head = NULL;
+			break;
+		}
+
+		(*next)->doubleBufferMode = 1;
+		(*next)->visualID = GLX_DONT_CARE;
+		(*next)->visualType = GLX_DONT_CARE;
+		(*next)->visualRating = GLX_NONE;
+		(*next)->transparentPixel = GLX_NONE;
+		(*next)->transparentRed = GLX_DONT_CARE;
+		(*next)->transparentGreen = GLX_DONT_CARE;
+		(*next)->transparentBlue = GLX_DONT_CARE;
+		(*next)->transparentAlpha = GLX_DONT_CARE;
+		(*next)->transparentIndex = GLX_DONT_CARE;
+		(*next)->xRenderable = GLX_DONT_CARE;
+		(*next)->fbconfigID = GLX_DONT_CARE;
+		(*next)->swapMethod = GLX_SWAP_UNDEFINED_OML;
+		(*next)->bindToTextureRgb = GLX_DONT_CARE;
+		(*next)->bindToTextureRgba = GLX_DONT_CARE;
+		(*next)->bindToMipmapTexture = GLX_DONT_CARE;
+		(*next)->bindToTextureTargets = 0;
+		(*next)->yInverted = GLX_DONT_CARE;
+
+		next = & ((*next)->next);
+	}
+
+	return head;
+}
+
+__GLcontextModes *
+drm_visual_from_config(_EGLConfig *conf)
+{
+	__GLcontextModes *visual;
+	(void)conf;
+
+	visual = drm_visual_modes_create(1, sizeof(*visual));
+	visual->redBits = 8;
+	visual->greenBits = 8;
+	visual->blueBits = 8;
+	visual->alphaBits = 8;
+
+	visual->rgbBits = 32;
+	visual->doubleBufferMode = 1;
+
+	visual->depthBits = 24;
+	visual->haveDepthBuffer = visual->depthBits > 0;
+	visual->stencilBits = 8;
+	visual->haveStencilBuffer = visual->stencilBits > 0;
+
+	return visual;
+}
diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
index 3ce93cf49d1..da119ff1bdb 100644
--- a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
+++ b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
@@ -71,7 +71,10 @@ static int vlResizeFrameBuffer
 	basic_csc->viewport.translate[3] = 0;
 	
 	if (basic_csc->framebuffer_tex)
-		pipe_texture_release(&basic_csc->framebuffer_tex);
+	{
+		pipe_surface_reference(&basic_csc->framebuffer.cbufs[0], NULL);
+		pipe_texture_reference(&basic_csc->framebuffer_tex, NULL);
+	}
 	
 	memset(&template, 0, sizeof(struct pipe_texture));
 	template.target = PIPE_TEXTURE_2D;
@@ -153,11 +156,11 @@ static int vlPutPictureCSC
 	basic_csc = (struct vlBasicCSC*)csc;
 	pipe = basic_csc->pipe;
 
-	vs_consts = pipe->winsys->buffer_map
+	vs_consts = pipe_buffer_map
 	(
-		pipe->winsys,
+		pipe->screen,
 		basic_csc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE
+		PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
 	);
 
 	vs_consts->dst_scale.x = destw / (float)basic_csc->framebuffer.cbufs[0]->width;
@@ -178,7 +181,7 @@ static int vlPutPictureCSC
 	vs_consts->src_trans.z = 0;
 	vs_consts->src_trans.w = 0;
 
-	pipe->winsys->buffer_unmap(pipe->winsys, basic_csc->vs_const_buf.buffer);
+	pipe_buffer_unmap(pipe->screen, basic_csc->vs_const_buf.buffer);
 
 	pipe->set_sampler_textures(pipe, 1, &surface->texture);
 	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
@@ -225,17 +228,20 @@ static int vlDestroy
 	pipe = basic_csc->pipe;
 
 	if (basic_csc->framebuffer_tex)
-		pipe_texture_release(&basic_csc->framebuffer_tex);
+	{
+		pipe_surface_reference(&basic_csc->framebuffer.cbufs[0], NULL);
+		pipe_texture_reference(&basic_csc->framebuffer_tex, NULL);
+	}
 
 	pipe->delete_sampler_state(pipe, basic_csc->sampler);
 	pipe->delete_vs_state(pipe, basic_csc->vertex_shader);
 	pipe->delete_fs_state(pipe, basic_csc->fragment_shader);
 
 	for (i = 0; i < 2; ++i)
-		pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->vertex_bufs[i].buffer);
+		pipe_buffer_reference(pipe->screen, &basic_csc->vertex_bufs[i].buffer, NULL);
 
-	pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->vs_const_buf.buffer);
-	pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->fs_const_buf.buffer);
+	pipe_buffer_reference(pipe->screen, &basic_csc->vs_const_buf.buffer, NULL);
+	pipe_buffer_reference(pipe->screen, &basic_csc->fs_const_buf.buffer, NULL);
 
 	FREE(basic_csc);
 
@@ -542,9 +548,9 @@ static int vlCreateDataBufs
 	csc->vertex_bufs[0].pitch = sizeof(struct vlVertex2f);
 	csc->vertex_bufs[0].max_index = 3;
 	csc->vertex_bufs[0].buffer_offset = 0;
-	csc->vertex_bufs[0].buffer = pipe->winsys->buffer_create
+	csc->vertex_bufs[0].buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		1,
 		PIPE_BUFFER_USAGE_VERTEX,
 		sizeof(struct vlVertex2f) * 4
@@ -552,12 +558,12 @@ static int vlCreateDataBufs
 
 	memcpy
 	(
-		pipe->winsys->buffer_map(pipe->winsys, csc->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+		pipe_buffer_map(pipe->screen, csc->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
 		surface_verts,
 		sizeof(struct vlVertex2f) * 4
 	);
 
-	pipe->winsys->buffer_unmap(pipe->winsys, csc->vertex_bufs[0].buffer);
+	pipe_buffer_unmap(pipe->screen, csc->vertex_bufs[0].buffer);
 
 	csc->vertex_elems[0].src_offset = 0;
 	csc->vertex_elems[0].vertex_buffer_index = 0;
@@ -571,9 +577,9 @@ static int vlCreateDataBufs
 	csc->vertex_bufs[1].pitch = sizeof(struct vlVertex2f);
 	csc->vertex_bufs[1].max_index = 3;
 	csc->vertex_bufs[1].buffer_offset = 0;
-	csc->vertex_bufs[1].buffer = pipe->winsys->buffer_create
+	csc->vertex_bufs[1].buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		1,
 		PIPE_BUFFER_USAGE_VERTEX,
 		sizeof(struct vlVertex2f) * 4
@@ -581,12 +587,12 @@ static int vlCreateDataBufs
 
 	memcpy
 	(
-		pipe->winsys->buffer_map(pipe->winsys, csc->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+		pipe_buffer_map(pipe->screen, csc->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
 		surface_texcoords,
 		sizeof(struct vlVertex2f) * 4
 	);
 
-	pipe->winsys->buffer_unmap(pipe->winsys, csc->vertex_bufs[1].buffer);
+	pipe_buffer_unmap(pipe->screen, csc->vertex_bufs[1].buffer);
 
 	csc->vertex_elems[1].src_offset = 0;
 	csc->vertex_elems[1].vertex_buffer_index = 1;
@@ -598,11 +604,11 @@ static int vlCreateDataBufs
 	 * Const buffer contains scaling and translation vectors
 	 */
 	csc->vs_const_buf.size = sizeof(struct vlVertexShaderConsts);
-	csc->vs_const_buf.buffer = pipe->winsys->buffer_create
+	csc->vs_const_buf.buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		1,
-		PIPE_BUFFER_USAGE_CONSTANT,
+		PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
 		csc->vs_const_buf.size
 	);
 
@@ -611,9 +617,9 @@ static int vlCreateDataBufs
 	 * Const buffer contains the color conversion matrix and bias vectors
 	 */
 	csc->fs_const_buf.size = sizeof(struct vlFragmentShaderConsts);
-	csc->fs_const_buf.buffer = pipe->winsys->buffer_create
+	csc->fs_const_buf.buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		1,
 		PIPE_BUFFER_USAGE_CONSTANT,
 		csc->fs_const_buf.size
@@ -625,12 +631,12 @@ static int vlCreateDataBufs
 	 */
 	memcpy
 	(
-		pipe->winsys->buffer_map(pipe->winsys, csc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+		pipe_buffer_map(pipe->screen, csc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
 		&bt_601_full,
 		sizeof(struct vlFragmentShaderConsts)
 	);
 
-	pipe->winsys->buffer_unmap(pipe->winsys, csc->fs_const_buf.buffer);
+	pipe_buffer_unmap(pipe->screen, csc->fs_const_buf.buffer);
 
 	return 0;
 }
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
index c5a73b2bf2a..f0f82944737 100644
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
+++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
@@ -297,6 +297,7 @@ static inline int vlGrabMacroBlock
 {
 	assert(mc);
 	assert(macroblock);
+	assert(mc->num_macroblocks < mc->macroblocks_per_picture);
 
 	mc->macroblocks[mc->num_macroblocks].mbx = macroblock->mbx;
 	mc->macroblocks[mc->num_macroblocks].mby = macroblock->mby;
@@ -330,6 +331,7 @@ static inline int vlGrabMacroBlock
 }
 
 #define SET_BLOCK(vb, cbp, mbx, mby, unitx, unity, ofsx, ofsy, hx, hy, lm, cbm, crm, zb)					\
+	do {															\
 	(vb)[0].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[0].pos.y = (mby) * (unity) + (ofsy);			\
 	(vb)[1].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[1].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
 	(vb)[2].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].pos.y = (mby) * (unity) + (ofsy);			\
@@ -392,7 +394,8 @@ static inline int vlGrabMacroBlock
 		(vb)[3].cr_tc.x = (zb)[2].x + (hx);	(vb)[3].cr_tc.y = (zb)[2].y;						\
 		(vb)[4].cr_tc.x = (zb)[2].x;		(vb)[4].cr_tc.y = (zb)[2].y + (hy);					\
 		(vb)[5].cr_tc.x = (zb)[2].x + (hx);	(vb)[5].cr_tc.y = (zb)[2].y + (hy);					\
-	}
+	}															\
+	} while (0)
 
 static inline int vlGenMacroblockVerts
 (
@@ -409,6 +412,7 @@ static inline int vlGenMacroblockVerts
 	assert(mc);
 	assert(macroblock);
 	assert(ycbcr_vb);
+	assert(pos < mc->macroblocks_per_picture);
 
 	switch (macroblock->mb_type)
 	{
@@ -581,6 +585,8 @@ static int vlFlush
 	if (mc->num_macroblocks < mc->macroblocks_per_picture)
 		return 0;
 
+	assert(mc->num_macroblocks <= mc->macroblocks_per_picture);
+
 	pipe = mc->pipe;
 
 	for (i = 0; i < mc->num_macroblocks; ++i)
@@ -599,19 +605,19 @@ static int vlFlush
 		struct vlMacroBlockVertexStream0	*ycbcr_vb;
 		struct vlVertex2f			*ref_vb[2];
 
-		ycbcr_vb = (struct vlMacroBlockVertexStream0*)mc->pipe->winsys->buffer_map
+		ycbcr_vb = (struct vlMacroBlockVertexStream0*)pipe_buffer_map
 		(
-			mc->pipe->winsys,
+			pipe->screen,
 			mc->vertex_bufs.ycbcr.buffer,
-			PIPE_BUFFER_USAGE_CPU_WRITE
+			PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
 		);
 
 		for (i = 0; i < 2; ++i)
-			ref_vb[i] = (struct vlVertex2f*)mc->pipe->winsys->buffer_map
+			ref_vb[i] = (struct vlVertex2f*)pipe_buffer_map
 			(
-				mc->pipe->winsys,
+				pipe->screen,
 				mc->vertex_bufs.ref[i].buffer,
-				PIPE_BUFFER_USAGE_CPU_WRITE
+				PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
 			);
 
 		for (i = 0; i < mc->num_macroblocks; ++i)
@@ -623,15 +629,15 @@ static int vlFlush
 			offset[mb_type_ex]++;
 		}
 
-		mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs.ycbcr.buffer);
+		pipe_buffer_unmap(pipe->screen, mc->vertex_bufs.ycbcr.buffer);
 		for (i = 0; i < 2; ++i)
-			mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs.ref[i].buffer);
+			pipe_buffer_unmap(pipe->screen, mc->vertex_bufs.ref[i].buffer);
 	}
 
 	for (i = 0; i < 3; ++i)
 	{
 		pipe_surface_unmap(mc->tex_surface[i]);
-		mc->pipe->screen->tex_surface_release(mc->pipe->screen, &mc->tex_surface[i]);
+		pipe_surface_reference(&mc->tex_surface[i], NULL);
 	}
 
 	mc->render_target.cbufs[0] = pipe->screen->get_tex_surface
@@ -647,13 +653,13 @@ static int vlFlush
 	(
 		pipe->winsys,
 		mc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE
+		PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
 	);
 
 	vs_consts->denorm.x = mc->buffered_surface->texture->width[0];
 	vs_consts->denorm.y = mc->buffered_surface->texture->height[0];
 
-	pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer);
+	pipe_buffer_unmap(pipe->screen, mc->vs_const_buf.buffer);
 	pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &mc->vs_const_buf);
 	pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &mc->fs_const_buf);
 
@@ -757,7 +763,7 @@ static int vlFlush
 	}
 
 	pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &mc->buffered_surface->render_fence);
-	pipe->screen->tex_surface_release(pipe->screen, &mc->render_target.cbufs[0]);
+	pipe_surface_reference(&mc->render_target.cbufs[0], NULL);
 
 	for (i = 0; i < 3; ++i)
 		mc->zero_block[i].x = -1.0f;
@@ -808,10 +814,10 @@ static int vlRenderMacroBlocksMpeg2R16SnormBuffered
 			(
 				mc->pipe->screen,
 				mc->textures.all[i],
-				0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
+				0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
 			);
 
-			mc->texels[i] = pipe_surface_map(mc->tex_surface[i], PIPE_BUFFER_USAGE_CPU_WRITE);
+			mc->texels[i] = pipe_surface_map(mc->tex_surface[i], PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD);
 		}
 	}
 
@@ -849,11 +855,11 @@ static int vlDestroy
 		pipe->delete_sampler_state(pipe, mc->samplers.all[i]);
 
 	for (i = 0; i < 3; ++i)
-		pipe->winsys->buffer_destroy(pipe->winsys, mc->vertex_bufs.all[i].buffer);
+		pipe_buffer_reference(pipe->screen, &mc->vertex_bufs.all[i].buffer, NULL);
 
 	/* Textures 3 & 4 are not created directly, no need to release them here */
 	for (i = 0; i < 3; ++i)
-		pipe_texture_release(&mc->textures.all[i]);
+		pipe_texture_reference(&mc->textures.all[i], NULL);
 
 	pipe->delete_vs_state(pipe, mc->i_vs);
 	pipe->delete_fs_state(pipe, mc->i_fs);
@@ -866,8 +872,8 @@ static int vlDestroy
 		pipe->delete_fs_state(pipe, mc->b_fs[i]);
 	}
 
-	pipe->winsys->buffer_destroy(pipe->winsys, mc->vs_const_buf.buffer);
-	pipe->winsys->buffer_destroy(pipe->winsys, mc->fs_const_buf.buffer);
+	pipe_buffer_reference(pipe->screen, &mc->vs_const_buf.buffer, NULL);
+	pipe_buffer_reference(pipe->screen, &mc->fs_const_buf.buffer, NULL);
 
 	FREE(mc->macroblocks);
 	FREE(mc);
@@ -909,11 +915,11 @@ static int vlCreateDataBufs
 	mc->vertex_bufs.ycbcr.pitch = sizeof(struct vlVertex2f) * 4;
 	mc->vertex_bufs.ycbcr.max_index = 24 * mc->macroblocks_per_picture - 1;
 	mc->vertex_bufs.ycbcr.buffer_offset = 0;
-	mc->vertex_bufs.ycbcr.buffer = pipe->winsys->buffer_create
+	mc->vertex_bufs.ycbcr.buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		DEFAULT_BUF_ALIGNMENT,
-		PIPE_BUFFER_USAGE_VERTEX,
+		PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
 		sizeof(struct vlVertex2f) * 4 * 24 * mc->macroblocks_per_picture
 	);
 
@@ -922,11 +928,11 @@ static int vlCreateDataBufs
 		mc->vertex_bufs.all[i].pitch = sizeof(struct vlVertex2f) * 2;
 		mc->vertex_bufs.all[i].max_index = 24 * mc->macroblocks_per_picture - 1;
 		mc->vertex_bufs.all[i].buffer_offset = 0;
-		mc->vertex_bufs.all[i].buffer = pipe->winsys->buffer_create
+		mc->vertex_bufs.all[i].buffer = pipe_buffer_create
 		(
-			pipe->winsys,
+			pipe->screen,
 			DEFAULT_BUF_ALIGNMENT,
-			PIPE_BUFFER_USAGE_VERTEX,
+			PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
 			sizeof(struct vlVertex2f) * 2 * 24 * mc->macroblocks_per_picture
 		);
 	}
@@ -981,18 +987,18 @@ static int vlCreateDataBufs
 
 	/* Create our constant buffer */
 	mc->vs_const_buf.size = sizeof(struct vlVertexShaderConsts);
-	mc->vs_const_buf.buffer = pipe->winsys->buffer_create
+	mc->vs_const_buf.buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		DEFAULT_BUF_ALIGNMENT,
-		PIPE_BUFFER_USAGE_CONSTANT,
+		PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
 		mc->vs_const_buf.size
 	);
 
 	mc->fs_const_buf.size = sizeof(struct vlFragmentShaderConsts);
-	mc->fs_const_buf.buffer = pipe->winsys->buffer_create
+	mc->fs_const_buf.buffer = pipe_buffer_create
 	(
-		pipe->winsys,
+		pipe->screen,
 		DEFAULT_BUF_ALIGNMENT,
 		PIPE_BUFFER_USAGE_CONSTANT,
 		mc->fs_const_buf.size
@@ -1000,12 +1006,12 @@ static int vlCreateDataBufs
 
 	memcpy
 	(
-		pipe->winsys->buffer_map(pipe->winsys, mc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+		pipe_buffer_map(pipe->screen, mc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
 		&fs_consts,
 		sizeof(struct vlFragmentShaderConsts)
 	);
 
-	pipe->winsys->buffer_unmap(pipe->winsys, mc->fs_const_buf.buffer);
+	pipe_buffer_unmap(pipe->screen, mc->fs_const_buf.buffer);
 
 	mc->macroblocks = MALLOC(sizeof(struct vlMpeg2MacroBlock) * mc->macroblocks_per_picture);
 
diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.c b/src/gallium/state_trackers/g3dvl/vl_surface.c
index 911469f966a..0fa7b25b92e 100644
--- a/src/gallium/state_trackers/g3dvl/vl_surface.c
+++ b/src/gallium/state_trackers/g3dvl/vl_surface.c
@@ -51,6 +51,12 @@ int vlCreateSurface
 
 	sfc->texture = vlGetPipeScreen(screen)->texture_create(vlGetPipeScreen(screen), &template);
 
+	if (!sfc->texture)
+	{
+		FREE(sfc);
+		return 1;
+	}
+
 	*surface = sfc;
 
 	return 0;
@@ -63,7 +69,7 @@ int vlDestroySurface
 {
 	assert(surface);
 
-	pipe_texture_release(&surface->texture);
+	pipe_texture_reference(&surface->texture, NULL);
 	FREE(surface);
 
 	return 0;
diff --git a/src/gallium/winsys/drm/Makefile.template b/src/gallium/winsys/drm/Makefile.template
index 80e817b8082..211f4d875ed 100644
--- a/src/gallium/winsys/drm/Makefile.template
+++ b/src/gallium/winsys/drm/Makefile.template
@@ -84,14 +84,14 @@ default: depend symlinks $(LIBNAME) $(TOP)/$(LIB_DIR)/$(LIBNAME) $(LIBNAME_EGL)
 
 $(LIBNAME): $(OBJECTS) $(MESA_MODULES) $(PIPE_DRIVERS) $(WINOBJ) Makefile $(TOP)/src/mesa/drivers/dri/Makefile.template
 	$(TOP)/bin/mklib -noprefix -o $@ \
-		$(OBJECTS) $(PIPE_DRIVERS) $(MESA_MODULES)  $(WINOBJ) $(DRI_LIB_DEPS)
+		$(OBJECTS) $(PIPE_DRIVERS) $(MESA_MODULES)  $(WINOBJ) $(DRI_LIB_DEPS) $(DRIVER_EXTRAS)
 
 $(LIBNAME_EGL): $(WINSYS_OBJECTS) $(LIBS)
 	$(TOP)/bin/mklib -o $(LIBNAME_EGL) \
 		-linker "$(CC)" \
 		-noprefix \
 		$(OBJECTS) $(MKLIB_OPTIONS) $(WINSYS_OBJECTS) $(PIPE_DRIVERS) $(WINOBJ) $(DRI_LIB_DEPS) \
-		--whole-archive $(LIBS) $(GALLIUM_AUXILIARIES) --no-whole-archive
+		--whole-archive $(LIBS) $(GALLIUM_AUXILIARIES) --no-whole-archive $(DRIVER_EXTRAS)
 
 $(TOP)/$(LIB_DIR)/$(LIBNAME): $(LIBNAME)
 	$(INSTALL) $(LIBNAME) $(TOP)/$(LIB_DIR) 
diff --git a/src/gallium/winsys/drm/intel/Makefile b/src/gallium/winsys/drm/intel/Makefile
index a670ac044d0..eede9fc866c 100644
--- a/src/gallium/winsys/drm/intel/Makefile
+++ b/src/gallium/winsys/drm/intel/Makefile
@@ -2,7 +2,7 @@ TOP = ../../../../..
 include $(TOP)/configs/current
 
 
-SUBDIRS = common dri egl
+SUBDIRS = gem egl
 
 
 default: subdirs
diff --git a/src/gallium/winsys/drm/intel/egl/Makefile b/src/gallium/winsys/drm/intel/egl/Makefile
index f0b5a443894..4b22d17ccfe 100644
--- a/src/gallium/winsys/drm/intel/egl/Makefile
+++ b/src/gallium/winsys/drm/intel/egl/Makefile
@@ -6,21 +6,23 @@ LIBNAME = EGL_i915.so
 PIPE_DRIVERS = \
 	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(TOP)/src/gallium/drivers/i915simple/libi915simple.a \
-	../common/libinteldrm.a
+	$(TOP)/src/gallium/state_trackers/egl/libegldrm.a \
+	../gem/libinteldrm.a
 
 DRIVER_SOURCES = \
-	intel_swapbuffers.c \
 	intel_context.c \
 	intel_device.c \
-	intel_egl.c
+	intel_api.c
 
 C_SOURCES = \
 	$(COMMON_GALLIUM_SOURCES) \
 	$(DRIVER_SOURCES)
 
+DRIVER_EXTRAS = -ldrm_intel
+
 ASM_SOURCES = 
 
-DRIVER_DEFINES = -I../common $(shell pkg-config libdrm --atleast-version=2.3.1 \
+DRIVER_DEFINES = -I../gem $(shell pkg-config libdrm --atleast-version=2.3.1 \
 				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
 
 include ../../Makefile.template
diff --git a/src/gallium/winsys/drm/intel/egl/SConscript b/src/gallium/winsys/drm/intel/egl/SConscript
deleted file mode 100644
index 0ad19d42a85..00000000000
--- a/src/gallium/winsys/drm/intel/egl/SConscript
+++ /dev/null
@@ -1,39 +0,0 @@
-Import('*')
-
-env = drienv.Clone()
-
-env.Append(CPPPATH = [
-	'../intel',
-	'server'
-])
-
-#MINIGLX_SOURCES = server/intel_dri.c
-
-DRIVER_SOURCES = [
-	'intel_winsys_pipe.c',
-	'intel_winsys_softpipe.c',
-	'intel_winsys_i915.c',
-	'intel_batchbuffer.c',
-	'intel_swapbuffers.c',
-	'intel_context.c',
-	'intel_lock.c',
-	'intel_screen.c',
-	'intel_batchpool.c',
-]
-
-sources = \
-	COMMON_GALLIUM_SOURCES + \
-	COMMON_BM_SOURCES + \
-	DRIVER_SOURCES
-
-drivers = [
-	softpipe,
-	i915simple
-]
-
-# TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions
-env.SharedLibrary(
-	target ='i915tex_dri.so',
-	source = sources,
-	LIBS = drivers + mesa + auxiliaries + env['LIBS'],
-)
\ No newline at end of file
diff --git a/src/gallium/winsys/drm/intel/egl/intel_api.c b/src/gallium/winsys/drm/intel/egl/intel_api.c
new file mode 100644
index 00000000000..5dc4a7b052b
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/egl/intel_api.c
@@ -0,0 +1,10 @@
+
+#include "intel_api.h"
+
+struct drm_api drm_api_hocks =
+{
+	.create_screen = intel_create_screen,
+	.create_context = intel_create_context,
+	.buffer_from_handle = intel_be_buffer_from_handle,
+	.handle_from_buffer = intel_be_handle_from_buffer,
+};
diff --git a/src/gallium/winsys/drm/intel/egl/intel_api.h b/src/gallium/winsys/drm/intel/egl/intel_api.h
new file mode 100644
index 00000000000..8ec165ab017
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/egl/intel_api.h
@@ -0,0 +1,14 @@
+
+#ifndef _INTEL_API_H_
+#define _INTEL_API_H_
+
+#include "pipe/p_compiler.h"
+
+#include "state_tracker/drm_api.h"
+
+#include "intel_be_device.h"
+
+struct pipe_screen *intel_create_screen(int drmFD, int pciID);
+struct pipe_context *intel_create_context(struct pipe_screen *screen);
+
+#endif
diff --git a/src/gallium/winsys/drm/intel/egl/intel_batchbuffer.h b/src/gallium/winsys/drm/intel/egl/intel_batchbuffer.h
deleted file mode 100644
index 3e953261689..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_batchbuffer.h
+++ /dev/null
@@ -1,24 +0,0 @@
-#ifndef INTEL_BATCHBUFFER_H
-#define INTEL_BATCHBUFFER_H
-
-#include "intel_be_batchbuffer.h"
-
-/*
- * Need to redefine the BATCH defines
- */
-
-#undef BEGIN_BATCH
-#define BEGIN_BATCH(dwords, relocs) \
-   (i915_batchbuffer_check(&intel->base.batch->base, dwords, relocs))
-
-#undef OUT_BATCH
-#define OUT_BATCH(d) \
-   i915_batchbuffer_dword(&intel->base.batch->base, d)
-
-#undef OUT_RELOC
-#define OUT_RELOC(buf,flags,mask,delta) do { 					\
-   assert((delta) >= 0);							\
-   intel_be_offset_relocation(intel->base.batch, delta, buf, flags, mask); 	\
-} while (0)
-
-#endif
diff --git a/src/gallium/winsys/drm/intel/egl/intel_context.c b/src/gallium/winsys/drm/intel/egl/intel_context.c
index 927addb834c..57e5ff7bc12 100644
--- a/src/gallium/winsys/drm/intel/egl/intel_context.c
+++ b/src/gallium/winsys/drm/intel/egl/intel_context.c
@@ -1,119 +1,21 @@
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
 
 #include "i915simple/i915_screen.h"
 
-#include "intel_device.h"
-#include "intel_context.h"
-#include "intel_batchbuffer.h"
+#include "intel_be_device.h"
+#include "intel_be_context.h"
 
-#include "state_tracker/st_public.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "intel_egl.h"
-#include "utils.h"
 
-#ifdef DEBUG
-int __intel_debug = 0;
-#endif
+#include "intel_api.h"
 
+struct intel_context
+{
+	struct intel_be_context base;
 
-#define need_GL_ARB_multisample
-#define need_GL_ARB_point_parameters
-#define need_GL_ARB_texture_compression
-#define need_GL_ARB_vertex_buffer_object
-#define need_GL_ARB_vertex_program
-#define need_GL_ARB_window_pos
-#define need_GL_EXT_blend_color
-#define need_GL_EXT_blend_equation_separate
-#define need_GL_EXT_blend_func_separate
-#define need_GL_EXT_blend_minmax
-#define need_GL_EXT_cull_vertex
-#define need_GL_EXT_fog_coord
-#define need_GL_EXT_framebuffer_object
-#define need_GL_EXT_multi_draw_arrays
-#define need_GL_EXT_secondary_color
-#define need_GL_NV_vertex_program
-#include "extension_helper.h"
-
-
-/**
- * Extension strings exported by the intel driver.
- *
- * \note
- * It appears that ARB_texture_env_crossbar has "disappeared" compared to the
- * old i830-specific driver.
- */
-const struct dri_extension card_extensions[] = {
-	{"GL_ARB_multisample", GL_ARB_multisample_functions},
-	{"GL_ARB_multitexture", NULL},
-	{"GL_ARB_point_parameters", GL_ARB_point_parameters_functions},
-	{"GL_ARB_texture_border_clamp", NULL},
-	{"GL_ARB_texture_compression", GL_ARB_texture_compression_functions},
-	{"GL_ARB_texture_cube_map", NULL},
-	{"GL_ARB_texture_env_add", NULL},
-	{"GL_ARB_texture_env_combine", NULL},
-	{"GL_ARB_texture_env_dot3", NULL},
-	{"GL_ARB_texture_mirrored_repeat", NULL},
-	{"GL_ARB_texture_rectangle", NULL},
-	{"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions},
-	{"GL_ARB_pixel_buffer_object", NULL},
-	{"GL_ARB_vertex_program", GL_ARB_vertex_program_functions},
-	{"GL_ARB_window_pos", GL_ARB_window_pos_functions},
-	{"GL_EXT_blend_color", GL_EXT_blend_color_functions},
-	{"GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions},
-	{"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions},
-	{"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions},
-	{"GL_EXT_blend_subtract", NULL},
-	{"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions},
-	{"GL_EXT_fog_coord", GL_EXT_fog_coord_functions},
-	{"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions},
-	{"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions},
-	{"GL_EXT_packed_depth_stencil", NULL},
-	{"GL_EXT_pixel_buffer_object", NULL},
-	{"GL_EXT_secondary_color", GL_EXT_secondary_color_functions},
-	{"GL_EXT_stencil_wrap", NULL},
-	{"GL_EXT_texture_edge_clamp", NULL},
-	{"GL_EXT_texture_env_combine", NULL},
-	{"GL_EXT_texture_env_dot3", NULL},
-	{"GL_EXT_texture_filter_anisotropic", NULL},
-	{"GL_EXT_texture_lod_bias", NULL},
-	{"GL_3DFX_texture_compression_FXT1", NULL},
-	{"GL_APPLE_client_storage", NULL},
-	{"GL_MESA_pack_invert", NULL},
-	{"GL_MESA_ycbcr_texture", NULL},
-	{"GL_NV_blend_square", NULL},
-	{"GL_NV_vertex_program", GL_NV_vertex_program_functions},
-	{"GL_NV_vertex_program1_1", NULL},
-	{"GL_SGIS_generate_mipmap", NULL },
-	{NULL, NULL}
+	/* stuff */
 };
 
-
 /*
  * Hardware lock functions.
  * Doesn't do anything in EGL
@@ -142,101 +44,40 @@ intel_locked_hardware(struct intel_be_context *context)
 /*
  * Misc functions.
  */
+static void
+intel_destroy_be_context(struct i915_winsys *winsys)
+{
+	struct intel_context *intel = (struct intel_context *)winsys;
+
+	intel_be_destroy_context(&intel->base);
+	free(intel);
+}
 
-int
-intel_create_context(struct egl_drm_context *egl_context, const __GLcontextModes *visual, void *sharedContextPrivate)
+struct pipe_context *
+intel_create_context(struct pipe_screen *screen)
 {
-	struct intel_context *intel = CALLOC_STRUCT(intel_context);
-	struct intel_device *device = (struct intel_device *)egl_context->device->priv;
+	struct intel_context *intel;
 	struct pipe_context *pipe;
-	struct st_context *st_share = NULL;
+	struct intel_be_device *device = (struct intel_be_device *)screen->winsys;
 
-	egl_context->priv = intel;
-
-	intel->intel_device = device;
-	intel->egl_context = egl_context;
-	intel->egl_device = egl_context->device;
+	intel = (struct intel_context *)malloc(sizeof(*intel));
+	memset(intel, 0, sizeof(*intel));
 
 	intel->base.hardware_lock = intel_lock_hardware;
 	intel->base.hardware_unlock = intel_unlock_hardware;
 	intel->base.hardware_locked = intel_locked_hardware;
 
-	intel_be_init_context(&intel->base, &device->base);
+	intel_be_init_context(&intel->base, device);
+
+	intel->base.base.destroy = intel_destroy_be_context;
 
 #if 0
 	pipe = intel_create_softpipe(intel, screen->winsys);
 #else
-	pipe = i915_create_context(device->pipe, &device->base.base, &intel->base.base);
+	pipe = i915_create_context(screen, &device->base, &intel->base.base);
 #endif
 
 	pipe->priv = intel;
 
-	intel->st = st_create_context(pipe, visual, st_share);
-
-	device->dummy = intel;
-
-	return TRUE;
-}
-
-int
-intel_destroy_context(struct egl_drm_context *egl_context)
-{
-	struct intel_context *intel = egl_context->priv;
-
-	if (intel->intel_device->dummy == intel)
-		intel->intel_device->dummy = NULL;
-
-	st_destroy_context(intel->st);
-	intel_be_destroy_context(&intel->base);
-	free(intel);
-	return TRUE;
-}
-
-void
-intel_make_current(struct egl_drm_context *context, struct egl_drm_drawable *draw, struct egl_drm_drawable *read)
-{
-	if (context) {
-		struct intel_context *intel = (struct intel_context *)context->priv;
-		struct intel_framebuffer *draw_fb = (struct intel_framebuffer *)draw->priv;
-		struct intel_framebuffer *read_fb = (struct intel_framebuffer *)read->priv;
-
-		assert(draw_fb->stfb);
-		assert(read_fb->stfb);
-
-		st_make_current(intel->st, draw_fb->stfb, read_fb->stfb);
-
-		intel->egl_drawable = draw;
-
-		st_resize_framebuffer(draw_fb->stfb, draw->w, draw->h);
-
-		if (draw != read)
-			st_resize_framebuffer(read_fb->stfb, read->w, read->h);
-
-	} else {
-		st_make_current(NULL, NULL, NULL);
-	}
-}
-
-void
-intel_bind_frontbuffer(struct egl_drm_drawable *draw, struct egl_drm_frontbuffer *front)
-{
-	struct intel_device *device = (struct intel_device *)draw->device->priv;
-	struct intel_framebuffer *draw_fb = (struct intel_framebuffer *)draw->priv;
-
-	if (draw_fb->front_buffer)
-		driBOUnReference(draw_fb->front_buffer);
-
-	draw_fb->front_buffer = NULL;
-	draw_fb->front = NULL;
-
-	/* to unbind just call this function with front == NULL */
-	if (!front)
-		return;
-
-	draw_fb->front = front;
-
-	driGenBuffers(device->base.staticPool, "front", 1, &draw_fb->front_buffer, 0, 0, 0);
-	driBOSetReferenced(draw_fb->front_buffer, front->handle);
-
-	st_resize_framebuffer(draw_fb->stfb, draw->w, draw->h);
+	return pipe;
 }
diff --git a/src/gallium/winsys/drm/intel/egl/intel_context.h b/src/gallium/winsys/drm/intel/egl/intel_context.h
deleted file mode 100644
index 477fdec7f70..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_context.h
+++ /dev/null
@@ -1,118 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-#ifndef INTEL_CONTEXT_H
-#define INTEL_CONTEXT_H
-
-#include "pipe/p_debug.h"
-#include "intel_be_context.h"
-
-
-struct st_context;
-struct egl_drm_device;
-struct egl_drm_context;
-struct egl_drm_frontbuffer;
-
-
-/**
- * Intel rendering context, contains a state tracker and intel-specific info.
- */
-struct intel_context
-{
-	struct intel_be_context base;
-
-	struct st_context *st;
-
-	struct intel_device *intel_device;
-
-	/* new egl stuff */
-	struct egl_drm_device *egl_device;
-	struct egl_drm_context *egl_context;
-	struct egl_drm_drawable *egl_drawable;
-};
-
-
-
-/**
- * Intel framebuffer.
- */
-struct intel_framebuffer
-{
-	struct st_framebuffer *stfb;
-
-	struct intel_device *device;
-	struct _DriBufferObject *front_buffer;
-	struct egl_drm_frontbuffer *front;
-};
-
-
-
-
-/* These are functions now:
- */
-void LOCK_HARDWARE( struct intel_context *intel );
-void UNLOCK_HARDWARE( struct intel_context *intel );
-
-extern char *__progname;
-
-
-
-/* ================================================================
- * Debugging:
- */
-#ifdef DEBUG
-extern int __intel_debug;
-
-#define DEBUG_SWAP	0x1
-#define DEBUG_LOCK      0x2
-#define DEBUG_IOCTL	0x4
-#define DEBUG_BATCH     0x8
-
-#define DBG(flag, ...)  do { 			\
-   if (__intel_debug & (DEBUG_##flag)) 		\
-      printf(__VA_ARGS__); 		\
-} while(0)
-
-#else
-#define DBG(flag, ...)
-#endif
-
-
-#define PCI_CHIP_845_G			0x2562
-#define PCI_CHIP_I830_M			0x3577
-#define PCI_CHIP_I855_GM		0x3582
-#define PCI_CHIP_I865_G			0x2572
-#define PCI_CHIP_I915_G			0x2582
-#define PCI_CHIP_I915_GM		0x2592
-#define PCI_CHIP_I945_G			0x2772
-#define PCI_CHIP_I945_GM		0x27A2
-#define PCI_CHIP_I945_GME		0x27AE
-#define PCI_CHIP_G33_G			0x29C2
-#define PCI_CHIP_Q35_G			0x29B2
-#define PCI_CHIP_Q33_G			0x29D2
-
-#endif
diff --git a/src/gallium/winsys/drm/intel/egl/intel_device.c b/src/gallium/winsys/drm/intel/egl/intel_device.c
index 1964745c994..6b281402d53 100644
--- a/src/gallium/winsys/drm/intel/egl/intel_device.c
+++ b/src/gallium/winsys/drm/intel/egl/intel_device.c
@@ -1,137 +1,48 @@
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
 
-
-#include "utils.h"
-
-#include "state_tracker/st_public.h"
+#include <stdio.h>
+#include "pipe/p_defines.h"
+#include "intel_be_device.h"
 #include "i915simple/i915_screen.h"
 
-#include "intel_context.h"
-#include "intel_device.h"
-#include "intel_batchbuffer.h"
-#include "intel_egl.h"
-
-
-extern const struct dri_extension card_extensions[];
-
+#include "intel_api.h"
 
-int
-intel_create_device(struct egl_drm_device *device)
+struct intel_device
 {
-	struct intel_device *intel_device;
-
-	/* Allocate the private area */
-	intel_device = CALLOC_STRUCT(intel_device);
-	if (!intel_device)
-		return FALSE;
-
-	device->priv = (void *)intel_device;
-	intel_device->device = device;
-
-	intel_device->deviceID = device->deviceID;
-
-	intel_be_init_device(&intel_device->base, device->drmFD, intel_device->deviceID);
+	struct intel_be_device base;
 
-	intel_device->pipe = i915_create_screen(&intel_device->base.base, intel_device->deviceID);
+	int deviceID;
+};
 
-	/* hack */
-	driInitExtensions(NULL, card_extensions, GL_FALSE);
-
-	return TRUE;
-}
-
-int
-intel_destroy_device(struct egl_drm_device *device)
+static void
+intel_destroy_winsys(struct pipe_winsys *winsys)
 {
-	struct intel_device *intel_device = (struct intel_device *)device->priv;
-	
-	intel_be_destroy_device(&intel_device->base);
+	struct intel_device *dev = (struct intel_device *)winsys;
 
-	free(intel_device);
-	device->priv = NULL;
+	intel_be_destroy_device(&dev->base);
 
-	return TRUE;
+	free(dev);
 }
 
-int
-intel_create_drawable(struct egl_drm_drawable *drawable,
-                      const __GLcontextModes * visual)
+struct pipe_screen *
+intel_create_screen(int drmFD, int deviceID)
 {
-	enum pipe_format colorFormat, depthFormat, stencilFormat;
-	struct intel_framebuffer *intelfb = CALLOC_STRUCT(intel_framebuffer);
-
-	if (!intelfb)
-		return GL_FALSE;
-
-	intelfb->device = drawable->device->priv;
+	struct intel_device *dev;
+	struct pipe_screen *screen;
 
-	if (visual->redBits == 5)
-		colorFormat = PIPE_FORMAT_R5G6B5_UNORM;
-	else
-		colorFormat = PIPE_FORMAT_A8R8G8B8_UNORM;
-
-	if (visual->depthBits == 16)
-		depthFormat = PIPE_FORMAT_Z16_UNORM;
-	else if (visual->depthBits == 24)
-		depthFormat = PIPE_FORMAT_S8Z24_UNORM;
-	else
-		depthFormat = PIPE_FORMAT_NONE;
+	/* Allocate the private area */
+	dev = malloc(sizeof(*dev));
+	if (!dev)
+		return NULL;
+	memset(dev, 0, sizeof(*dev));
 
-	if (visual->stencilBits == 8)
-		stencilFormat = PIPE_FORMAT_S8Z24_UNORM;
-	else
-		stencilFormat = PIPE_FORMAT_NONE;
+	dev->deviceID = deviceID;
 
-	intelfb->stfb = st_create_framebuffer(visual,
-			colorFormat,
-			depthFormat,
-			stencilFormat,
-			drawable->w,
-			drawable->h,
-			(void*) intelfb);
+	intel_be_init_device(&dev->base, drmFD, deviceID);
 
-	if (!intelfb->stfb) {
-		free(intelfb);
-		return GL_FALSE;
-	}
+	/* we need to hock our own destroy function in here */
+	dev->base.base.destroy = intel_destroy_winsys;
 
-	drawable->priv = (void *) intelfb;
-	return GL_TRUE;
-}
-
-int
-intel_destroy_drawable(struct egl_drm_drawable *drawable)
-{
-	struct intel_framebuffer *intelfb = (struct intel_framebuffer *)drawable->priv;
-	drawable->priv = NULL;
+	screen = i915_create_screen(&dev->base.base, deviceID);
 
-	assert(intelfb->stfb);
-	st_unreference_framebuffer(intelfb->stfb);
-	free(intelfb);
-	return TRUE;
+	return screen;
 }
diff --git a/src/gallium/winsys/drm/intel/egl/intel_device.h b/src/gallium/winsys/drm/intel/egl/intel_device.h
deleted file mode 100644
index 323a7c2aef7..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_device.h
+++ /dev/null
@@ -1,50 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-#ifndef _INTEL_SCREEN_H_
-#define _INTEL_SCREEN_H_
-
-#include "intel_be_device.h"
-
-#include "pipe/p_compiler.h"
-
-struct pipe_screen;
-struct egl_drm_device;
-struct intel_context;
-
-struct intel_device
-{
-	struct intel_be_device base;
-	struct pipe_screen *pipe;
-
-	int deviceID;
-	struct egl_drm_device *device;
-
-	struct intel_context *dummy;
-};
-
-#endif
diff --git a/src/gallium/winsys/drm/intel/egl/intel_egl.c b/src/gallium/winsys/drm/intel/egl/intel_egl.c
deleted file mode 100644
index ed464076ee3..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_egl.c
+++ /dev/null
@@ -1,796 +0,0 @@
-
-#include <assert.h>
-#include <stdlib.h>
-#include <string.h>
-#include <stdio.h>
-#include <stdint.h>
-
-#include "eglconfig.h"
-#include "eglcontext.h"
-#include "egldisplay.h"
-#include "egldriver.h"
-#include "eglglobals.h"
-#include "eglmode.h"
-#include "eglscreen.h"
-#include "eglsurface.h"
-#include "egllog.h"
-
-#include "intel_egl.h"
-
-#include "xf86drm.h"
-#include "xf86drmMode.h"
-
-#include "intel_context.h"
-
-#include "state_tracker/st_public.h"
-
-#define MAX_SCREENS 16
-
-static void
-drm_get_device_id(struct egl_drm_device *device)
-{
-	char path[512];
-	FILE *file;
-
-	/* TODO get the real minor */
-	int minor = 0;
-
-	snprintf(path, sizeof(path), "/sys/class/drm/card%d/device/device", minor);
-	file = fopen(path, "r");
-	if (!file) {
-		_eglLog(_EGL_WARNING, "Could not retrive device ID\n");
-		return;
-	}
-
-	fgets(path, sizeof( path ), file);
-	sscanf(path, "%x", &device->deviceID);
-	fclose(file);
-}
-
-static struct egl_drm_device*
-egl_drm_create_device(int drmFD)
-{
-	struct egl_drm_device *device = malloc(sizeof(*device));
-	memset(device, 0, sizeof(*device));
-	device->drmFD = drmFD;
-
-	device->version = drmGetVersion(device->drmFD);
-
-	drm_get_device_id(device);
-
-	if (!intel_create_device(device)) {
-		free(device);
-		return NULL;
-	}
-
-	return device;
-}
-
-static void
-_egl_context_modes_destroy(__GLcontextModes *modes)
-{
-   _eglLog(_EGL_DEBUG, "%s", __FUNCTION__);
-
-   while (modes) {
-      __GLcontextModes * const next = modes->next;
-      free(modes);
-      modes = next;
-   }
-}
-/**
- * Create a linked list of 'count' GLcontextModes.
- * These are used during the client/server visual negotiation phase,
- * then discarded.
- */
-static __GLcontextModes *
-_egl_context_modes_create(unsigned count, size_t minimum_size)
-{
-   /* This code copied from libGLX, and modified */
-   const size_t size = (minimum_size > sizeof(__GLcontextModes))
-       ? minimum_size : sizeof(__GLcontextModes);
-   __GLcontextModes * head = NULL;
-   __GLcontextModes ** next;
-   unsigned   i;
-
-   _eglLog(_EGL_DEBUG, "%s %d %d", __FUNCTION__, count, minimum_size);
-
-   next = & head;
-   for (i = 0 ; i < count ; i++) {
-      *next = (__GLcontextModes *) calloc(1, size);
-      if (*next == NULL) {
-	 _egl_context_modes_destroy(head);
-	 head = NULL;
-	 break;
-      }
-      
-      (*next)->doubleBufferMode = 1;
-      (*next)->visualID = GLX_DONT_CARE;
-      (*next)->visualType = GLX_DONT_CARE;
-      (*next)->visualRating = GLX_NONE;
-      (*next)->transparentPixel = GLX_NONE;
-      (*next)->transparentRed = GLX_DONT_CARE;
-      (*next)->transparentGreen = GLX_DONT_CARE;
-      (*next)->transparentBlue = GLX_DONT_CARE;
-      (*next)->transparentAlpha = GLX_DONT_CARE;
-      (*next)->transparentIndex = GLX_DONT_CARE;
-      (*next)->xRenderable = GLX_DONT_CARE;
-      (*next)->fbconfigID = GLX_DONT_CARE;
-      (*next)->swapMethod = GLX_SWAP_UNDEFINED_OML;
-      (*next)->bindToTextureRgb = GLX_DONT_CARE;
-      (*next)->bindToTextureRgba = GLX_DONT_CARE;
-      (*next)->bindToMipmapTexture = GLX_DONT_CARE;
-      (*next)->bindToTextureTargets = 0;
-      (*next)->yInverted = GLX_DONT_CARE;
-
-      next = & ((*next)->next);
-   }
-
-   return head;
-}
-
-struct drm_screen;
-
-struct drm_driver
-{
-	_EGLDriver base;  /* base class/object */
-
-	drmModeResPtr res;
-
-	struct drm_screen *screens[MAX_SCREENS];
-	size_t count_screens;
-
-	struct egl_drm_device *device;
-};
-
-struct drm_surface
-{
-	_EGLSurface base;  /* base class/object */
-
-	struct egl_drm_drawable *drawable;
-};
-
-struct drm_context
-{
-	_EGLContext base;  /* base class/object */
-
-	struct egl_drm_context *context;
-};
-
-struct drm_screen
-{
-	_EGLScreen base;
-
-	/* currently only support one connector */
-	drmModeConnectorPtr connector;
-
-	/* Has this screen been shown */
-	int shown;
-
-	/* Surface that is currently attached to this screen */
-	struct drm_surface *surf;
-
-	/* backing buffer */
-	drmBO buffer;
-
-	/* framebuffer */
-	drmModeFBPtr fb;
-	uint32_t fbID;
-
-	/* crtc and mode used */
-	drmModeCrtcPtr crtc;
-	uint32_t crtcID;
-
-	struct drm_mode_modeinfo *mode;
-
-	/* geometry of the screen */
-	struct egl_drm_frontbuffer front;
-};
-
-static void
-drm_update_res(struct drm_driver *drm_drv)
-{
-	drmModeFreeResources(drm_drv->res);
-	drm_drv->res = drmModeGetResources(drm_drv->device->drmFD);
-}
-
-static void
-drm_add_modes_from_connector(_EGLScreen *screen, drmModeConnectorPtr connector)
-{
-	struct drm_mode_modeinfo *m;
-	int i;
-
-	for (i = 0; i < connector->count_modes; i++) {
-		m = &connector->modes[i];
-		_eglAddNewMode(screen, m->hdisplay, m->vdisplay, m->vrefresh, m->name);
-	}
-}
-
-
-static EGLBoolean
-drm_initialize(_EGLDriver *drv, EGLDisplay dpy, EGLint *major, EGLint *minor)
-{
-	_EGLDisplay *disp = _eglLookupDisplay(dpy);
-	struct drm_driver *drm_drv = (struct drm_driver *)drv;
-	struct drm_screen *screen = NULL;
-	drmModeConnectorPtr connector = NULL;
-	drmModeResPtr res = NULL;
-	unsigned count_connectors = 0;
-	int num_screens = 0;
-
-	EGLint i;
-	int fd;
-
-	fd = drmOpen("i915", NULL);
-	if (fd < 0) {
-		return EGL_FALSE;
-	}
-
-	drm_drv->device = egl_drm_create_device(fd);
-	if (!drm_drv->device) {
-		drmClose(fd);
-		return EGL_FALSE;
-	}
-
-	drm_update_res(drm_drv);
-	res = drm_drv->res;
-	if (res)
-		count_connectors = res->count_connectors;
-
-	for(i = 0; i < count_connectors && i < MAX_SCREENS; i++) {
-		connector = drmModeGetConnector(fd, res->connectors[i]);
-
-		if (!connector)
-			continue;
-
-		if (connector->connection != DRM_MODE_CONNECTED) {
-			drmModeFreeConnector(connector);
-			continue;
-		}
-
-		screen = malloc(sizeof(struct drm_screen));
-		memset(screen, 0, sizeof(*screen));
-		screen->connector = connector;
-		_eglInitScreen(&screen->base);
-		_eglAddScreen(disp, &screen->base);
-		drm_add_modes_from_connector(&screen->base, connector);
-		drm_drv->screens[num_screens++] = screen;
-	}
-	drm_drv->count_screens = num_screens;
-
-	/* for now we only have one config */
-	_EGLConfig *config = calloc(1, sizeof(*config));
-	memset(config, 1, sizeof(*config));
-	_eglInitConfig(config, 1);
-	_eglSetConfigAttrib(config, EGL_RED_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_GREEN_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_BLUE_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_ALPHA_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_BUFFER_SIZE, 32);
-	_eglSetConfigAttrib(config, EGL_DEPTH_SIZE, 24);
-	_eglSetConfigAttrib(config, EGL_STENCIL_SIZE, 8);
-	_eglSetConfigAttrib(config, EGL_SURFACE_TYPE, EGL_PBUFFER_BIT);
-	_eglAddConfig(disp, config);
-
-	drv->Initialized = EGL_TRUE;
-
-	*major = 1;
-	*minor = 4;
-
-	return EGL_TRUE;
-}
-
-static void
-drm_takedown_shown_screen(_EGLDriver *drv, struct drm_screen *screen)
-{
-	struct drm_driver *drm_drv = (struct drm_driver *)drv;
-	unsigned int i;
-
-	intel_bind_frontbuffer(screen->surf->drawable, NULL);
-	screen->surf = NULL;
-
-	for (i = 0; i < drm_drv->res->count_crtcs; i++) {
-		drmModeSetCrtc(
-			drm_drv->device->drmFD,
-			drm_drv->res->crtcs[i],
-			0, // FD
-			0, 0,
-			NULL, 0, // List of output ids
-			NULL);
-	}
-
-	drmModeRmFB(drm_drv->device->drmFD, screen->fbID);
-	drmModeFreeFB(screen->fb);
-	screen->fb = NULL;
-
-	drmBOUnreference(drm_drv->device->drmFD, &screen->buffer);
-
-	screen->shown = 0;
-}
-
-static EGLBoolean
-drm_terminate(_EGLDriver *drv, EGLDisplay dpy)
-{
-	struct drm_driver *drm_drv = (struct drm_driver *)drv;
-	struct drm_screen *screen;
-	int i = 0;
-
-	intel_destroy_device(drm_drv->device);
-	drmFreeVersion(drm_drv->device->version);
-
-	for (i = 0; i < drm_drv->count_screens; i++) {
-		screen = drm_drv->screens[i];
-
-		if (screen->shown)
-			drm_takedown_shown_screen(drv, screen);
-
-		drmModeFreeConnector(screen->connector);
-		_eglDestroyScreen(&screen->base);
-		drm_drv->screens[i] = NULL;
-	}
-
-	drmClose(drm_drv->device->drmFD);
-
-	free(drm_drv->device);
-
-	_eglCleanupDisplay(_eglLookupDisplay(dpy));
-	free(drm_drv);
-
-	return EGL_TRUE;
-}
-
-
-static struct drm_context *
-lookup_drm_context(EGLContext context)
-{
-	_EGLContext *c = _eglLookupContext(context);
-	return (struct drm_context *) c;
-}
-
-
-static struct drm_surface *
-lookup_drm_surface(EGLSurface surface)
-{
-	_EGLSurface *s = _eglLookupSurface(surface);
-	return (struct drm_surface *) s;
-}
-
-static struct drm_screen *
-lookup_drm_screen(EGLDisplay dpy, EGLScreenMESA screen)
-{
-	_EGLScreen *s = _eglLookupScreen(dpy, screen);
-	return (struct drm_screen *) s;
-}
-
-static __GLcontextModes*
-visual_from_config(_EGLConfig *conf)
-{
-	__GLcontextModes *visual;
-	(void)conf;
-
-	visual = _egl_context_modes_create(1, sizeof(*visual));
-	visual->redBits = 8;
-	visual->greenBits = 8;
-	visual->blueBits = 8;
-	visual->alphaBits = 8;
-
-	visual->rgbBits = 32;
-	visual->doubleBufferMode = 1;
-
-	visual->depthBits = 24;
-	visual->haveDepthBuffer = visual->depthBits > 0;
-	visual->stencilBits = 8;
-	visual->haveStencilBuffer = visual->stencilBits > 0;
-
-	return visual;
-}
-
-
-
-static EGLContext
-drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext share_list, const EGLint *attrib_list)
-{
-	struct drm_driver *drm_drv = (struct drm_driver *)drv;
-	struct drm_context *c;
-	struct drm_egl_context *share = NULL;
-	_EGLConfig *conf;
-	int i;
-	int ret;
-	__GLcontextModes *visual;
-	struct egl_drm_context *context;
-
-	conf = _eglLookupConfig(drv, dpy, config);
-	if (!conf) {
-		_eglError(EGL_BAD_CONFIG, "eglCreateContext");
-		return EGL_NO_CONTEXT;
-	}
-
-	for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) {
-		switch (attrib_list[i]) {
-			/* no attribs defined for now */
-			default:
-				_eglError(EGL_BAD_ATTRIBUTE, "eglCreateContext");
-				return EGL_NO_CONTEXT;
-		}
-	}
-
-	c = (struct drm_context *) calloc(1, sizeof(struct drm_context));
-	if (!c)
-		return EGL_NO_CONTEXT;
-
-	_eglInitContext(drv, dpy, &c->base, config, attrib_list);
-
-	context = malloc(sizeof(*context));
-	memset(context, 0, sizeof(*context));
-
-	if (!context)
-		goto err_c;
-
-	context->device = drm_drv->device;
-	visual = visual_from_config(conf);
-
-	ret = intel_create_context(context, visual, share);
-	free(visual);
-
-	if (!ret)
-		goto err_gl;
-
-	c->context = context;
-
-	/* generate handle and insert into hash table */
-	_eglSaveContext(&c->base);
-	assert(_eglGetContextHandle(&c->base));
-
-	return _eglGetContextHandle(&c->base);
-err_gl:
-	free(context);
-err_c:
-	free(c);
-	return EGL_NO_CONTEXT;
-}
-
-static EGLBoolean
-drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context)
-{
-	struct drm_context *fc = lookup_drm_context(context);
-	_eglRemoveContext(&fc->base);
-	if (fc->base.IsBound) {
-		fc->base.DeletePending = EGL_TRUE;
-	} else {
-		intel_destroy_context(fc->context);
-		free(fc->context);
-		free(fc);
-	}
-	return EGL_TRUE;
-}
-
-
-static EGLSurface
-drm_create_window_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativeWindowType window, const EGLint *attrib_list)
-{
-	return EGL_NO_SURFACE;
-}
-
-
-static EGLSurface
-drm_create_pixmap_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, NativePixmapType pixmap, const EGLint *attrib_list)
-{
-	return EGL_NO_SURFACE;
-}
-
-
-static EGLSurface
-drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
-                         const EGLint *attrib_list)
-{
-	struct drm_driver *drm_drv = (struct drm_driver *)drv;
-	int i;
-	int ret;
-	int width = -1;
-	int height = -1;
-	struct drm_surface *surf = NULL;
-	struct egl_drm_drawable *drawable = NULL;
-	__GLcontextModes *visual;
-	_EGLConfig *conf;
-
-	conf = _eglLookupConfig(drv, dpy, config);
-	if (!conf) {
-		_eglError(EGL_BAD_CONFIG, "eglCreatePbufferSurface");
-		return EGL_NO_CONTEXT;
-	}
-
-	for (i = 0; attrib_list && attrib_list[i] != EGL_NONE; i++) {
-		switch (attrib_list[i]) {
-			case EGL_WIDTH:
-				width = attrib_list[++i];
-				break;
-			case EGL_HEIGHT:
-				height = attrib_list[++i];
-				break;
-			default:
-				_eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface");
-				return EGL_NO_SURFACE;
-		}
-	}
-
-	if (width < 1 || height < 1) {
-		_eglError(EGL_BAD_ATTRIBUTE, "eglCreatePbufferSurface");
-		return EGL_NO_SURFACE;
-	}
-
-	surf = (struct drm_surface *) calloc(1, sizeof(struct drm_surface));
-	if (!surf)
-		goto err;
-
-	if (!_eglInitSurface(drv, dpy, &surf->base, EGL_PBUFFER_BIT, config, attrib_list))
-		goto err_surf;
-
-	drawable = malloc(sizeof(*drawable));
-	memset(drawable, 0, sizeof(*drawable));
-
-	drawable->w = width;
-	drawable->h = height;
-
-	visual = visual_from_config(conf);
-
-	drawable->device = drm_drv->device;
-	ret = intel_create_drawable(drawable, visual);
-	free(visual);
-
-	if (!ret)
-		goto err_draw;
-
-	surf->drawable = drawable;
-
-	_eglSaveSurface(&surf->base);
-	return surf->base.Handle;
-
-err_draw:
-	free(drawable);
-err_surf:
-	free(surf);
-err:
-	return EGL_NO_SURFACE;
-}
-
-static EGLSurface
-drm_create_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy, EGLConfig cfg,
-                               const EGLint *attrib_list)
-{
-	EGLSurface surf = drm_create_pbuffer_surface(drv, dpy, cfg, attrib_list);
-
-	return surf;
-}
-
-static struct drm_mode_modeinfo *
-drm_find_mode(drmModeConnectorPtr connector, _EGLMode *mode)
-{
-	int i;
-	struct drm_mode_modeinfo *m = NULL;
-
-	for (i = 0; i < connector->count_modes; i++) {
-		m = &connector->modes[i];
-		if (m->hdisplay == mode->Width && m->vdisplay == mode->Height && m->vrefresh == mode->RefreshRate)
-			break;
-		m = &connector->modes[0]; /* if we can't find one, return first */
-	}
-
-	return m;
-}
-static void
-draw(size_t x, size_t y, size_t w, size_t h, size_t pitch, size_t v, unsigned int *ptr)
-{
-    int i, j;
-
-    for (i = x; i < x + w; i++)
-        for(j = y; j < y + h; j++)
-            ptr[(i * pitch / 4) + j] = v;
-
-}
-
-static void
-prettyColors(int fd, unsigned int handle, size_t pitch)
-{
-	drmBO bo;
-	unsigned int *ptr;
-	void *p;
-	int i;
-
-	drmBOReference(fd, handle, &bo);
-	drmBOMap(fd, &bo, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0, &p);
-	ptr = (unsigned int*)p;
-
-	for (i = 0; i < (bo.size / 4); i++)
-		ptr[i] = 0xFFFFFFFF;
-
-	for (i = 0; i < 4; i++)
-		draw(i * 40, i * 40, 40, 40, pitch, 0, ptr);
-
-
-	draw(200, 100, 40, 40, pitch, 0xff00ff, ptr);
-	draw(100, 200, 40, 40, pitch, 0xff00ff, ptr);
-
-	drmBOUnmap(fd, &bo);
-}
-
-static EGLBoolean
-drm_show_screen_surface_mesa(_EGLDriver *drv, EGLDisplay dpy,
-                         EGLScreenMESA screen,
-                         EGLSurface surface, EGLModeMESA m)
-{
-	struct drm_driver *drm_drv = (struct drm_driver *)drv;
-	struct drm_surface *surf = lookup_drm_surface(surface);
-	struct drm_screen *scrn = lookup_drm_screen(dpy, screen);
-	_EGLMode *mode = _eglLookupMode(dpy, m);
-	size_t pitch = mode->Width * 4;
-	size_t size = mode->Height * pitch;
-	int ret;
-	unsigned int i,j,k;
-
-	if (scrn->shown)
-		drm_takedown_shown_screen(drv, scrn);
-
-	ret = drmBOCreate(drm_drv->device->drmFD, size, 0, 0,
-		DRM_BO_FLAG_READ |
-		DRM_BO_FLAG_WRITE |
-		DRM_BO_FLAG_MEM_TT |
-		DRM_BO_FLAG_MEM_VRAM |
-		DRM_BO_FLAG_NO_EVICT,
-		DRM_BO_HINT_DONT_FENCE, &scrn->buffer);
-
-	if (ret)
-		return EGL_FALSE;
-
-	prettyColors(drm_drv->device->drmFD, scrn->buffer.handle, pitch);
-
-	ret = drmModeAddFB(drm_drv->device->drmFD, mode->Width, mode->Height,
-			32, 32, pitch,
-			scrn->buffer.handle,
-			&scrn->fbID);
-
-	if (ret)
-		goto err_bo;
-
-	scrn->fb = drmModeGetFB(drm_drv->device->drmFD, scrn->fbID);
-	if (!scrn->fb)
-		goto err_bo;
-
-	for (j = 0; j < drm_drv->res->count_connectors; j++) {
-		drmModeConnector *con = drmModeGetConnector(drm_drv->device->drmFD, drm_drv->res->connectors[j]);
-		scrn->mode = drm_find_mode(con, mode);
-		if (!scrn->mode)
-			goto err_fb;
-
-		for (k = 0; k < con->count_encoders; k++) {
-			drmModeEncoder *enc = drmModeGetEncoder(drm_drv->device->drmFD, con->encoders[k]);
-			for (i = 0; i < drm_drv->res->count_crtcs; i++) {
-				if (enc->possible_crtcs & (1<<i)) {
-					ret = drmModeSetCrtc(
-						drm_drv->device->drmFD,
-						drm_drv->res->crtcs[i],
-						scrn->fbID,
-						0, 0,
-						&drm_drv->res->connectors[j], 1,
-						scrn->mode);	
-					/* skip the other crtcs now */
-					i = drm_drv->res->count_crtcs;
-				}
-			}
-		}
-	}
-
-	scrn->front.handle = scrn->buffer.handle;
-	scrn->front.pitch = pitch;
-	scrn->front.width = mode->Width;
-	scrn->front.height = mode->Height;
-
-	scrn->surf = surf;
-	intel_bind_frontbuffer(surf->drawable, &scrn->front);
-
-	scrn->shown = 1;
-
-	return EGL_TRUE;
-
-err_fb:
-	/* TODO remove fb */
-
-err_bo:
-	drmBOUnreference(drm_drv->device->drmFD, &scrn->buffer);
-	return EGL_FALSE;
-}
-
-static EGLBoolean
-drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface)
-{
-	struct drm_surface *fs = lookup_drm_surface(surface);
-	_eglRemoveSurface(&fs->base);
-	if (fs->base.IsBound) {
-		fs->base.DeletePending = EGL_TRUE;
-	} else {
-		intel_bind_frontbuffer(fs->drawable, NULL);
-		intel_destroy_drawable(fs->drawable);
-		free(fs->drawable);
-		free(fs);
-	}
-	return EGL_TRUE;
-}
-
-
-static EGLBoolean
-drm_make_current(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext context)
-{
-	struct drm_surface *readSurf = lookup_drm_surface(read);
-	struct drm_surface *drawSurf = lookup_drm_surface(draw);
-	struct drm_context *ctx = lookup_drm_context(context);
-	EGLBoolean b;
-
-	b = _eglMakeCurrent(drv, dpy, draw, read, context);
-	if (!b)
-		return EGL_FALSE;
-
-	if (ctx) {
-		if (!drawSurf || !readSurf)
-			return EGL_FALSE;
-
-		intel_make_current(ctx->context, drawSurf->drawable, readSurf->drawable);
-	} else {
-		intel_make_current(NULL, NULL, NULL);
-	}
-
-	return EGL_TRUE;
-}
-
-static EGLBoolean
-drm_swap_buffers(_EGLDriver *drv, EGLDisplay dpy, EGLSurface draw)
-{
-	struct drm_surface *surf = lookup_drm_surface(draw);
-	if (!surf)
-		return EGL_FALSE;
-
-	/* error checking */
-	if (!_eglSwapBuffers(drv, dpy, draw))
-		return EGL_FALSE;
-
-	intel_swap_buffers(surf->drawable);
-	return EGL_TRUE;
-}
-
-
-/**
- * The bootstrap function.  Return a new drm_driver object and
- * plug in API functions.
- */
-_EGLDriver *
-_eglMain(_EGLDisplay *dpy, const char *args)
-{
-	struct drm_driver *drm;
-
-	drm = (struct drm_driver *) calloc(1, sizeof(struct drm_driver));
-	if (!drm) {
-		return NULL;
-	}
-
-	/* First fill in the dispatch table with defaults */
-	_eglInitDriverFallbacks(&drm->base);
-	/* then plug in our Drm-specific functions */
-	drm->base.API.Initialize = drm_initialize;
-	drm->base.API.Terminate = drm_terminate;
-	drm->base.API.CreateContext = drm_create_context;
-	drm->base.API.MakeCurrent = drm_make_current;
-	drm->base.API.CreateWindowSurface = drm_create_window_surface;
-	drm->base.API.CreatePixmapSurface = drm_create_pixmap_surface;
-	drm->base.API.CreatePbufferSurface = drm_create_pbuffer_surface;
-	drm->base.API.DestroySurface = drm_destroy_surface;
-	drm->base.API.DestroyContext = drm_destroy_context;
-	drm->base.API.CreateScreenSurfaceMESA = drm_create_screen_surface_mesa;
-	drm->base.API.ShowScreenSurfaceMESA = drm_show_screen_surface_mesa;
-	drm->base.API.SwapBuffers = drm_swap_buffers;
-
-	drm->base.ClientAPIsMask = EGL_OPENGL_BIT /*| EGL_OPENGL_ES_BIT*/;
-	drm->base.Name = "DRM/Gallium";
-
-	/* enable supported extensions */
-	drm->base.Extensions.MESA_screen_surface = EGL_TRUE;
-	drm->base.Extensions.MESA_copy_context = EGL_TRUE;
-
-	return &drm->base;
-}
diff --git a/src/gallium/winsys/drm/intel/egl/intel_egl.h b/src/gallium/winsys/drm/intel/egl/intel_egl.h
deleted file mode 100644
index 1ee27e0847a..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_egl.h
+++ /dev/null
@@ -1,53 +0,0 @@
-
-#ifndef _INTEL_EGL_H_
-#define _INTEL_EGL_H_
-
-#include <xf86drm.h>
-
-struct egl_drm_device
-{
-	void *priv;
-	int drmFD;
-
-	drmVersionPtr version;
-	int deviceID;
-};
-
-struct egl_drm_context
-{
-	void *priv;
-	struct egl_drm_device *device;
-};
-
-struct egl_drm_drawable
-{
-	void *priv;
-	struct egl_drm_device *device;
-	size_t h;
-	size_t w;
-};
-
-struct egl_drm_frontbuffer
-{
-	uint32_t handle;
-	uint32_t pitch;
-	uint32_t width;
-	uint32_t height;
-};
-
-#include "GL/internal/glcore.h"
-
-int intel_create_device(struct egl_drm_device *device);
-int intel_destroy_device(struct egl_drm_device *device);
-
-int intel_create_context(struct egl_drm_context *context, const __GLcontextModes *visual, void *sharedContextPrivate);
-int intel_destroy_context(struct egl_drm_context *context);
-
-int intel_create_drawable(struct egl_drm_drawable *drawable, const __GLcontextModes * visual);
-int intel_destroy_drawable(struct egl_drm_drawable *drawable);
-
-void intel_make_current(struct egl_drm_context *context, struct egl_drm_drawable *draw, struct egl_drm_drawable *read);
-void intel_swap_buffers(struct egl_drm_drawable *draw);
-void intel_bind_frontbuffer(struct egl_drm_drawable *draw, struct egl_drm_frontbuffer *front);
-
-#endif
diff --git a/src/gallium/winsys/drm/intel/egl/intel_reg.h b/src/gallium/winsys/drm/intel/egl/intel_reg.h
deleted file mode 100644
index 4f33bee4385..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_reg.h
+++ /dev/null
@@ -1,53 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-
-#ifndef _INTEL_REG_H_
-#define _INTEL_REG_H_
-
-
-#define BR00_BITBLT_CLIENT   0x40000000
-#define BR00_OP_COLOR_BLT    0x10000000
-#define BR00_OP_SRC_COPY_BLT 0x10C00000
-#define BR13_SOLID_PATTERN   0x80000000
-
-#define XY_COLOR_BLT_CMD		((2<<29)|(0x50<<22)|0x4)
-#define XY_COLOR_BLT_WRITE_ALPHA	(1<<21)
-#define XY_COLOR_BLT_WRITE_RGB		(1<<20)
-
-#define XY_SRC_COPY_BLT_CMD             ((2<<29)|(0x53<<22)|6)
-#define XY_SRC_COPY_BLT_WRITE_ALPHA     (1<<21)
-#define XY_SRC_COPY_BLT_WRITE_RGB       (1<<20)
-
-#define MI_WAIT_FOR_EVENT               ((0x3<<23))
-#define MI_WAIT_FOR_PLANE_B_FLIP        (1<<6)
-#define MI_WAIT_FOR_PLANE_A_FLIP        (1<<2)
-
-#define MI_BATCH_BUFFER_END 	        (0xA<<23)
-
-
-#endif
diff --git a/src/gallium/winsys/drm/intel/egl/intel_swapbuffers.c b/src/gallium/winsys/drm/intel/egl/intel_swapbuffers.c
deleted file mode 100644
index 2edcbc79fff..00000000000
--- a/src/gallium/winsys/drm/intel/egl/intel_swapbuffers.c
+++ /dev/null
@@ -1,111 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-#include "intel_device.h"
-#include "intel_context.h"
-#include "intel_batchbuffer.h"
-#include "intel_reg.h"
-
-#include "pipe/p_context.h"
-#include "state_tracker/st_public.h"
-#include "state_tracker/st_context.h"
-#include "state_tracker/st_cb_fbo.h"
-#include "intel_egl.h"
-
-
-static void
-intel_display_surface(struct egl_drm_drawable *draw,
-                      struct pipe_surface *surf);
-
-void intel_swap_buffers(struct egl_drm_drawable *draw)
-{
-	struct intel_framebuffer *intel_fb = (struct intel_framebuffer *)draw->priv;
-	struct pipe_surface *back_surf;
-
-	assert(intel_fb);
-	assert(intel_fb->stfb);
-
-	back_surf = st_get_framebuffer_surface(intel_fb->stfb, ST_SURFACE_BACK_LEFT);
-	if (back_surf) {
-		st_notify_swapbuffers(intel_fb->stfb);
-		if (intel_fb->front)
-			intel_display_surface(draw, back_surf);
-		st_notify_swapbuffers_complete(intel_fb->stfb);
-	}
-}
-
-static void
-intel_display_surface(struct egl_drm_drawable *draw,
-                      struct pipe_surface *surf)
-{
-	struct intel_context *intel = NULL;
-	struct intel_framebuffer *intel_fb = (struct intel_framebuffer *)draw->priv;
-	struct _DriFenceObject *fence;
-
-	//const int srcWidth = surf->width;
-	//const int srcHeight = surf->height;
-
-	intel = intel_fb->device->dummy;
-	if (!intel) {
-		printf("No dummy context\n");
-		return;
-	}
-
-	const int dstWidth = intel_fb->front->width;
-	const int dstHeight = intel_fb->front->height;
-	const int dstPitch = intel_fb->front->pitch / 4;//draw->front.cpp;
-
-	const int cpp = 4;//intel_fb->front->cpp;
-	const int srcPitch = surf->stride / cpp;
-
-	int BR13, CMD;
-	//int i;
-
-	BR13 = (dstPitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
-	CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
-			XY_SRC_COPY_BLT_WRITE_RGB);
-
-	BEGIN_BATCH(8, 2);
-	OUT_BATCH(CMD);
-	OUT_BATCH(BR13);
-	OUT_BATCH((0 << 16) | 0);
-	OUT_BATCH((dstHeight << 16) | dstWidth);
-
-	OUT_RELOC(intel_fb->front_buffer,
-		DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
-		DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
-
-	OUT_BATCH((0 << 16) | 0);
-	OUT_BATCH((srcPitch * cpp) & 0xffff);
-	OUT_RELOC(dri_bo(surf->buffer),
-			DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
-			DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
-
-	fence = intel_be_batchbuffer_flush(intel->base.batch);
-	driFenceUnReference(&fence);
-	intel_be_batchbuffer_finish(intel->base.batch);
-}
diff --git a/src/gallium/winsys/drm/intel/gem/Makefile b/src/gallium/winsys/drm/intel/gem/Makefile
new file mode 100644
index 00000000000..b25fc258f45
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/Makefile
@@ -0,0 +1,18 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = inteldrm
+
+C_SOURCES = \
+	intel_be_batchbuffer.c \
+	intel_be_context.c \
+	intel_be_device.c
+
+
+include ./Makefile.template
+
+DRIVER_DEFINES = $(shell pkg-config libdrm --cflags \
+                && pkg-config libdrm --atleast-version=2.3.1 \
+                && echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
+symlinks:
+
diff --git a/src/gallium/winsys/drm/intel/gem/Makefile.template b/src/gallium/winsys/drm/intel/gem/Makefile.template
new file mode 100644
index 00000000000..557070ae02a
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/Makefile.template
@@ -0,0 +1,64 @@
+# -*-makefile-*-
+
+
+# We still have a dependency on the "dri" buffer manager.  Most likely
+# the interface can be reused in non-dri environments, and also as a
+# frontend to simpler memory managers.
+#
+COMMON_SOURCES =
+
+OBJECTS = $(C_SOURCES:.c=.o) \
+	$(CPP_SOURCES:.cpp=.o) \
+	$(ASM_SOURCES:.S=.o)
+
+
+### Include directories
+INCLUDES = \
+	-I. \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/auxiliary \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/include \
+	$(DRIVER_INCLUDES)
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.cpp.o:
+	$(CXX) -c $(INCLUDES) $(CXXFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES)  $< -o $@
+
+
+##### TARGETS #####
+
+default: depend symlinks $(LIBNAME)
+
+
+$(LIBNAME): $(OBJECTS) Makefile Makefile.template
+	$(TOP)/bin/mklib -o $@ -static $(OBJECTS) $(DRIVER_LIBS)
+
+
+depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DRIVER_DEFINES) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) \
+		$(ASM_SOURCES) 2> /dev/null
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean::
+	-rm -f *.o */*.o *~ *.so *.a *~ server/*.o $(SYMLINKS)
+	-rm -f depend depend.bak
+
+
+include depend
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c
new file mode 100644
index 00000000000..e83a4c42cd0
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c
@@ -0,0 +1,139 @@
+
+#include "intel_be_batchbuffer.h"
+#include "intel_be_context.h"
+#include "intel_be_device.h"
+#include "intel_be_fence.h"
+#include <errno.h>
+
+#include "util/u_memory.h"
+
+struct intel_be_batchbuffer *
+intel_be_batchbuffer_alloc(struct intel_be_context *intel)
+{
+	struct intel_be_batchbuffer *batch = CALLOC_STRUCT(intel_be_batchbuffer);
+
+
+	batch->base.buffer = NULL;
+	batch->base.winsys = &intel->base;
+	batch->base.map = NULL;
+	batch->base.ptr = NULL;
+	batch->base.size = 0;
+	batch->base.actual_size = intel->device->max_batch_size;
+	batch->base.relocs = 0;
+	batch->base.max_relocs = INTEL_DEFAULT_RELOCS;
+
+	batch->base.map = malloc(batch->base.actual_size);
+	memset(batch->base.map, 0, batch->base.actual_size);
+
+	batch->base.ptr = batch->base.map;
+
+	intel_be_batchbuffer_reset(batch);
+
+	return batch;
+}
+
+void
+intel_be_batchbuffer_reset(struct intel_be_batchbuffer *batch)
+{
+	struct intel_be_context *intel = intel_be_context(batch->base.winsys);
+	struct intel_be_device *dev = intel->device;
+
+	if (batch->bo)
+		drm_intel_bo_unreference(batch->bo);
+
+	memset(batch->base.map, 0, batch->base.actual_size);
+	batch->base.ptr = batch->base.map;
+	batch->base.size = batch->base.actual_size - BATCH_RESERVED;
+
+	batch->base.relocs = 0;
+	batch->base.max_relocs = INTEL_DEFAULT_RELOCS;
+
+	batch->bo = drm_intel_bo_alloc(dev->pools.gem,
+	                               "gallium3d_batch_buffer",
+	                               batch->base.actual_size, 0);
+}
+
+int
+intel_be_offset_relocation(struct intel_be_batchbuffer *batch,
+			   unsigned pre_add,
+			   drm_intel_bo *bo,
+			   uint32_t read_domains,
+			   uint32_t write_domain)
+{
+	unsigned offset;
+	int ret = 0;
+
+	assert(batch->base.relocs < batch->base.max_relocs);
+
+	offset = (unsigned)(batch->base.ptr - batch->base.map);
+	batch->base.ptr += 4;
+
+	ret = drm_intel_bo_emit_reloc(bo, pre_add,
+	                              batch->bo, offset,
+	                              read_domains,
+	                              write_domain);
+
+	if (!ret)
+		batch->base.relocs++;
+
+	return ret;
+}
+
+void
+intel_be_batchbuffer_flush(struct intel_be_batchbuffer *batch,
+			   struct intel_be_fence **fence)
+{
+	struct i915_batchbuffer *i915 = &batch->base;
+	unsigned used = 0;
+	int ret = 0;
+
+	assert(i915_batchbuffer_space(i915) >= 0);
+
+	used = batch->base.ptr - batch->base.map;
+	assert((used & 3) == 0);
+
+	if (used & 4) {
+		((uint32_t *) batch->base.ptr)[0] = ((0<<29)|(4<<23)); // MI_FLUSH;
+		((uint32_t *) batch->base.ptr)[1] = 0;
+		((uint32_t *) batch->base.ptr)[2] = (0xA<<23); // MI_BATCH_BUFFER_END;
+		batch->base.ptr += 12;
+	} else {
+		((uint32_t *) batch->base.ptr)[0] = ((0<<29)|(4<<23)); // MI_FLUSH;
+		((uint32_t *) batch->base.ptr)[1] = (0xA<<23); // MI_BATCH_BUFFER_END;
+		batch->base.ptr += 8;
+	}
+
+	used = batch->base.ptr - batch->base.map;
+
+	drm_intel_bo_subdata(batch->bo, 0, used, batch->base.map);
+	ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0);
+
+	assert(ret == 0);
+
+	intel_be_batchbuffer_reset(batch);
+
+	if (fence) {
+		if (*fence)
+			intel_be_fence_unreference(*fence);
+
+		(*fence) = CALLOC_STRUCT(intel_be_fence);
+		(*fence)->refcount = 1;
+		(*fence)->bo = NULL;
+	}
+}
+
+void
+intel_be_batchbuffer_finish(struct intel_be_batchbuffer *batch)
+{
+
+}
+
+void
+intel_be_batchbuffer_free(struct intel_be_batchbuffer *batch)
+{
+	if (batch->bo)
+		drm_intel_bo_unreference(batch->bo);
+
+	free(batch->base.map);
+	free(batch);
+}
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.h b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.h
new file mode 100644
index 00000000000..195bf8dee77
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.h
@@ -0,0 +1,55 @@
+
+#ifndef INTEL_BE_BATCHBUFFER_H
+#define INTEL_BE_BATCHBUFFER_H
+
+#include "i915simple/i915_batch.h"
+
+#include "drm.h"
+#include "intel_bufmgr.h"
+
+#define BATCH_RESERVED 16
+
+#define INTEL_DEFAULT_RELOCS 100
+#define INTEL_MAX_RELOCS 400
+
+#define INTEL_BATCH_NO_CLIPRECTS 0x1
+#define INTEL_BATCH_CLIPRECTS    0x2
+
+struct intel_be_context;
+struct intel_be_device;
+struct intel_be_fence;
+
+struct intel_be_batchbuffer
+{
+	struct i915_batchbuffer base;
+
+	struct intel_be_context *intel;
+	struct intel_be_device *device;
+
+	drm_intel_bo *bo;
+};
+
+struct intel_be_batchbuffer *
+intel_be_batchbuffer_alloc(struct intel_be_context *intel);
+
+void
+intel_be_batchbuffer_free(struct intel_be_batchbuffer *batch);
+
+void
+intel_be_batchbuffer_finish(struct intel_be_batchbuffer *batch);
+
+void
+intel_be_batchbuffer_flush(struct intel_be_batchbuffer *batch,
+			   struct intel_be_fence **fence);
+
+void
+intel_be_batchbuffer_reset(struct intel_be_batchbuffer *batch);
+
+int
+intel_be_offset_relocation(struct intel_be_batchbuffer *batch,
+			   unsigned pre_add,
+			   drm_intel_bo *bo,
+			   uint32_t read_domains,
+			   uint32_t write_doman);
+
+#endif
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_context.c b/src/gallium/winsys/drm/intel/gem/intel_be_context.c
new file mode 100644
index 00000000000..3e472e1e433
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_context.c
@@ -0,0 +1,78 @@
+
+#include "intel_be_device.h"
+#include "intel_be_context.h"
+#include "intel_be_batchbuffer.h"
+
+#include "i915_drm.h"
+
+static struct i915_batchbuffer *
+intel_be_batch_get(struct i915_winsys *sws)
+{
+	struct intel_be_context *intel = intel_be_context(sws);
+	return &intel->batch->base;
+}
+
+static void
+intel_be_batch_reloc(struct i915_winsys *sws,
+		     struct pipe_buffer *buf,
+		     unsigned access_flags,
+		     unsigned delta)
+{
+	struct intel_be_context *intel = intel_be_context(sws);
+	drm_intel_bo *bo = intel_bo(buf);
+	int ret;
+	uint32_t read = 0;
+	uint32_t write = 0;
+
+	if (access_flags & I915_BUFFER_ACCESS_WRITE) {
+		write = I915_GEM_DOMAIN_RENDER;
+	}
+
+	if (access_flags & I915_BUFFER_ACCESS_READ) {
+		read = I915_GEM_DOMAIN_SAMPLER |
+		       I915_GEM_DOMAIN_VERTEX;
+	}
+
+	ret = intel_be_offset_relocation(intel->batch,
+	                                 delta,
+	                                 bo,
+	                                 read,
+	                                 write);
+	/* TODO change return type */
+	/* return ret; */
+}
+
+static void
+intel_be_batch_flush(struct i915_winsys *sws,
+		     struct pipe_fence_handle **fence)
+{
+	struct intel_be_context *intel = intel_be_context(sws);
+	struct intel_be_fence **f = (struct intel_be_fence **)fence;
+
+	if (fence && *fence)
+		assert(0);
+
+	intel_be_batchbuffer_flush(intel->batch, f);
+}
+
+boolean
+intel_be_init_context(struct intel_be_context *intel, struct intel_be_device *device)
+{
+	assert(intel);
+	assert(device);
+	intel->device = device;
+
+	intel->base.batch_get = intel_be_batch_get;
+	intel->base.batch_reloc = intel_be_batch_reloc;
+	intel->base.batch_flush = intel_be_batch_flush;
+
+	intel->batch = intel_be_batchbuffer_alloc(intel);
+
+	return true;
+}
+
+void
+intel_be_destroy_context(struct intel_be_context *intel)
+{
+	intel_be_batchbuffer_free(intel->batch);
+}
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_context.h b/src/gallium/winsys/drm/intel/gem/intel_be_context.h
new file mode 100644
index 00000000000..9cee1a4e52b
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_context.h
@@ -0,0 +1,48 @@
+
+#ifndef INTEL_BE_CONTEXT_H
+#define INTEL_BE_CONTEXT_H
+
+#include "i915simple/i915_winsys.h"
+
+struct intel_be_context
+{
+	/** Interface to i915simple driver */
+	struct i915_winsys base;
+
+	struct intel_be_device *device;
+	struct intel_be_batchbuffer *batch;
+
+	/*
+	 * Hardware lock functions.
+	 *
+	 * Needs to be filled in by the winsys.
+	 */
+	void (*hardware_lock)(struct intel_be_context *context);
+	void (*hardware_unlock)(struct intel_be_context *context);
+	boolean (*hardware_locked)(struct intel_be_context *context);
+};
+
+static INLINE struct intel_be_context *
+intel_be_context(struct i915_winsys *sws)
+{
+	return (struct intel_be_context *)sws;
+}
+
+/**
+ * Intialize a allocated intel_be_context struct.
+ *
+ * Remember to set the hardware_* functions.
+ */
+boolean
+intel_be_init_context(struct intel_be_context *intel,
+		      struct intel_be_device *device);
+
+/**
+ * Destroy a intel_be_context.
+ *
+ * Does not free the struct that is up to the winsys.
+ */
+void
+intel_be_destroy_context(struct intel_be_context *intel);
+
+#endif
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_device.c b/src/gallium/winsys/drm/intel/gem/intel_be_device.c
new file mode 100644
index 00000000000..201a4850498
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_device.c
@@ -0,0 +1,267 @@
+
+#include "intel_be_device.h"
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+#include "pipe/p_inlines.h"
+#include "util/u_memory.h"
+
+#include "intel_be_fence.h"
+
+#include "i915simple/i915_screen.h"
+
+
+/**
+ *  Turn a pipe winsys into an intel/pipe winsys:
+ */
+static INLINE struct intel_be_device *
+intel_be_device(struct pipe_winsys *winsys)
+{
+	return (struct intel_be_device *)winsys;
+}
+
+/*
+ * Buffer
+ */
+
+static void *
+intel_be_buffer_map(struct pipe_winsys *winsys,
+		    struct pipe_buffer *buf,
+		    unsigned flags)
+{
+	drm_intel_bo *bo = intel_bo(buf);
+	int write = 0;
+	int ret;
+
+	if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
+		write = 1;
+
+	ret = drm_intel_bo_map(bo, write);
+
+	if (ret)
+		return NULL;
+
+	return bo->virtual;
+}
+
+static void
+intel_be_buffer_unmap(struct pipe_winsys *winsys,
+		      struct pipe_buffer *buf)
+{
+	drm_intel_bo_unmap(intel_bo(buf));
+}
+
+static void
+intel_be_buffer_destroy(struct pipe_winsys *winsys,
+			struct pipe_buffer *buf)
+{
+	drm_intel_bo_unreference(intel_bo(buf));
+	free(buf);
+}
+
+static struct pipe_buffer *
+intel_be_buffer_create(struct pipe_winsys *winsys,
+		       unsigned alignment,
+		       unsigned usage,
+		       unsigned size)
+{
+	struct intel_be_buffer *buffer = CALLOC_STRUCT(intel_be_buffer);
+	struct intel_be_device *dev = intel_be_device(winsys);
+	drm_intel_bufmgr *pool;
+	char *name;
+
+	if (!buffer)
+		return NULL;
+
+	buffer->base.refcount = 1;
+	buffer->base.alignment = alignment;
+	buffer->base.usage = usage;
+	buffer->base.size = size;
+
+	if (usage & (PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_CONSTANT)) {
+		/* Local buffer */
+		name = "gallium3d_local";
+		pool = dev->pools.gem;
+	} else if (usage & PIPE_BUFFER_USAGE_CUSTOM) {
+		/* For vertex buffers */
+		name = "gallium3d_internal_vertex";
+		pool = dev->pools.gem;
+	} else {
+		/* Regular buffers */
+		name = "gallium3d_regular";
+		pool = dev->pools.gem;
+	}
+
+	buffer->bo = drm_intel_bo_alloc(pool, name, size, alignment);
+
+	if (!buffer->bo)
+		goto err;
+
+	return &buffer->base;
+
+err:
+	free(buffer);
+	return NULL;
+}
+
+static struct pipe_buffer *
+intel_be_user_buffer_create(struct pipe_winsys *winsys, void *ptr, unsigned bytes)
+{
+	struct intel_be_buffer *buffer = CALLOC_STRUCT(intel_be_buffer);
+	struct intel_be_device *dev = intel_be_device(winsys);
+	int ret;
+
+	if (!buffer)
+		return NULL;
+
+	buffer->base.refcount = 1;
+	buffer->base.alignment = 0;
+	buffer->base.usage = 0;
+	buffer->base.size = bytes;
+
+	buffer->bo = drm_intel_bo_alloc(dev->pools.gem,
+	                                "gallium3d_user_buffer",
+	                                bytes, 0);
+
+	if (!buffer->bo)
+		goto err;
+
+	ret = drm_intel_bo_subdata(buffer->bo,
+	                           0, bytes, ptr);
+
+	if (ret)
+		goto err;
+
+	return &buffer->base;
+
+err:
+	free(buffer);
+	return NULL;
+}
+
+struct pipe_buffer *
+intel_be_buffer_from_handle(struct pipe_winsys *winsys,
+                            const char* name, unsigned handle)
+{
+	struct intel_be_device *dev = intel_be_device(winsys);
+	struct intel_be_buffer *buffer = CALLOC_STRUCT(intel_be_buffer);
+
+	if (!buffer)
+		return NULL;
+
+	buffer->bo = drm_intel_bo_gem_create_from_name(dev->pools.gem, name, handle);
+
+	if (!buffer->bo)
+		goto err;
+
+	buffer->base.refcount = 1;
+	buffer->base.alignment = buffer->bo->align;
+	buffer->base.usage = PIPE_BUFFER_USAGE_GPU_READ |
+	                     PIPE_BUFFER_USAGE_GPU_WRITE |
+	                     PIPE_BUFFER_USAGE_CPU_READ |
+	                     PIPE_BUFFER_USAGE_CPU_WRITE;
+	buffer->base.size = buffer->bo->size;
+
+	return &buffer->base;
+
+err:
+	free(buffer);
+	return NULL;
+}
+
+unsigned
+intel_be_handle_from_buffer(struct pipe_winsys *winsys,
+                            struct pipe_buffer *buf)
+{
+	drm_intel_bo *bo = intel_bo(buf);
+	return bo->handle;
+}
+
+/*
+ * Fence
+ */
+
+static void
+intel_be_fence_refunref(struct pipe_winsys *sws,
+			 struct pipe_fence_handle **ptr,
+			 struct pipe_fence_handle *fence)
+{
+	struct intel_be_fence **p = (struct intel_be_fence **)ptr;
+	struct intel_be_fence *f = (struct intel_be_fence *)fence;
+
+	assert(p);
+
+	if (f)
+		intel_be_fence_reference(f);
+
+	if (*p)
+		intel_be_fence_unreference(*p);
+
+	*p = f;
+}
+
+static int
+intel_be_fence_signalled(struct pipe_winsys *sws,
+			 struct pipe_fence_handle *fence,
+			 unsigned flag)
+{
+	assert(0);
+
+	return 0;
+}
+
+static int
+intel_be_fence_finish(struct pipe_winsys *sws,
+		      struct pipe_fence_handle *fence,
+		      unsigned flag)
+{
+	struct intel_be_fence *f = (struct intel_be_fence *)fence;
+
+	/* fence already expired */
+	if (!f->bo)
+		return 0;
+
+	drm_intel_bo_wait_rendering(f->bo);
+	drm_intel_bo_unreference(f->bo);
+	f->bo = NULL;
+
+	return 0;
+}
+
+/*
+ * Misc functions
+ */
+
+boolean
+intel_be_init_device(struct intel_be_device *dev, int fd, unsigned id)
+{
+	dev->fd = fd;
+	dev->max_batch_size = 16 * 4096;
+	dev->max_vertex_size = 128 * 4096;
+
+	dev->base.buffer_create = intel_be_buffer_create;
+	dev->base.user_buffer_create = intel_be_user_buffer_create;
+	dev->base.buffer_map = intel_be_buffer_map;
+	dev->base.buffer_unmap = intel_be_buffer_unmap;
+	dev->base.buffer_destroy = intel_be_buffer_destroy;
+
+	/* Not used anymore */
+	dev->base.surface_alloc = NULL;
+	dev->base.surface_alloc_storage = NULL;
+	dev->base.surface_release = NULL;
+
+	dev->base.fence_reference = intel_be_fence_refunref;
+	dev->base.fence_signalled = intel_be_fence_signalled;
+	dev->base.fence_finish = intel_be_fence_finish;
+
+	dev->pools.gem = drm_intel_bufmgr_gem_init(dev->fd, dev->max_batch_size);
+
+	return true;
+}
+
+void
+intel_be_destroy_device(struct intel_be_device *dev)
+{
+	drm_intel_bufmgr_destroy(dev->pools.gem);
+}
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_device.h b/src/gallium/winsys/drm/intel/gem/intel_be_device.h
new file mode 100644
index 00000000000..96e94c47e71
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_device.h
@@ -0,0 +1,74 @@
+
+#ifndef INTEL_DRM_DEVICE_H
+#define INTEL_DRM_DEVICE_H
+
+#include "pipe/p_winsys.h"
+#include "pipe/p_context.h"
+
+#include "drm.h"
+#include "intel_bufmgr.h"
+
+/*
+ * Device
+ */
+
+struct intel_be_device
+{
+	struct pipe_winsys base;
+
+	int fd; /**< Drm file discriptor */
+
+	size_t max_batch_size;
+	size_t max_vertex_size;
+
+	struct {
+		drm_intel_bufmgr *gem;
+	} pools;
+};
+
+boolean
+intel_be_init_device(struct intel_be_device *device, int fd, unsigned id);
+
+void
+intel_be_destroy_device(struct intel_be_device *dev);
+
+/*
+ * Buffer
+ */
+
+struct intel_be_buffer {
+	struct pipe_buffer base;
+	drm_intel_bo *bo;
+};
+
+/**
+ * Create a be buffer from a drm bo handle.
+ *
+ * Takes a reference.
+ */
+struct pipe_buffer *
+intel_be_buffer_from_handle(struct pipe_winsys *winsys,
+                            const char* name, unsigned handle);
+
+/**
+ * Gets a handle from a buffer.
+ *
+ * If buffer is destroyed handle may become invalid.
+ */
+unsigned
+intel_be_handle_from_buffer(struct pipe_winsys *winsys,
+                            struct pipe_buffer *buffer);
+
+static INLINE struct intel_be_buffer *
+intel_be_buffer(struct pipe_buffer *buf)
+{
+	return (struct intel_be_buffer *)buf;
+}
+
+static INLINE drm_intel_bo *
+intel_bo(struct pipe_buffer *buf)
+{
+	return intel_be_buffer(buf)->bo;
+}
+
+#endif
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_fence.h b/src/gallium/winsys/drm/intel/gem/intel_be_fence.h
new file mode 100644
index 00000000000..0fe18f66f83
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_fence.h
@@ -0,0 +1,38 @@
+
+#ifndef INTEL_BE_FENCE_H
+#define INTEL_BE_FENCE_H
+
+#include "pipe/p_defines.h"
+
+#include "drm.h"
+#include "intel_bufmgr.h"
+
+/**
+ * Because gem does not have fence's we have to create our own fences.
+ *
+ * They work by keeping the batchbuffer around and checking if that has
+ * been idled. If bo is NULL fence has expired.
+ */
+struct intel_be_fence
+{
+	uint32_t refcount;
+	drm_intel_bo *bo;
+};
+
+static INLINE void
+intel_be_fence_reference(struct intel_be_fence *f)
+{
+	f->refcount++;
+}
+
+static INLINE void
+intel_be_fence_unreference(struct intel_be_fence *f)
+{
+	if (!--f->refcount) {
+		if (f->bo)
+			drm_intel_bo_unreference(f->bo);
+		free(f);
+	}
+}
+
+#endif
diff --git a/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.c b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.c
index 68951375066..5b3101fbba9 100644
--- a/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.c
@@ -69,7 +69,7 @@ nouveau_pipe_bo_create(struct pipe_winsys *pws, unsigned alignment,
 	nvbuf->base.usage = usage;
 	nvbuf->base.size = size;
 
-	flags = nouveau_flags_from_usage(nv, flags);
+	flags = nouveau_flags_from_usage(nv, usage);
 
 	if (nouveau_bo_new(dev, flags, alignment, size, &nvbuf->bo)) {
 		FREE(nvbuf);
@@ -121,14 +121,9 @@ nouveau_pipe_bo_map(struct pipe_winsys *pws, struct pipe_buffer *buf,
 	if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
 		map_flags |= NOUVEAU_BO_WR;
 
-	/* XXX: Technically incorrect. If the client maps a buffer for write-only
-	 * and leaves part of the buffer untouched it probably expects those parts
-	 * to remain intact. This is violated because we allocate a whole new buffer
-	 * and don't copy the previous buffer's contents, so this optimization is
-	 * only valid if the client intends to overwrite the whole buffer.
-	 */
-	if ((map_flags & NOUVEAU_BO_RDWR) == NOUVEAU_BO_WR &&
-	    !nouveau_bo_busy(nvbuf->bo, map_flags)) {
+	if (flags & PIPE_BUFFER_USAGE_DISCARD &&
+	    !(flags & PIPE_BUFFER_USAGE_CPU_READ) &&
+	    nouveau_bo_busy(nvbuf->bo, map_flags)) {
 		struct nouveau_pipe_winsys *nvpws = (struct nouveau_pipe_winsys *)pws;
 		struct nouveau_context *nv = nvpws->nv;
 		struct nouveau_device *dev = nv->nv_screen->device;
diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c
index 006978b1821..aacfe984d18 100644
--- a/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c
@@ -42,7 +42,7 @@ nouveau_context_create(const __GLcontextModes *glVis,
 
 	if (nouveau_context_init(&nv_screen->base, driContextPriv->hHWContext,
 	                         (drmLock *)&driScrnPriv->pSAREA->lock,
-	                         nv_share, &nv->base)) {
+	                         &nv_share->base, &nv->base)) {
 		return GL_FALSE;
 	}
 
diff --git a/src/glx/x11/drisw_glx.c b/src/glx/x11/drisw_glx.c
index fee45952b38..35bbd9151ca 100644
--- a/src/glx/x11/drisw_glx.c
+++ b/src/glx/x11/drisw_glx.c
@@ -113,7 +113,7 @@ swrastGetDrawableInfo(__DRIdrawable * draw,
                       int *x, int *y, int *w, int *h, void *loaderPrivate)
 {
    __GLXDRIdrawablePrivate *pdp = loaderPrivate;
-   __GLXDRIdrawable *pdraw = &(pdp->base);;
+   __GLXDRIdrawable *pdraw = &(pdp->base);
    Display *dpy = pdraw->psc->dpy;
    Drawable drawable;
 
@@ -141,7 +141,7 @@ swrastPutImage(__DRIdrawable * draw, int op,
                int x, int y, int w, int h, char *data, void *loaderPrivate)
 {
    __GLXDRIdrawablePrivate *pdp = loaderPrivate;
-   __GLXDRIdrawable *pdraw = &(pdp->base);;
+   __GLXDRIdrawable *pdraw = &(pdp->base);
    Display *dpy = pdraw->psc->dpy;
    Drawable drawable;
    XImage *ximage;
@@ -176,7 +176,7 @@ swrastGetImage(__DRIdrawable * draw,
                int x, int y, int w, int h, char *data, void *loaderPrivate)
 {
    __GLXDRIdrawablePrivate *pdp = loaderPrivate;
-   __GLXDRIdrawable *pdraw = &(pdp->base);;
+   __GLXDRIdrawable *pdraw = &(pdp->base);
    Display *dpy = pdraw->psc->dpy;
    Drawable drawable;
    XImage *ximage;
diff --git a/src/glx/x11/glx_pbuffer.c b/src/glx/x11/glx_pbuffer.c
index c63d53439d5..a602cd28817 100644
--- a/src/glx/x11/glx_pbuffer.c
+++ b/src/glx/x11/glx_pbuffer.c
@@ -220,14 +220,14 @@ GetDrawableAttribute(Display * dpy, GLXDrawable drawable,
    unsigned int length;
    unsigned int i;
    unsigned int num_attributes;
+   GLboolean use_glx_1_3;
 
    if ((dpy == NULL) || (drawable == 0)) {
       return 0;
    }
 
    priv = __glXInitialize(dpy);
-   GLboolean use_glx_1_3 = ((priv->majorVersion > 1)
-                            || (priv->minorVersion >= 3));
+   use_glx_1_3 = ((priv->majorVersion > 1) || (priv->minorVersion >= 3));
 
    *value = 0;
 
diff --git a/src/glx/x11/glxcurrent.c b/src/glx/x11/glxcurrent.c
index f3eb045d9f9..5af46cf0ba7 100644
--- a/src/glx/x11/glxcurrent.c
+++ b/src/glx/x11/glxcurrent.c
@@ -365,7 +365,7 @@ MakeContextCurrent(Display * dpy, GLXDrawable draw,
    const CARD8 oldOpcode = ((gc == oldGC) || (oldGC == &dummyContext))
       ? opcode : __glXSetupForCommand(oldGC->currentDpy);
    Bool bindReturnValue;
-
+   __GLXattribute *state;
 
    if (!opcode || !oldOpcode) {
       return GL_FALSE;
@@ -489,8 +489,7 @@ MakeContextCurrent(Display * dpy, GLXDrawable draw,
             } while (0);
 #endif
 
-            __GLXattribute *state =
-               (__GLXattribute *) (gc->client_state_private);
+            state = (__GLXattribute *) (gc->client_state_private);
 
             gc->currentContextTag = reply.contextTag;
             if (state->array_state == NULL) {
diff --git a/src/mesa/drivers/allegro/amesa.c b/src/mesa/drivers/allegro/amesa.c
index ade62518489..a9d8f62f92d 100644
--- a/src/mesa/drivers/allegro/amesa.c
+++ b/src/mesa/drivers/allegro/amesa.c
@@ -26,7 +26,7 @@
 #include "main/imports.h"
 #include "main/matrix.h"
 #include "main/mtypes.h"
-#include "GL/amesa.h"
+#include "amesa.h"
 
 
 struct amesa_visual
diff --git a/src/mesa/drivers/allegro/amesa.h b/src/mesa/drivers/allegro/amesa.h
new file mode 100644
index 00000000000..852d34cf4f5
--- /dev/null
+++ b/src/mesa/drivers/allegro/amesa.h
@@ -0,0 +1,65 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.3
+ * 
+ * Copyright (C) 1999-2000  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/* Allegro (DJGPP) driver by Bernhard Tschirren ([email protected]) */
+
+
+#ifndef AMESA_H
+#define AMESA_H
+
+
+#define AMESA_MAJOR_VERSION 3
+#define AMESA_MINOR_VERSION 3
+
+
+typedef struct amesa_visual  *AMesaVisual;
+typedef struct amesa_buffer  *AMesaBuffer;
+typedef struct amesa_context *AMesaContext;
+
+
+extern AMesaVisual AMesaCreateVisual(GLboolean dbFlag, GLint depth,
+                                     GLint depthSize,
+                                     GLint stencilSize,
+                                     GLint accumSize);
+
+extern void AMesaDestroyVisual(AMesaVisual visual);
+
+extern AMesaBuffer AMesaCreateBuffer(AMesaVisual visual,
+                                     GLint width, GLint height);
+
+extern void AMesaDestroyBuffer(AMesaBuffer buffer);
+
+
+extern AMesaContext AMesaCreateContext(AMesaVisual visual,
+                                       AMesaContext sharelist);
+
+extern void AMesaDestroyContext(AMesaContext context);
+
+extern GLboolean AMesaMakeCurrent(AMesaContext context, AMesaBuffer buffer);
+
+extern void AMesaSwapBuffers(AMesaBuffer buffer);
+
+
+#endif /* AMESA_H */
diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c
index a415e378fff..d7a2bd95ee2 100644
--- a/src/mesa/drivers/dri/i965/brw_context.c
+++ b/src/mesa/drivers/dri/i965/brw_context.c
@@ -143,6 +143,9 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
    ctx->Const.MaxCubeTextureLevels = 12;
    ctx->Const.MaxTextureRectSize = (1<<11);
    
+   /* if conformance mode is set, swrast can handle any size AA point */
+   ctx->Const.MaxPointSizeAA = 255.0;
+
 /*    ctx->Const.MaxNativeVertexProgramTemps = 32; */
 
    brw_init_attribs( brw );
diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c
index a18dee85e80..0c86911044b 100644
--- a/src/mesa/drivers/dri/i965/brw_program.c
+++ b/src/mesa/drivers/dri/i965/brw_program.c
@@ -111,9 +111,15 @@ static void brwProgramStringNotify( GLcontext *ctx,
 				    struct gl_program *prog )
 {
    if (target == GL_FRAGMENT_PROGRAM_ARB) {
+      struct gl_fragment_program *fprog = (struct gl_fragment_program *) prog;
       struct brw_context *brw = brw_context(ctx);
       struct brw_fragment_program *p = (struct brw_fragment_program *)prog;
       struct brw_fragment_program *fp = (struct brw_fragment_program *)brw->fragment_program;
+      if (fprog->FogOption) {
+         _mesa_append_fog_code(ctx, fprog);
+         fprog->FogOption = GL_NONE;
+      }
+
       if (p == fp)
 	 brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
       p->id = brw->program_id++;      
diff --git a/src/mesa/drivers/dri/i965/brw_sf.c b/src/mesa/drivers/dri/i965/brw_sf.c
index 9dce6cd8e6c..1a11d546210 100644
--- a/src/mesa/drivers/dri/i965/brw_sf.c
+++ b/src/mesa/drivers/dri/i965/brw_sf.c
@@ -73,10 +73,12 @@ static void compile_sf_prog( struct brw_context *brw,
 	 c.attr_to_idx[i] = idx;
 	 c.idx_to_attr[idx] = i;
 	 if (i >= VERT_RESULT_TEX0 && i <= VERT_RESULT_TEX7) {
-		 c.point_attrs[i].CoordReplace = 
-			brw->attribs.Point->CoordReplace[i - VERT_RESULT_TEX0];
-	 } else
-		 c.point_attrs[i].CoordReplace = GL_FALSE;
+            c.point_attrs[i].CoordReplace = 
+               brw->attribs.Point->CoordReplace[i - VERT_RESULT_TEX0];
+	 }
+         else {
+            c.point_attrs[i].CoordReplace = GL_FALSE;
+         }
 	 idx++;
       }
    
@@ -106,7 +108,6 @@ static void compile_sf_prog( struct brw_context *brw,
       assert(0);
       return;
    }
-	 
 
    /* get the program
     */
diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c
index 506126fcfb0..741a7c53c47 100644
--- a/src/mesa/drivers/dri/i965/brw_sf_state.c
+++ b/src/mesa/drivers/dri/i965/brw_sf_state.c
@@ -229,7 +229,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
    /* XXX clamp max depends on AA vs. non-AA */
 
    sf.sf7.sprite_point = key->point_sprite;
-   sf.sf7.point_size = CLAMP(nearbyint(key->point_size), 1, 255) * (1<<3);
+   sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
    sf.sf7.use_point_size_state = !key->point_attenuated;
    sf.sf7.aa_line_distance_mode = 0;
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c
index f4583877aa5..1a00b698256 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_fp.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c
@@ -811,57 +811,6 @@ static void precalc_txp( struct brw_wm_compile *c,
 
 
 
-
-
-/***********************************************************************
- * Add instructions to perform fog blending
- */
-
-static void fog_blend( struct brw_wm_compile *c,
-			     struct prog_src_register fog_factor )
-{
-   struct prog_dst_register outcolor = dst_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLR);
-   struct prog_src_register fogcolor = search_or_add_param5( c, STATE_FOG_COLOR, 0,0,0,0 );
-
-   /* color.xyz = LRP fog_factor.xxxx, output_color, fog_color */
-   
-   emit_op(c, 
-	   OPCODE_LRP,
-	   dst_mask(outcolor, WRITEMASK_XYZ),
-	   0, 0, 0,
-	   fog_factor,
-	   src_reg_from_dst(outcolor),
-	   fogcolor);
-}
-
-
-
-/* This one is simple - just take the interpolated fog coordinate and
- * use it as the fog blend factor.
- */
-static void fog_interpolated( struct brw_wm_compile *c )
-{
-   struct prog_src_register fogc = src_reg(PROGRAM_INPUT, FRAG_ATTRIB_FOGC);
-   
-   if (!(c->fp_interp_emitted & (1<<FRAG_ATTRIB_FOGC))) 
-      emit_interp(c, FRAG_ATTRIB_FOGC);
-
-   fog_blend( c, src_swizzle1(fogc, GET_SWZ(fogc.Swizzle,X)));
-}
-
-static void emit_fog( struct brw_wm_compile *c ) 
-{
-   if (!c->fp->program.FogOption)
-      return;
-
-   if (1) 
-      fog_interpolated( c );
-   else {
-      /* TODO: per-pixel fog */
-      assert(0);
-   }
-}
-
 static void emit_fb_write( struct brw_wm_compile *c )
 {
    struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH);
@@ -1059,7 +1008,6 @@ void brw_wm_pass_fp( struct brw_wm_compile *c )
          emit_ddy(c, inst);
 	break;
       case OPCODE_END:
-	 emit_fog(c);
 	 emit_fb_write(c);
 	 break;
       case OPCODE_PRINT:
diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
index 942ebe1ed4e..8fd776ac393 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
@@ -892,15 +892,19 @@ static void emit_lrp(struct brw_wm_compile *c,
     }
 }
 
+/**
+ * For GLSL shaders, this KIL will be unconditional.
+ * It may be contained inside an IF/ENDIF structure of course.
+ */
 static void emit_kil(struct brw_wm_compile *c)
 {
-	struct brw_compile *p = &c->func;
-	struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
-	brw_push_insn_state(p);
-	brw_set_mask_control(p, BRW_MASK_DISABLE);
-	brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK
-	brw_AND(p, depth, c->emit_mask_reg, depth);
-	brw_pop_insn_state(p);
+    struct brw_compile *p = &c->func;
+    struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
+    brw_push_insn_state(p);
+    brw_set_mask_control(p, BRW_MASK_DISABLE);
+    brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK
+    brw_AND(p, depth, c->emit_mask_reg, depth);
+    brw_pop_insn_state(p);
 }
 
 static void emit_mad(struct brw_wm_compile *c,
diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
index f12ef47a7d7..8c9cb789453 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c
@@ -244,7 +244,7 @@ brw_wm_sampler_populate_key(struct brw_context *brw,
 	 entry->minfilter = texObj->MinFilter;
 	 entry->magfilter = texObj->MagFilter;
 	 entry->comparemode = texObj->CompareMode;
-    entry->comparefunc = texObj->CompareFunc;
+         entry->comparefunc = texObj->CompareFunc;
 
 	 dri_bo_unreference(brw->wm.sdc_bo[unit]);
 	 if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index 63e14cc3900..06e71e6d694 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -192,21 +192,27 @@ brw_create_texture_surface( struct brw_context *brw,
    if (key->bo) 
       surf.ss0.surface_format = translate_tex_format(key->format, key->depthmode);
    else {
-     switch(key->depth) {
-     case 32: surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; break;
-     default:
-     case 24: surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8X8_UNORM; break;
-     case 16: surf.ss0.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM; break;
-     }
+      switch (key->depth) {
+      case 32:
+         surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
+         break;
+      default:
+      case 24:
+         surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8X8_UNORM;
+         break;
+      case 16:
+         surf.ss0.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM;
+         break;
+      }
    }
 
    /* This is ok for all textures with channel width 8bit or less:
     */
 /*    surf.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
    if (key->bo)
-     surf.ss1.base_addr = key->bo->offset; /* reloc */
+      surf.ss1.base_addr = key->bo->offset; /* reloc */
    else
-     surf.ss1.base_addr = key->offset;
+      surf.ss1.base_addr = key->offset;
 
    surf.ss2.mip_count = key->last_level - key->first_level;
    surf.ss2.width = key->width - 1;
diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c
index 44b276a123e..7b7f7d8c145 100644
--- a/src/mesa/drivers/dri/intel/intel_context.c
+++ b/src/mesa/drivers/dri/intel/intel_context.c
@@ -95,7 +95,7 @@ int INTEL_DEBUG = (0);
 
 #include "extension_helper.h"
 
-#define DRIVER_DATE                     "20080716"
+#define DRIVER_DATE                     "20090114"
 #define DRIVER_DATE_GEM                 "GEM " DRIVER_DATE
 
 static const GLubyte *
diff --git a/src/mesa/drivers/dri/intel/intel_depthstencil.c b/src/mesa/drivers/dri/intel/intel_depthstencil.c
index c2b4d7728b4..354b3bf0d70 100644
--- a/src/mesa/drivers/dri/intel/intel_depthstencil.c
+++ b/src/mesa/drivers/dri/intel/intel_depthstencil.c
@@ -110,7 +110,10 @@ intel_unpair_depth_stencil(GLcontext *ctx, struct intel_renderbuffer *irb)
 	 ASSERT(stencilIrb->PairedDepth == rb->Name);
 	 intel_renderbuffer_map(intel, rb);
 	 intel_renderbuffer_map(intel, stencilRb);
+#if 0
+         /* disable for now */
 	 _mesa_extract_stencil(ctx, rb, stencilRb);
+#endif
 	 intel_renderbuffer_unmap(intel, stencilRb);
 	 intel_renderbuffer_unmap(intel, rb);
          stencilIrb->PairedDepth = 0;
@@ -132,7 +135,10 @@ intel_unpair_depth_stencil(GLcontext *ctx, struct intel_renderbuffer *irb)
 	 ASSERT(depthIrb->PairedStencil == rb->Name);
 	 intel_renderbuffer_map(intel, rb);
 	 intel_renderbuffer_map(intel, depthRb);
+#if 0
+         /* disable for now */
 	 _mesa_extract_stencil(ctx, depthRb, rb);
+#endif
 	 intel_renderbuffer_unmap(intel, depthRb);
 	 intel_renderbuffer_unmap(intel, rb);
          depthIrb->PairedStencil = 0;
@@ -177,8 +183,11 @@ intel_validate_paired_depth_stencil(GLcontext * ctx,
       }
       else {
          /* Separate depth/stencil buffers, need to interleave now */
-         ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_COMPONENT);
-         ASSERT(stencilRb->Base._BaseFormat == GL_STENCIL_INDEX);
+         ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_COMPONENT ||
+                depthRb->Base._BaseFormat == GL_DEPTH_STENCIL);
+         ASSERT(stencilRb->Base._BaseFormat == GL_STENCIL_INDEX ||
+                stencilRb->Base._BaseFormat == GL_DEPTH_STENCIL);
+
          /* may need to interleave depth/stencil now */
          if (depthRb->PairedStencil == stencilRb->Base.Name) {
             /* OK, the depth and stencil buffers are already interleaved */
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
index 7453b96dc5d..54f2fa5287b 100644
--- a/src/mesa/drivers/dri/intel/intel_fbo.c
+++ b/src/mesa/drivers/dri/intel/intel_fbo.c
@@ -248,11 +248,18 @@ intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
       cpp = 4;
       break;
    case GL_DEPTH_COMPONENT16:
+#if 0
       rb->_ActualFormat = GL_DEPTH_COMPONENT16;
       rb->DataType = GL_UNSIGNED_SHORT;
       rb->DepthBits = 16;
       cpp = 2;
       break;
+#else
+      /* fall-through.
+       * 16bpp depth renderbuffer can't be paired with a stencil buffer so
+       * always used combined depth/stencil format.
+       */
+#endif
    case GL_DEPTH_COMPONENT:
    case GL_DEPTH_COMPONENT24:
    case GL_DEPTH_COMPONENT32:
@@ -529,20 +536,25 @@ intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb,
    if (texImage->TexFormat == &_mesa_texformat_argb8888) {
       irb->Base._ActualFormat = GL_RGBA8;
       irb->Base._BaseFormat = GL_RGBA;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
       DBG("Render to RGBA8 texture OK\n");
    }
    else if (texImage->TexFormat == &_mesa_texformat_rgb565) {
       irb->Base._ActualFormat = GL_RGB5;
       irb->Base._BaseFormat = GL_RGB;
+      irb->Base.DataType = GL_UNSIGNED_SHORT;
       DBG("Render to RGB5 texture OK\n");
    }
    else if (texImage->TexFormat == &_mesa_texformat_z16) {
       irb->Base._ActualFormat = GL_DEPTH_COMPONENT16;
       irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      irb->Base.DataType = GL_UNSIGNED_SHORT;
       DBG("Render to DEPTH16 texture OK\n");
-   } else if (texImage->TexFormat == &_mesa_texformat_s8_z24) {
+   }
+   else if (texImage->TexFormat == &_mesa_texformat_s8_z24) {
       irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
       irb->Base._BaseFormat = GL_DEPTH_STENCIL_EXT;
+      irb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT;
       DBG("Render to DEPTH_STENCIL texture OK\n");
    }
    else {
@@ -554,7 +566,6 @@ intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb,
    irb->Base.InternalFormat = irb->Base._ActualFormat;
    irb->Base.Width = texImage->Width;
    irb->Base.Height = texImage->Height;
-   irb->Base.DataType = GL_UNSIGNED_BYTE;       /* FBO XXX fix */
    irb->Base.RedBits = texImage->TexFormat->RedBits;
    irb->Base.GreenBits = texImage->TexFormat->GreenBits;
    irb->Base.BlueBits = texImage->TexFormat->BlueBits;
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
index 1d7f15f10a5..3a01f63dc72 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
@@ -204,6 +204,14 @@ do_blit_bitmap( GLcontext *ctx,
    /* Update draw buffer bounds */
    _mesa_update_state(ctx);
 
+   if (ctx->Depth.Test) {
+      /* The blit path produces incorrect results when depth testing is on.
+       * It seems the blit Z coord is always 1.0 (the far plane) so fragments
+       * will likely be obscured by other, closer geometry.
+       */
+      return GL_FALSE;
+   }
+
    if (!dst)
        return GL_FALSE;
 
@@ -357,6 +365,7 @@ intel_texture_bitmap(GLcontext * ctx,
    GLubyte *unpacked_bitmap;
    GLubyte *a8_bitmap;
    int x, y;
+   GLfloat dst_z;
 
    /* We need a fragment program for the KIL effect */
    if (!ctx->Extensions.ARB_fragment_program ||
@@ -456,21 +465,24 @@ intel_texture_bitmap(GLcontext * ctx,
    intel_meta_set_passthrough_vertex_program(intel);
    intel_meta_set_passthrough_transform(intel);
 
+   /* convert rasterpos Z from [0,1] to NDC coord in [-1,1] */
+   dst_z = -1.0 + 2.0 * ctx->Current.RasterPos[2];
+
    vertices[0][0] = dst_x;
    vertices[0][1] = dst_y;
-   vertices[0][2] = ctx->Current.RasterPos[2];
+   vertices[0][2] = dst_z;
    vertices[0][3] = 1.0;
    vertices[1][0] = dst_x + width;
    vertices[1][1] = dst_y;
-   vertices[1][2] = ctx->Current.RasterPos[2];
+   vertices[1][2] = dst_z;
    vertices[1][3] = 1.0;
    vertices[2][0] = dst_x + width;
    vertices[2][1] = dst_y + height;
-   vertices[2][2] = ctx->Current.RasterPos[2];
+   vertices[2][2] = dst_z;
    vertices[2][3] = 1.0;
    vertices[3][0] = dst_x;
    vertices[3][1] = dst_y + height;
-   vertices[3][2] = ctx->Current.RasterPos[2];
+   vertices[3][2] = dst_z;
    vertices[3][3] = 1.0;
 
    texcoords[0][0] = 0.0;
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_copy.c b/src/mesa/drivers/dri/intel/intel_pixel_copy.c
index 447c6494e79..7c7aa6097c8 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_copy.c
@@ -119,6 +119,12 @@ do_texture_copypixels(GLcontext * ctx,
    if (!src || !dst || type != GL_COLOR)
       return GL_FALSE;
 
+   if (ctx->_ImageTransferState) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         fprintf(stderr, "%s: check_color failed\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
    /* Can't handle overlapping regions.  Don't have sufficient control
     * over rasterization to pull it off in-place.  Punt on these for
     * now.
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_draw.c b/src/mesa/drivers/dri/intel/intel_pixel_draw.c
index 2af839b8031..0e83afa6451 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_draw.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_draw.c
@@ -71,6 +71,7 @@ intel_texture_drawpixels(GLcontext * ctx,
    GLuint texname;
    GLfloat vertices[4][4];
    GLfloat texcoords[4][2];
+   GLfloat z;
 
    /* We're going to mess with texturing with no regard to existing texture
     * state, so if there is some set up we have to bail.
@@ -140,6 +141,9 @@ intel_texture_drawpixels(GLcontext * ctx,
 
    intel_meta_set_passthrough_transform(intel);
 
+   /* convert rasterpos Z from [0,1] to NDC coord in [-1,1] */
+   z = -1.0 + 2.0 * ctx->Current.RasterPos[2];
+
    /* Create the vertex buffer based on the current raster pos.  The x and y
     * we're handed are ctx->Current.RasterPos[0,1] rounded to integers.
     * We also apply the depth.  However, the W component is already multiplied
@@ -147,19 +151,19 @@ intel_texture_drawpixels(GLcontext * ctx,
     */
    vertices[0][0] = x;
    vertices[0][1] = y;
-   vertices[0][2] = ctx->Current.RasterPos[2];
+   vertices[0][2] = z;
    vertices[0][3] = 1.0;
    vertices[1][0] = x + width * ctx->Pixel.ZoomX;
    vertices[1][1] = y;
-   vertices[1][2] = ctx->Current.RasterPos[2];
+   vertices[1][2] = z;
    vertices[1][3] = 1.0;
    vertices[2][0] = x + width * ctx->Pixel.ZoomX;
    vertices[2][1] = y + height * ctx->Pixel.ZoomY;
-   vertices[2][2] = ctx->Current.RasterPos[2];
+   vertices[2][2] = z;
    vertices[2][3] = 1.0;
    vertices[3][0] = x;
    vertices[3][1] = y + height * ctx->Pixel.ZoomY;
-   vertices[3][2] = ctx->Current.RasterPos[2];
+   vertices[3][2] = z;
    vertices[3][3] = 1.0;
 
    texcoords[0][0] = 0.0;
diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c
index 8f4e681ffea..d9315043e6e 100644
--- a/src/mesa/drivers/dri/intel/intel_span.c
+++ b/src/mesa/drivers/dri/intel/intel_span.c
@@ -633,7 +633,7 @@ intelSpanRenderStart(GLcontext * ctx)
    intelFlush(&intel->ctx);
    LOCK_HARDWARE(intel);
 
-   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
+   for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
       if (ctx->Texture.Unit[i]._ReallyEnabled) {
          struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
          intel_tex_map_images(intel, intel_texture_object(texObj));
@@ -655,7 +655,7 @@ intelSpanRenderFinish(GLcontext * ctx)
 
    _swrast_flush(ctx);
 
-   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
+   for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
       if (ctx->Texture.Unit[i]._ReallyEnabled) {
          struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
          intel_tex_unmap_images(intel, intel_texture_object(texObj));
diff --git a/src/mesa/drivers/dri/intel/intel_tex_format.c b/src/mesa/drivers/dri/intel/intel_tex_format.c
index 2be060dd3e3..5e418ac4463 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_format.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_format.c
@@ -134,8 +134,14 @@ intelChooseTextureFormat(GLcontext * ctx, GLint internalFormat,
    case GL_DEPTH_COMPONENT16:
    case GL_DEPTH_COMPONENT24:
    case GL_DEPTH_COMPONENT32:
+#if 0
       return &_mesa_texformat_z16;
-
+#else
+      /* fall-through.
+       * 16bpp depth texture can't be paired with a stencil buffer so
+       * always used combined depth/stencil format.
+       */
+#endif
    case GL_DEPTH_STENCIL_EXT:
    case GL_DEPTH24_STENCIL8_EXT:
       return &_mesa_texformat_s8_z24;
@@ -158,7 +164,7 @@ intelChooseTextureFormat(GLcontext * ctx, GLint internalFormat,
    case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT:
    case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT:
    case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
-     return &_mesa_texformat_srgb_dxt1;
+      return &_mesa_texformat_srgb_dxt1;
 #endif
 
    default:
diff --git a/src/mesa/drivers/dri/mach64/mach64_context.h b/src/mesa/drivers/dri/mach64/mach64_context.h
index 55e0618ff80..854751626d0 100644
--- a/src/mesa/drivers/dri/mach64/mach64_context.h
+++ b/src/mesa/drivers/dri/mach64/mach64_context.h
@@ -294,7 +294,13 @@ extern GLboolean mach64UnbindContext( __DRIcontextPrivate *driContextPriv );
 #define LE32_OUT( x, y )	do { *(GLuint *)(x) = (y); } while (0)
 #define LE32_OUT_FLOAT( x, y )	do { *(GLfloat *)(x) = (y); } while (0)
 #else
+#ifndef __OpenBSD__
 #include <byteswap.h>
+#else
+#include <machine/endian.h>
+#define bswap_32 bswap32
+#endif
+
 #define LE32_IN( x )		bswap_32( *(GLuint *)(x) )
 #define LE32_IN_FLOAT( x )						\
 ({									\
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index 6a5c3633a23..a63dbac3436 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1675,6 +1675,13 @@ static void r300SetupRSUnit(GLcontext * ctx)
 		rs_col_count += count;
 	}
 
+	if (InputsRead & FRAG_BIT_FOGC) {
+		/* XXX FIX THIS
+		 * Just turn off the bit for now.
+		 * Need to do something similar to the color/texcoord inputs.
+		 */
+		InputsRead &= ~FRAG_BIT_FOGC;
+	}
 
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
 		int swiz;
diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.c b/src/mesa/drivers/dri/r300/radeon_program_pair.c
index 5ad50d2863a..58bc0d5843f 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_pair.c
+++ b/src/mesa/drivers/dri/r300/radeon_program_pair.c
@@ -473,6 +473,11 @@ static void allocate_input_registers(struct pair_state *s)
 		alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_COL1, hwindex++);
 	InputsRead &= ~FRAG_BIT_COL1;
 
+	/* Fog coordinate */
+	if (InputsRead & FRAG_BIT_FOGC)
+		alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_FOGC, hwindex++);
+	InputsRead &= ~FRAG_BIT_FOGC;
+
 	/* Anything else */
 	if (InputsRead)
 		error("Don't know how to handle inputs 0x%x\n", InputsRead);
diff --git a/src/mesa/drivers/dri/swrast/swrast.c b/src/mesa/drivers/dri/swrast/swrast.c
index 49f1b8b9ee7..15b57244dcd 100644
--- a/src/mesa/drivers/dri/swrast/swrast.c
+++ b/src/mesa/drivers/dri/swrast/swrast.c
@@ -66,6 +66,7 @@
 #define need_GL_ARB_vertex_program
 #define need_GL_APPLE_vertex_array_object
 #define need_GL_ATI_fragment_shader
+#define need_GL_ATI_separate_stencil
 #define need_GL_EXT_depth_bounds_test
 #define need_GL_EXT_framebuffer_object
 #define need_GL_EXT_framebuffer_blit
@@ -96,6 +97,7 @@ const struct dri_extension card_extensions[] =
     { "GL_ARB_vertex_program",		GL_ARB_vertex_program_functions },
     { "GL_APPLE_vertex_array_object",	GL_APPLE_vertex_array_object_functions },
     { "GL_ATI_fragment_shader",		GL_ATI_fragment_shader_functions },
+    { "GL_ATI_separate_stencil",	GL_ATI_separate_stencil_functions },
     { "GL_EXT_depth_bounds_test",	GL_EXT_depth_bounds_test_functions },
     { "GL_EXT_framebuffer_object",	GL_EXT_framebuffer_object_functions },
     { "GL_EXT_framebuffer_blit",	GL_EXT_framebuffer_blit_functions },
diff --git a/src/mesa/drivers/windows/gdi/mesa.def b/src/mesa/drivers/windows/gdi/mesa.def
index 3f2d644e86e..b386e34aadf 100644
--- a/src/mesa/drivers/windows/gdi/mesa.def
+++ b/src/mesa/drivers/windows/gdi/mesa.def
@@ -867,6 +867,7 @@ EXPORTS
 	_glapi_get_proc_address
 	_mesa_add_soft_renderbuffers
 	_mesa_add_renderbuffer
+	_mesa_begin_query
 	_mesa_buffer_data
 	_mesa_buffer_get_subdata
 	_mesa_buffer_map
@@ -881,6 +882,7 @@ EXPORTS
 	_mesa_delete_array_object
 	_mesa_delete_buffer_object
 	_mesa_delete_program
+	_mesa_delete_query
 	_mesa_delete_texture_object
 	_mesa_destroy_framebuffer
 	_mesa_destroy_visual
@@ -890,6 +892,7 @@ EXPORTS
 	_mesa_enable_2_0_extensions
 	_mesa_enable_2_1_extensions
 	_mesa_enable_sw_extensions
+	_mesa_end_query
 	_mesa_error
 	_mesa_finish_render_texture
 	_mesa_framebuffer_renderbuffer
@@ -941,6 +944,7 @@ EXPORTS
 	_mesa_update_framebuffer_visual
 	_mesa_use_program
 	_mesa_Viewport
+	_mesa_wait_query
 	_swrast_Accum
 	_swrast_Bitmap
 	_swrast_BlitFramebuffer
@@ -973,3 +977,4 @@ EXPORTS
 	_tnl_InvalidateState
 	_tnl_run_pipeline
 	_tnl_program_string
+	_tnl_RasterPos
\ No newline at end of file
diff --git a/src/mesa/drivers/windows/gdi/wgl.c b/src/mesa/drivers/windows/gdi/wgl.c
index 9f0bb9122a9..8d8087067f5 100644
--- a/src/mesa/drivers/windows/gdi/wgl.c
+++ b/src/mesa/drivers/windows/gdi/wgl.c
@@ -601,8 +601,9 @@ WINGDIAPI BOOL GLAPIENTRY wglCopyContext(HGLRC hglrcSrc,
 WINGDIAPI HGLRC GLAPIENTRY wglCreateLayerContext(HDC hdc,
 						 int iLayerPlane)
 {
-    (void) hdc; (void) iLayerPlane;
     SetLastError(0);
+    if (iLayerPlane == 0)
+        return wglCreateContext( hdc );
     return(NULL);
 }
 
diff --git a/src/mesa/drivers/x11/fakeglx.c b/src/mesa/drivers/x11/fakeglx.c
index 827d39f9957..ea3585258d8 100644
--- a/src/mesa/drivers/x11/fakeglx.c
+++ b/src/mesa/drivers/x11/fakeglx.c
@@ -42,7 +42,6 @@
 
 #include "glxheader.h"
 #include "glxapi.h"
-#include "GL/xmesa.h"
 #include "main/context.h"
 #include "main/config.h"
 #include "main/macros.h"
diff --git a/src/mesa/drivers/x11/fxmesa.h b/src/mesa/drivers/x11/fxmesa.h
new file mode 100644
index 00000000000..f8e9661f9cc
--- /dev/null
+++ b/src/mesa/drivers/x11/fxmesa.h
@@ -0,0 +1,103 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  4.0
+ * Copyright (C) 1995-2001  Brian Paul
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Library General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU Library General Public
+ * License along with this library; if not, write to the Free
+ * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+
+/*
+ * FXMesa - 3Dfx Glide driver for Mesa.  Contributed by David Bucciarelli
+ *
+ * NOTE: This version requires Glide3 (http://sourceforge.net/projects/glide)
+ */
+
+
+#ifndef FXMESA_H
+#define FXMESA_H
+
+
+#include <glide.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+#define FXMESA_MAJOR_VERSION 6
+#define FXMESA_MINOR_VERSION 3
+
+
+/*
+ * Values for attribList parameter to fxMesaCreateContext():
+ */
+#define FXMESA_NONE		0	/* to terminate attribList */
+#define FXMESA_DOUBLEBUFFER	10
+#define FXMESA_ALPHA_SIZE	11      /* followed by an integer */
+#define FXMESA_DEPTH_SIZE	12      /* followed by an integer */
+#define FXMESA_STENCIL_SIZE	13      /* followed by an integer */
+#define FXMESA_ACCUM_SIZE	14      /* followed by an integer */
+#define FXMESA_COLORDEPTH	20      /* followed by an integer */
+#define FXMESA_SHARE_CONTEXT 990099	/* keep in sync with xmesa1.c! */
+
+
+
+typedef struct tfxMesaContext *fxMesaContext;
+
+
+#if defined (__BEOS__)
+#pragma export on
+#endif
+
+
+GLAPI fxMesaContext GLAPIENTRY fxMesaCreateContext(GLuint win, GrScreenResolution_t,
+						  GrScreenRefresh_t,
+						  const GLint attribList[]);
+
+GLAPI fxMesaContext GLAPIENTRY fxMesaCreateBestContext(GLuint win,
+						      GLint width, GLint height,
+						      const GLint attribList[]);
+GLAPI void GLAPIENTRY fxMesaDestroyContext(fxMesaContext ctx);
+
+GLAPI GLint GLAPIENTRY fxMesaSelectCurrentBoard(int n);
+
+GLAPI void GLAPIENTRY fxMesaMakeCurrent(fxMesaContext ctx);
+
+GLAPI fxMesaContext GLAPIENTRY fxMesaGetCurrentContext(void);
+
+GLAPI void GLAPIENTRY fxMesaSwapBuffers(void);
+
+GLAPI void GLAPIENTRY fxMesaSetNearFar(GLfloat nearVal, GLfloat farVal);
+
+GLAPI void GLAPIENTRY fxMesaUpdateScreenSize(fxMesaContext ctx);
+
+GLAPI void GLAPIENTRY fxCloseHardware(void);
+
+GLAPI void GLAPIENTRY fxGetScreenGeometry (GLint *w, GLint *h);
+
+
+#if defined (__BEOS__)
+#pragma export off
+#endif
+
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/src/mesa/drivers/x11/xm_api.c b/src/mesa/drivers/x11/xm_api.c
index c9009bad031..18aa8bcc096 100644
--- a/src/mesa/drivers/x11/xm_api.c
+++ b/src/mesa/drivers/x11/xm_api.c
@@ -63,7 +63,6 @@
 #endif
 
 #include "glxheader.h"
-#include "GL/xmesa.h"
 #include "xmesaP.h"
 #include "main/context.h"
 #include "main/extensions.h"
diff --git a/src/mesa/drivers/x11/xm_buffer.c b/src/mesa/drivers/x11/xm_buffer.c
index f104d44d051..7ad67bc34da 100644
--- a/src/mesa/drivers/x11/xm_buffer.c
+++ b/src/mesa/drivers/x11/xm_buffer.c
@@ -30,7 +30,6 @@
 
 
 #include "glxheader.h"
-#include "GL/xmesa.h"
 #include "xmesaP.h"
 #include "main/imports.h"
 #include "main/framebuffer.h"
diff --git a/src/mesa/drivers/x11/xmesa.h b/src/mesa/drivers/x11/xmesa.h
new file mode 100644
index 00000000000..98139af8336
--- /dev/null
+++ b/src/mesa/drivers/x11/xmesa.h
@@ -0,0 +1,424 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ * 
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ * 
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/*
+ * Mesa/X11 interface.  This header file serves as the documentation for
+ * the Mesa/X11 interface functions.
+ *
+ * Note: this interface isn't intended for user programs.  It's primarily
+ * just for implementing the pseudo-GLX interface.
+ */
+
+
+/* Sample Usage:
+
+In addition to the usual X calls to select a visual, create a colormap
+and create a window, you must do the following to use the X/Mesa interface:
+
+1. Call XMesaCreateVisual() to make an XMesaVisual from an XVisualInfo.
+
+2. Call XMesaCreateContext() to create an X/Mesa rendering context, given
+   the XMesaVisual.
+
+3. Call XMesaCreateWindowBuffer() to create an XMesaBuffer from an X window
+   and XMesaVisual.
+
+4. Call XMesaMakeCurrent() to bind the XMesaBuffer to an XMesaContext and
+   to make the context the current one.
+
+5. Make gl* calls to render your graphics.
+
+6. Use XMesaSwapBuffers() when double buffering to swap front/back buffers.
+
+7. Before the X window is destroyed, call XMesaDestroyBuffer().
+
+8. Before exiting, call XMesaDestroyVisual and XMesaDestroyContext.
+
+*/
+
+
+
+
+#ifndef XMESA_H
+#define XMESA_H
+
+#ifdef __VMS
+#include <GL/vms_x_fix.h>
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef XFree86Server
+#include "xmesa_xf86.h"
+#else
+#include <X11/Xlib.h>
+#include <X11/Xutil.h>
+#include "xmesa_x.h"
+#endif
+#include "GL/gl.h"
+
+#ifdef AMIWIN
+#include <pragmas/xlib_pragmas.h>
+extern struct Library *XLibBase;
+#endif
+
+
+#define XMESA_MAJOR_VERSION 6
+#define XMESA_MINOR_VERSION 3
+
+
+
+/*
+ * Values passed to XMesaGetString:
+ */
+#define XMESA_VERSION 1
+#define XMESA_EXTENSIONS 2
+
+
+/*
+ * Values passed to XMesaSetFXmode:
+ */
+#define XMESA_FX_WINDOW       1
+#define XMESA_FX_FULLSCREEN   2
+
+
+
+typedef struct xmesa_context *XMesaContext;
+
+typedef struct xmesa_visual *XMesaVisual;
+
+typedef struct xmesa_buffer *XMesaBuffer;
+
+
+
+/*
+ * Create a new X/Mesa visual.
+ * Input:  display - X11 display
+ *         visinfo - an XVisualInfo pointer
+ *         rgb_flag - GL_TRUE = RGB mode,
+ *                    GL_FALSE = color index mode
+ *         alpha_flag - alpha buffer requested?
+ *         db_flag - GL_TRUE = double-buffered,
+ *                   GL_FALSE = single buffered
+ *         stereo_flag - stereo visual?
+ *         ximage_flag - GL_TRUE = use an XImage for back buffer,
+ *                       GL_FALSE = use an off-screen pixmap for back buffer
+ *         depth_size - requested bits/depth values, or zero
+ *         stencil_size - requested bits/stencil values, or zero
+ *         accum_red_size - requested bits/red accum values, or zero
+ *         accum_green_size - requested bits/green accum values, or zero
+ *         accum_blue_size - requested bits/blue accum values, or zero
+ *         accum_alpha_size - requested bits/alpha accum values, or zero
+ *         num_samples - number of samples/pixel if multisampling, or zero
+ *         level - visual level, usually 0
+ *         visualCaveat - ala the GLX extension, usually GLX_NONE_EXT
+ * Return;  a new XMesaVisual or 0 if error.
+ */
+extern XMesaVisual XMesaCreateVisual( XMesaDisplay *display,
+                                      XMesaVisualInfo visinfo,
+                                      GLboolean rgb_flag,
+                                      GLboolean alpha_flag,
+                                      GLboolean db_flag,
+                                      GLboolean stereo_flag,
+                                      GLboolean ximage_flag,
+                                      GLint depth_size,
+                                      GLint stencil_size,
+                                      GLint accum_red_size,
+                                      GLint accum_green_size,
+                                      GLint accum_blue_size,
+                                      GLint accum_alpha_size,
+                                      GLint num_samples,
+                                      GLint level,
+                                      GLint visualCaveat );
+
+/*
+ * Destroy an XMesaVisual, but not the associated XVisualInfo.
+ */
+extern void XMesaDestroyVisual( XMesaVisual v );
+
+
+
+/*
+ * Create a new XMesaContext for rendering into an X11 window.
+ *
+ * Input:  visual - an XMesaVisual
+ *         share_list - another XMesaContext with which to share display
+ *                      lists or NULL if no sharing is wanted.
+ * Return:  an XMesaContext or NULL if error.
+ */
+extern XMesaContext XMesaCreateContext( XMesaVisual v,
+					XMesaContext share_list );
+
+
+/*
+ * Destroy a rendering context as returned by XMesaCreateContext()
+ */
+extern void XMesaDestroyContext( XMesaContext c );
+
+
+#ifdef XFree86Server
+/*
+ * These are the extra routines required for integration with XFree86.
+ * None of these routines should be user visible. -KEM
+ */
+extern GLboolean XMesaForceCurrent( XMesaContext c );
+
+extern GLboolean XMesaLoseCurrent( XMesaContext c );
+
+extern GLboolean XMesaCopyContext( XMesaContext src,
+				   XMesaContext dst,
+				   GLuint mask );
+#endif /* XFree86Server */
+
+
+/*
+ * Create an XMesaBuffer from an X window.
+ */
+extern XMesaBuffer XMesaCreateWindowBuffer( XMesaVisual v, XMesaWindow w );
+
+
+/*
+ * Create an XMesaBuffer from an X pixmap.
+ */
+extern XMesaBuffer XMesaCreatePixmapBuffer( XMesaVisual v,
+					    XMesaPixmap p,
+					    XMesaColormap cmap );
+
+
+/*
+ * Destroy an XMesaBuffer, but not the corresponding window or pixmap.
+ */
+extern void XMesaDestroyBuffer( XMesaBuffer b );
+
+
+/*
+ * Return the XMesaBuffer handle which corresponds to an X drawable, if any.
+ *
+ * New in Mesa 2.3.
+ */
+extern XMesaBuffer XMesaFindBuffer( XMesaDisplay *dpy,
+				    XMesaDrawable d );
+
+
+
+/*
+ * Bind a buffer to a context and make the context the current one.
+ */
+extern GLboolean XMesaMakeCurrent( XMesaContext c,
+				   XMesaBuffer b );
+
+
+/*
+ * Bind two buffers (read and draw) to a context and make the
+ * context the current one.
+ * New in Mesa 3.3
+ */
+extern GLboolean XMesaMakeCurrent2( XMesaContext c,
+                                    XMesaBuffer drawBuffer,
+                                    XMesaBuffer readBuffer );
+
+
+/*
+ * Unbind the current context from its buffer.
+ */
+extern GLboolean XMesaUnbindContext( XMesaContext c );
+
+
+/*
+ * Return a handle to the current context.
+ */
+extern XMesaContext XMesaGetCurrentContext( void );
+
+
+/*
+ * Return handle to the current (draw) buffer.
+ */
+extern XMesaBuffer XMesaGetCurrentBuffer( void );
+
+
+/*
+ * Return handle to the current read buffer.
+ * New in Mesa 3.3
+ */
+extern XMesaBuffer XMesaGetCurrentReadBuffer( void );
+
+
+/*
+ * Swap the front and back buffers for the given buffer.  No action is
+ * taken if the buffer is not double buffered.
+ */
+extern void XMesaSwapBuffers( XMesaBuffer b );
+
+
+/*
+ * Copy a sub-region of the back buffer to the front buffer.
+ *
+ * New in Mesa 2.6
+ */
+extern void XMesaCopySubBuffer( XMesaBuffer b,
+				int x,
+				int y,
+				int width,
+				int height );
+
+
+/*
+ * Return a pointer to the the Pixmap or XImage being used as the back
+ * color buffer of an XMesaBuffer.  This function is a way to get "under
+ * the hood" of X/Mesa so one can manipulate the back buffer directly.
+ * Input:  b - the XMesaBuffer
+ * Output:  pixmap - pointer to back buffer's Pixmap, or 0
+ *          ximage - pointer to back buffer's XImage, or NULL
+ * Return:  GL_TRUE = context is double buffered
+ *          GL_FALSE = context is single buffered
+ */
+extern GLboolean XMesaGetBackBuffer( XMesaBuffer b,
+				     XMesaPixmap *pixmap,
+				     XMesaImage **ximage );
+
+
+
+/*
+ * Return the depth buffer associated with an XMesaBuffer.
+ * Input:  b - the XMesa buffer handle
+ * Output:  width, height - size of buffer in pixels
+ *          bytesPerValue - bytes per depth value (2 or 4)
+ *          buffer - pointer to depth buffer values
+ * Return:  GL_TRUE or GL_FALSE to indicate success or failure.
+ *
+ * New in Mesa 2.4.
+ */
+extern GLboolean XMesaGetDepthBuffer( XMesaBuffer b,
+				      GLint *width,
+				      GLint *height,
+				      GLint *bytesPerValue,
+				      void **buffer );
+
+
+
+/*
+ * Flush/sync a context
+ */
+extern void XMesaFlush( XMesaContext c );
+
+
+
+/*
+ * Get an X/Mesa-specific string.
+ * Input:  name - either XMESA_VERSION or XMESA_EXTENSIONS
+ */
+extern const char *XMesaGetString( XMesaContext c, int name );
+
+
+
+/*
+ * Scan for XMesaBuffers whose window/pixmap has been destroyed, then free
+ * any memory used by that buffer.
+ *
+ * New in Mesa 2.3.
+ */
+extern void XMesaGarbageCollect( void );
+
+
+
+/*
+ * Return a dithered pixel value.
+ * Input:  c - XMesaContext
+ *         x, y - window coordinate
+ *         red, green, blue, alpha - color components in [0,1]
+ * Return:  pixel value
+ *
+ * New in Mesa 2.3.
+ */
+extern unsigned long XMesaDitherColor( XMesaContext xmesa,
+				       GLint x,
+				       GLint y,
+				       GLfloat red,
+				       GLfloat green,
+				       GLfloat blue,
+				       GLfloat alpha );
+
+
+
+/*
+ * 3Dfx Glide driver only!
+ * Set 3Dfx/Glide full-screen or window rendering mode.
+ * Input:  mode - either XMESA_FX_WINDOW (window rendering mode) or
+ *                XMESA_FX_FULLSCREEN (full-screen rendering mode)
+ * Return:  GL_TRUE if success
+ *          GL_FALSE if invalid mode or if not using 3Dfx driver
+ *
+ * New in Mesa 2.6.
+ */
+extern GLboolean XMesaSetFXmode( GLint mode );
+
+
+
+/*
+ * Reallocate the back/depth/stencil/accum/etc/ buffers associated with
+ * buffer <b> if its size has changed.
+ *
+ * New in Mesa 4.0.2
+ */
+extern void XMesaResizeBuffers( XMesaBuffer b );
+
+
+
+/*
+ * Create a pbuffer.
+ * New in Mesa 4.1
+ */
+extern XMesaBuffer XMesaCreatePBuffer(XMesaVisual v, XMesaColormap cmap,
+                                      unsigned int width, unsigned int height);
+
+
+
+/*
+ * Texture from Pixmap
+ * New in Mesa 7.1
+ */
+extern void
+XMesaBindTexImage(XMesaDisplay *dpy, XMesaBuffer drawable, int buffer,
+                  const int *attrib_list);
+
+extern void
+XMesaReleaseTexImage(XMesaDisplay *dpy, XMesaBuffer drawable, int buffer);
+
+
+extern XMesaBuffer
+XMesaCreatePixmapTextureBuffer(XMesaVisual v, XMesaPixmap p,
+                               XMesaColormap cmap,
+                               int format, int target, int mipmap);
+
+
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/src/mesa/drivers/x11/xmesaP.h b/src/mesa/drivers/x11/xmesaP.h
index 98867ac7106..65e747d7b9d 100644
--- a/src/mesa/drivers/x11/xmesaP.h
+++ b/src/mesa/drivers/x11/xmesaP.h
@@ -27,10 +27,10 @@
 #define XMESAP_H
 
 
-#include "GL/xmesa.h"
+#include "xmesa.h"
 #include "main/mtypes.h"
 #if defined(FX)
-#include "GL/fxmesa.h"
+#include "fxmesa.h"
 #include "xm_glide.h"
 #endif
 #ifdef XFree86Server
diff --git a/src/mesa/drivers/x11/xmesa_x.h b/src/mesa/drivers/x11/xmesa_x.h
new file mode 100644
index 00000000000..865bab4313b
--- /dev/null
+++ b/src/mesa/drivers/x11/xmesa_x.h
@@ -0,0 +1,86 @@
+
+/**************************************************************************
+
+Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sub license, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial portions
+of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
+ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <[email protected]>
+ *
+ * When we're building the XMesa driver for stand-alone Mesa we
+ * include this file when building the xm_*.c files.
+ * We need to define some types and macros differently when building
+ * in the Xserver vs. stand-alone Mesa.
+ */
+
+#ifndef _XMESA_X_H_
+#define _XMESA_X_H_
+
+typedef Display      XMesaDisplay;
+typedef Pixmap       XMesaPixmap;
+typedef Colormap     XMesaColormap;
+typedef Drawable     XMesaDrawable;
+typedef Window       XMesaWindow;
+typedef GC           XMesaGC;
+typedef XVisualInfo *XMesaVisualInfo;
+typedef XImage       XMesaImage;
+typedef XPoint       XMesaPoint;
+typedef XColor       XMesaColor;
+
+#define XMesaDestroyImage      XDestroyImage
+
+#define XMesaPutPixel          XPutPixel
+#define XMesaGetPixel          XGetPixel
+
+#define XMesaSetForeground     XSetForeground
+#define XMesaSetBackground     XSetBackground
+#define XMesaSetPlaneMask      XSetPlaneMask
+#define XMesaSetFunction       XSetFunction
+#define XMesaSetFillStyle      XSetFillStyle
+#define XMesaSetTile           XSetTile
+
+#define XMesaDrawPoint         XDrawPoint
+#define XMesaDrawPoints        XDrawPoints
+#define XMesaDrawLine          XDrawLine
+#define XMesaFillRectangle     XFillRectangle
+#define XMesaGetImage          XGetImage
+#define XMesaPutImage          XPutImage
+#define XMesaCopyArea          XCopyArea
+
+#define XMesaCreatePixmap      XCreatePixmap
+#define XMesaFreePixmap        XFreePixmap
+#define XMesaFreeGC            XFreeGC
+
+#define GET_COLORMAP_SIZE(__v)  __v->visinfo->colormap_size
+#define GET_REDMASK(__v)        __v->mesa_visual.redMask
+#define GET_GREENMASK(__v)      __v->mesa_visual.greenMask
+#define GET_BLUEMASK(__v)       __v->mesa_visual.blueMask
+#define GET_VISUAL_DEPTH(__v)   __v->visinfo->depth
+#define GET_BLACK_PIXEL(__v)    BlackPixel(__v->display, __v->mesa_visual.screen)
+#define CHECK_BYTE_ORDER(__v)   host_byte_order()==ImageByteOrder(__v->display)
+#define CHECK_FOR_HPCR(__v)     XInternAtom(__v->display, "_HP_RGB_SMOOTH_MAP_LIST", True)
+
+#endif
diff --git a/src/mesa/drivers/x11/xmesa_xf86.h b/src/mesa/drivers/x11/xmesa_xf86.h
new file mode 100644
index 00000000000..10f93c3ab62
--- /dev/null
+++ b/src/mesa/drivers/x11/xmesa_xf86.h
@@ -0,0 +1,198 @@
+
+/**************************************************************************
+
+Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sub license, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial portions
+of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+IN NO EVENT SHALL PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
+ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <[email protected]>
+ *
+ * When we're building the XMesa driver for use in the X server (as the
+ * indirect render) we include this file when building the xm_*.c files.
+ * We need to define some types and macros differently when building
+ * in the Xserver vs. stand-alone Mesa.
+ */
+
+#ifndef _XMESA_XF86_H_
+#define _XMESA_XF86_H_
+
+#include "GL/glxtokens.h"
+#include "scrnintstr.h"
+#include "pixmapstr.h"
+#include "gcstruct.h"
+#include "servermd.h"
+
+
+typedef struct _XMesaImageRec {
+    int width, height;
+    char *data;
+    int bytes_per_line; /* Padded to 32 bits */
+    int bits_per_pixel;
+} XMesaImage;
+
+typedef ScreenRec   XMesaDisplay;
+typedef PixmapPtr   XMesaPixmap;
+typedef ColormapPtr XMesaColormap;
+typedef DrawablePtr XMesaDrawable;
+typedef WindowPtr   XMesaWindow;
+typedef GCPtr       XMesaGC;
+typedef VisualPtr   XMesaVisualInfo;
+typedef DDXPointRec XMesaPoint;
+typedef xColorItem  XMesaColor;
+
+#define XMesaSetGeneric(__d,__gc,__val,__mask) \
+do { \
+    CARD32 __v[1]; \
+    (void) __d; \
+    __v[0] = __val; \
+    dixChangeGC(NullClient, __gc, __mask, __v, NULL); \
+} while (0)
+
+#define XMesaSetGenericPtr(__d,__gc,__pval,__mask) \
+do { \
+    ChangeGCVal __v[1]; \
+    (void) __d; \
+    __v[0].ptr = __pval; \
+    dixChangeGC(NullClient, __gc, __mask, NULL, __v); \
+} while (0)
+
+#define XMesaSetForeground(d,gc,v) XMesaSetGeneric(d,gc,v,GCForeground)
+#define XMesaSetBackground(d,gc,v) XMesaSetGeneric(d,gc,v,GCBackground)
+#define XMesaSetPlaneMask(d,gc,v)  XMesaSetGeneric(d,gc,v,GCPlaneMask)
+#define XMesaSetFunction(d,gc,v)   XMesaSetGeneric(d,gc,v,GCFunction)
+#define XMesaSetFillStyle(d,gc,v)  XMesaSetGeneric(d,gc,v,GCFillStyle)
+
+#define XMesaSetTile(d,gc,v)       XMesaSetGenericPtr(d,gc,v,GCTile)
+
+#define XMesaDrawPoint(__d,__b,__gc,__x,__y) \
+do { \
+    XMesaPoint __p[1]; \
+    (void) __d; \
+    __p[0].x = __x; \
+    __p[0].y = __y; \
+    ValidateGC(__b, __gc); \
+    (*gc->ops->PolyPoint)(__b, __gc, CoordModeOrigin, 1, __p); \
+} while (0)
+
+#define XMesaDrawPoints(__d,__b,__gc,__p,__n,__m) \
+do { \
+    (void) __d; \
+    ValidateGC(__b, __gc); \
+    (*gc->ops->PolyPoint)(__b, __gc, __m, __n, __p); \
+} while (0)
+
+#define XMesaDrawLine(__d, __b, __gc, __x0, __y0, __x1, __y1) \
+do { \
+    XMesaPoint __p[2]; \
+    (void) __d; \
+    __p[0].x = __x0; \
+    __p[0].y = __y0; \
+    __p[1].x = __x1; \
+    __p[1].y = __y1; \
+    ValidateGC(__b, __gc); \
+    (*gc->ops->PolyLines)(__b, __gc, CoordModeOrigin, 2, __p); \
+} while (0)
+
+#define XMesaFillRectangle(__d,__b,__gc,__x,__y,__w,__h) \
+do { \
+    xRectangle __r[1]; \
+    (void) __d; \
+    ValidateGC((DrawablePtr)__b, __gc); \
+    __r[0].x = __x; \
+    __r[0].y = __y; \
+    __r[0].width = __w; \
+    __r[0].height = __h; \
+    (*__gc->ops->PolyFillRect)((DrawablePtr)__b, __gc, 1, __r); \
+} while (0)
+
+static _X_INLINE XMesaImage *XMesaGetImage(XMesaDisplay *dpy, PixmapPtr p, int x,
+					int y, unsigned int width,
+					unsigned int height,
+					unsigned long plane_mask, int format)
+{
+    XMesaImage *img = Xcalloc(sizeof(*img));
+
+    img->width = p->drawable.width;
+    img->height = p->drawable.height;
+    img->bits_per_pixel = p->drawable.bitsPerPixel;
+    img->bytes_per_line = PixmapBytePad(width, p->drawable.depth);
+    img->data = malloc(height * img->bytes_per_line);
+
+    /* Assumes: Images are always in ZPixmap format */
+    (*p->drawable.pScreen->GetImage)(&p->drawable, x, y, width, height,
+				     plane_mask, ZPixmap, img->data);
+
+    return img;
+}
+
+#define XMesaPutImage(__d,__b,__gc,__i,__sx,__sy,__x,__y,__w,__h) \
+do { \
+    /* Assumes: Images are always in ZPixmap format */ \
+    (void) __d; \
+    ASSERT(!__sx && !__sy); /* The SubImage case */     \
+    ValidateGC(__b, __gc); \
+    (*__gc->ops->PutImage)(__b, __gc, ((XMesaDrawable)(__b))->depth, \
+			   __x, __y, __w, __h, 0, ZPixmap, \
+			   ((XMesaImage *)(__i))->data); \
+} while (0)
+
+#define XMesaCopyArea(__d,__sb,__db,__gc,__sx,__sy,__w,__h,__x,__y) \
+do { \
+    (void) __d; \
+    ValidateGC(__db, __gc); \
+    (*__gc->ops->CopyArea)((DrawablePtr)__sb, __db, __gc, \
+			   __sx, __sy, __w, __h, __x, __y); \
+} while (0)
+
+
+/* CreatePixmap returns a PixmapPtr; so, it cannot be inside braces */
+#ifdef CREATE_PIXMAP_USAGE_SCRATCH
+#define XMesaCreatePixmap(__d,__b,__w,__h,__depth) \
+  (*__d->CreatePixmap)(__d, __w, __h, __depth, 0)
+#else
+#define XMesaCreatePixmap(__d,__b,__w,__h,__depth) \
+    (*__d->CreatePixmap)(__d, __w, __h, __depth)
+#endif
+
+#define XMesaFreePixmap(__d,__b) \
+    (*__d->DestroyPixmap)(__b)
+
+#define XMesaFreeGC(__d,__gc) \
+do { \
+    (void) __d; \
+    FreeScratchGC(__gc); \
+} while (0)
+
+#define GET_COLORMAP_SIZE(__v)  __v->ColormapEntries
+#define GET_REDMASK(__v)        __v->mesa_visual.redMask
+#define GET_GREENMASK(__v)      __v->mesa_visual.greenMask
+#define GET_BLUEMASK(__v)       __v->mesa_visual.blueMask
+#define GET_VISUAL_DEPTH(__v)   __v->nplanes
+#define GET_BLACK_PIXEL(__v)    __v->display->blackPixel
+#define CHECK_BYTE_ORDER(__v)   GL_TRUE
+#define CHECK_FOR_HPCR(__v)     GL_FALSE
+
+#endif
diff --git a/src/mesa/glapi/glthread.h b/src/mesa/glapi/glthread.h
index ad2a63584b7..ae2f79f37dd 100644
--- a/src/mesa/glapi/glthread.h
+++ b/src/mesa/glapi/glthread.h
@@ -338,6 +338,10 @@ _glthread_GetTSD(_glthread_TSD *);
 extern void
 _glthread_SetTSD(_glthread_TSD *, void *);
 
+#if !defined __GNUC__ || __GNUC__ < 3
+#  define __builtin_expect(x, y) x
+#endif
+
 #if defined(GLX_USE_TLS)
 
 extern __thread struct _glapi_table * _glapi_tls_Dispatch
diff --git a/src/mesa/main/api_validate.c b/src/mesa/main/api_validate.c
index bbc5933ab9f..5c8955d7c8d 100644
--- a/src/mesa/main/api_validate.c
+++ b/src/mesa/main/api_validate.c
@@ -78,6 +78,23 @@ max_buffer_index(GLcontext *ctx, GLuint count, GLenum type,
    return max;
 }
 
+static GLboolean
+check_valid_to_render(GLcontext *ctx, char *function)
+{
+   if (ctx->DrawBuffer->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) {
+      _mesa_error(ctx, GL_INVALID_FRAMEBUFFER_OPERATION_EXT,
+                  "glDraw%s(incomplete framebuffer)", function);
+      return GL_FALSE;
+   }
+
+   /* Always need vertex positions, unless a vertex program is in use */
+   if (!ctx->VertexProgram._Current &&
+       !ctx->Array.ArrayObj->Vertex.Enabled &&
+       !ctx->Array.ArrayObj->VertexAttrib[0].Enabled)
+      return GL_FALSE;
+
+   return GL_TRUE;
+}
 
 GLboolean
 _mesa_validate_DrawElements(GLcontext *ctx,
@@ -108,10 +125,7 @@ _mesa_validate_DrawElements(GLcontext *ctx,
    if (ctx->NewState)
       _mesa_update_state(ctx);
 
-   /* Always need vertex positions, unless a vertex program is in use */
-   if (!ctx->VertexProgram._Current &&
-       !ctx->Array.ArrayObj->Vertex.Enabled &&
-       !ctx->Array.ArrayObj->VertexAttrib[0].Enabled)
+   if (!check_valid_to_render(ctx, "Elements"))
       return GL_FALSE;
 
    /* Vertex buffer object tests */
@@ -161,7 +175,6 @@ _mesa_validate_DrawElements(GLcontext *ctx,
    return GL_TRUE;
 }
 
-
 GLboolean
 _mesa_validate_DrawRangeElements(GLcontext *ctx, GLenum mode,
 				 GLuint start, GLuint end,
@@ -196,10 +209,7 @@ _mesa_validate_DrawRangeElements(GLcontext *ctx, GLenum mode,
    if (ctx->NewState)
       _mesa_update_state(ctx);
 
-   /* Always need vertex positions, unless a vertex program is in use */
-   if (!ctx->VertexProgram._Current &&
-       !ctx->Array.ArrayObj->Vertex.Enabled &&
-       !ctx->Array.ArrayObj->VertexAttrib[0].Enabled)
+   if (!check_valid_to_render(ctx, "RangeElements"))
       return GL_FALSE;
 
    /* Vertex buffer object tests */
@@ -267,10 +277,7 @@ _mesa_validate_DrawArrays(GLcontext *ctx,
    if (ctx->NewState)
       _mesa_update_state(ctx);
 
-   /* Always need vertex positions, unless a vertex program is in use */
-   if (!ctx->VertexProgram._Current &&
-       !ctx->Array.ArrayObj->Vertex.Enabled &&
-       !ctx->Array.ArrayObj->VertexAttrib[0].Enabled)
+   if (!check_valid_to_render(ctx, "Arrays"))
       return GL_FALSE;
 
    if (ctx->Const.CheckArrayBounds) {
diff --git a/src/mesa/main/glheader.h b/src/mesa/main/glheader.h
index d69c7bbb21d..5657976711e 100644
--- a/src/mesa/main/glheader.h
+++ b/src/mesa/main/glheader.h
@@ -157,7 +157,8 @@
 #include <byteswap.h>
 #define CPU_TO_LE32( x )	bswap_32( x )
 #else /*__linux__*/
-#define CPU_TO_LE32( x )	( x )  /* fix me for non-Linux big-endian! */
+#include <sys/endian.h>
+#define CPU_TO_LE32( x )	bswap32( x )
 #endif /*__linux__*/
 #define MESA_BIG_ENDIAN 1
 #else
diff --git a/src/mesa/main/light.c b/src/mesa/main/light.c
index d4db960f1b9..a15c1f0be05 100644
--- a/src/mesa/main/light.c
+++ b/src/mesa/main/light.c
@@ -208,7 +208,8 @@ _mesa_Lightfv( GLenum light, GLenum pname, const GLfloat *params )
       if (_math_matrix_is_dirty(ctx->ModelviewMatrixStack.Top)) {
 	 _math_matrix_analyse(ctx->ModelviewMatrixStack.Top);
       }
-      TRANSFORM_NORMAL(temp, params, ctx->ModelviewMatrixStack.Top->inv);
+      TRANSFORM_DIRECTION(temp, params, ctx->ModelviewMatrixStack.Top->m);
+      NORMALIZE_3FV(temp);
       params = temp;
       break;
    case GL_SPOT_EXPONENT:
diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
index 25dee52ad07..a61f970685c 100644
--- a/src/mesa/main/mtypes.h
+++ b/src/mesa/main/mtypes.h
@@ -2097,6 +2097,13 @@ struct gl_query_state
 };
 
 
+/** Set by #pragma directives */
+struct gl_sl_pragmas
+{
+   GLboolean Optimize;  /**< defaults on */
+   GLboolean Debug;     /**< defaults off */
+};
+
 
 /**
  * A GLSL vertex or fragment shader object.
@@ -2107,12 +2114,12 @@ struct gl_shader
    GLuint Name;  /**< AKA the handle */
    GLint RefCount;  /**< Reference count */
    GLboolean DeletePending;
-
-   const GLchar *Source;  /**< Source code string */
    GLboolean CompileStatus;
+   GLboolean Main;  /**< shader defines main() */
+   const GLchar *Source;  /**< Source code string */
    struct gl_program *Program;  /**< Post-compile assembly code */
    GLchar *InfoLog;
-   GLboolean Main;  /**< shader defines main() */
+   struct gl_sl_pragmas Pragmas;
 };
 
 
diff --git a/src/mesa/main/version.h b/src/mesa/main/version.h
index 83aefe685ad..24495d608a5 100644
--- a/src/mesa/main/version.h
+++ b/src/mesa/main/version.h
@@ -31,7 +31,7 @@
 #define MESA_MAJOR 7
 #define MESA_MINOR 3
 #define MESA_PATCH 0
-#define MESA_VERSION_STRING "7.3-rc1"
+#define MESA_VERSION_STRING "7.3-rc2"
 
 /* To make version comparison easy */
 #define MESA_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
diff --git a/src/mesa/math/m_matrix.h b/src/mesa/math/m_matrix.h
index e8303f3ac5c..a8d9000e89b 100644
--- a/src/mesa/math/m_matrix.h
+++ b/src/mesa/math/m_matrix.h
@@ -189,6 +189,18 @@ do {								\
 } while (0)
 
 
+/**
+ * Transform a direction by a matrix.
+ */
+#define TRANSFORM_DIRECTION( TO, DIR, MAT )			\
+do {								\
+   TO[0] = DIR[0] * MAT[0] + DIR[1] * MAT[4] + DIR[2] * MAT[8];	\
+   TO[1] = DIR[0] * MAT[1] + DIR[1] * MAT[5] + DIR[2] * MAT[9];	\
+   TO[2] = DIR[0] * MAT[2] + DIR[1] * MAT[6] + DIR[2] * MAT[10];\
+} while (0)
+
+
+
 /*@}*/
 
 
diff --git a/src/mesa/shader/arbprogparse.c b/src/mesa/shader/arbprogparse.c
index 536404bf978..5b80d99d49c 100644
--- a/src/mesa/shader/arbprogparse.c
+++ b/src/mesa/shader/arbprogparse.c
@@ -979,6 +979,8 @@ parse_output_color_num (GLcontext * ctx, const GLubyte ** inst,
 
 
 /**
+ * Validate the index of a texture coordinate
+ *
  * \param coord The texture unit index
  * \return 0 on sucess, 1 on error
  */
@@ -988,8 +990,30 @@ parse_texcoord_num (GLcontext * ctx, const GLubyte ** inst,
 {
    GLint i = parse_integer (inst, Program);
 
-   if ((i < 0) || (i >= (int)ctx->Const.MaxTextureUnits)) {
-      program_error(ctx, Program->Position, "Invalid texture unit index");
+   if ((i < 0) || (i >= (int)ctx->Const.MaxTextureCoordUnits)) {
+      program_error(ctx, Program->Position, "Invalid texture coordinate index");
+      return 1;
+   }
+
+   *coord = (GLuint) i;
+   return 0;
+}
+
+
+/**
+ * Validate the index of a texture image unit
+ *
+ * \param coord The texture unit index
+ * \return 0 on sucess, 1 on error
+ */
+static GLuint
+parse_teximage_num (GLcontext * ctx, const GLubyte ** inst,
+                    struct arb_program *Program, GLuint * coord)
+{
+   GLint i = parse_integer (inst, Program);
+
+   if ((i < 0) || (i >= (int)ctx->Const.MaxTextureImageUnits)) {
+      program_error(ctx, Program->Position, "Invalid texture image index");
       return 1;
    }
 
@@ -997,6 +1021,7 @@ parse_texcoord_num (GLcontext * ctx, const GLubyte ** inst,
    return 0;
 }
 
+
 /**
  * \param coord The weight index
  * \return 0 on sucess, 1 on error
@@ -3035,7 +3060,7 @@ parse_fp_instruction (GLcontext * ctx, const GLubyte ** inst,
             return 1;
 
          /* texImageUnit */
-         if (parse_texcoord_num (ctx, inst, Program, &texcoord))
+         if (parse_teximage_num (ctx, inst, Program, &texcoord))
             return 1;
          fp->TexSrcUnit = texcoord;
 
diff --git a/src/mesa/shader/prog_instruction.c b/src/mesa/shader/prog_instruction.c
index f5c0a498fb0..4a6d0d670ac 100644
--- a/src/mesa/shader/prog_instruction.c
+++ b/src/mesa/shader/prog_instruction.c
@@ -1,8 +1,9 @@
 /*
  * Mesa 3-D graphics library
- * Version:  6.5.3
+ * Version:  7.3
  *
- * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2009  VMware, Inc.  All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -157,7 +158,7 @@ static const struct instruction_info InstInfo[MAX_OPCODE] = {
    { OPCODE_AND,    "AND",     2, 1 },
    { OPCODE_ARA,    "ARA",     1, 1 },
    { OPCODE_ARL,    "ARL",     1, 1 },
-   { OPCODE_ARL_NV, "ARL",     1, 1 },
+   { OPCODE_ARL_NV, "ARL_NV",  1, 1 },
    { OPCODE_ARR,    "ARL",     1, 1 },
    { OPCODE_BGNLOOP,"BGNLOOP", 0, 0 },
    { OPCODE_BGNSUB, "BGNSUB",  0, 0 },
@@ -186,7 +187,7 @@ static const struct instruction_info InstInfo[MAX_OPCODE] = {
    { OPCODE_FRC,    "FRC",     1, 1 },
    { OPCODE_IF,     "IF",      1, 0 },
    { OPCODE_KIL,    "KIL",     1, 0 },
-   { OPCODE_KIL_NV, "KIL",     0, 0 },
+   { OPCODE_KIL_NV, "KIL_NV",  0, 0 },
    { OPCODE_LG2,    "LG2",     1, 1 },
    { OPCODE_LIT,    "LIT",     1, 1 },
    { OPCODE_LOG,    "LOG",     1, 1 },
@@ -235,7 +236,7 @@ static const struct instruction_info InstInfo[MAX_OPCODE] = {
    { OPCODE_TXD,    "TXD",     3, 1 },
    { OPCODE_TXL,    "TXL",     1, 1 },
    { OPCODE_TXP,    "TXP",     1, 1 },
-   { OPCODE_TXP_NV, "TXP",     1, 1 },
+   { OPCODE_TXP_NV, "TXP_NV",  1, 1 },
    { OPCODE_TRUNC,  "TRUNC",   1, 1 },
    { OPCODE_UP2H,   "UP2H",    1, 1 },
    { OPCODE_UP2US,  "UP2US",   1, 1 },
diff --git a/src/mesa/shader/slang/library/slang_pp_directives.syn b/src/mesa/shader/slang/library/slang_pp_directives.syn
index d4a321034d1..b51d168cc03 100644
--- a/src/mesa/shader/slang/library/slang_pp_directives.syn
+++ b/src/mesa/shader/slang/library/slang_pp_directives.syn
@@ -79,6 +79,12 @@
 .emtcode BEHAVIOR_WARN    3
 .emtcode BEHAVIOR_DISABLE 4
 
+/*
+ * The PRAGMA_* symbols follow TOKEN_PRAGMA
+ */
+.emtcode PRAGMA_NO_PARAM 0
+.emtcode PRAGMA_PARAM    1
+
 source
    optional_directive .and .loop source_element .and '\0' .emit ESCAPE_TOKEN .emit TOKEN_END;
 
@@ -153,6 +159,7 @@ directive
    dir_elif .emit TOKEN_ELIF .or
    dir_endif .emit TOKEN_ENDIF .or
    dir_ext .emit TOKEN_EXTENSION .or
+   dir_pragma .emit TOKEN_PRAGMA .or
    dir_line .emit TOKEN_LINE;
 
 dir_define
@@ -187,6 +194,19 @@ dir_ext
 dir_line
    optional_space .and '#' .and optional_space .and "line" .and expression;
 
+dir_pragma
+   optional_space .and '#' .and optional_space .and "pragma" .and symbol .and opt_pragma_param;
+
+
+opt_pragma_param
+   pragma_param .or .true .emit PRAGMA_NO_PARAM;
+
+pragma_param
+   optional_space .and '(' .emit PRAGMA_PARAM .and optional_space .and symbol_no_space .and optional_space .and ')';
+
+symbol_no_space
+   symbol_character .emit * .and .loop symbol_character2 .emit * .and .true .emit '\0';
+
 symbol
    space .and symbol_character .emit * .and .loop symbol_character2 .emit * .and .true .emit '\0';
 
diff --git a/src/mesa/shader/slang/library/slang_pp_directives_syn.h b/src/mesa/shader/slang/library/slang_pp_directives_syn.h
index 35e7bc27616..430f8d8217c 100644
--- a/src/mesa/shader/slang/library/slang_pp_directives_syn.h
+++ b/src/mesa/shader/slang/library/slang_pp_directives_syn.h
@@ -20,6 +20,8 @@
 ".emtcode BEHAVIOR_ENABLE 2\n"
 ".emtcode BEHAVIOR_WARN 3\n"
 ".emtcode BEHAVIOR_DISABLE 4\n"
+".emtcode PRAGMA_NO_PARAM 0\n"
+".emtcode PRAGMA_PARAM 1\n"
 "source\n"
 " optional_directive .and .loop source_element .and '\\0' .emit ESCAPE_TOKEN .emit TOKEN_END;\n"
 "source_element\n"
@@ -76,6 +78,7 @@
 " dir_elif .emit TOKEN_ELIF .or\n"
 " dir_endif .emit TOKEN_ENDIF .or\n"
 " dir_ext .emit TOKEN_EXTENSION .or\n"
+" dir_pragma .emit TOKEN_PRAGMA .or\n"
 " dir_line .emit TOKEN_LINE;\n"
 "dir_define\n"
 " optional_space .and '#' .and optional_space .and \"define\" .and symbol .and opt_parameters .and\n"
@@ -99,6 +102,14 @@
 " optional_space .and ':' .and optional_space .and extension_behavior;\n"
 "dir_line\n"
 " optional_space .and '#' .and optional_space .and \"line\" .and expression;\n"
+"dir_pragma\n"
+" optional_space .and '#' .and optional_space .and \"pragma\" .and symbol .and opt_pragma_param;\n"
+"opt_pragma_param\n"
+" pragma_param .or .true .emit PRAGMA_NO_PARAM;\n"
+"pragma_param\n"
+" optional_space .and '(' .emit PRAGMA_PARAM .and optional_space .and symbol_no_space .and optional_space .and ')';\n"
+"symbol_no_space\n"
+" symbol_character .emit * .and .loop symbol_character2 .emit * .and .true .emit '\\0';\n"
 "symbol\n"
 " space .and symbol_character .emit * .and .loop symbol_character2 .emit * .and .true .emit '\\0';\n"
 "opt_parameters\n"
diff --git a/src/mesa/shader/slang/library/slang_shader.syn b/src/mesa/shader/slang/library/slang_shader.syn
index 0ad4a1cba62..cc5c70a02f8 100644
--- a/src/mesa/shader/slang/library/slang_shader.syn
+++ b/src/mesa/shader/slang/library/slang_shader.syn
@@ -1162,13 +1162,6 @@ compound_statement_3
     lbrace .and statement_list .and rbrace;
 
 /*
- * <statement_no_new_scope> ::= <compound_statement_no_new_scope>
- *                            | <simple_statement>
- */
-statement_no_new_scope
-    compound_statement_no_new_scope .or simple_statement;
-
-/*
  * <compound_statement_no_new_scope> ::= "{" "}"
  *                                     | "{" <statement_list> "}"
  */
@@ -1181,6 +1174,7 @@ compound_statement_no_new_scope_2
 compound_statement_no_new_scope_3
     lbrace .and statement_list .and rbrace;
 
+
 /*
  * <statement_list> ::= <statement>
  *                    | <statement_list> <statement>
@@ -1242,8 +1236,7 @@ condition_3
 /*
  * <iteration_statement> ::= "while" "(" <condition> ")" <statement>
  *                         | "do" <statement> "while" "(" <expression> ")" ";"
- *                         | "for" "(" <for_init_statement> <for_rest_statement> ")"
- *                            <statement_no_new_scope>
+ *                         | "for" "(" <for_init_statement> <for_rest_statement> ")" <statement>
  */
 iteration_statement
     iteration_statement_1 .or iteration_statement_2 .or iteration_statement_3;
@@ -1255,7 +1248,7 @@ iteration_statement_2
     expression .and rparen .error RPAREN_EXPECTED .emit OP_END .and semicolon;
 iteration_statement_3
     "for" .emit OP_FOR .and lparen .error LPAREN_EXPECTED .and for_init_statement .and
-    for_rest_statement .and rparen .error RPAREN_EXPECTED .and statement_no_new_scope;
+    for_rest_statement .and rparen .error RPAREN_EXPECTED .and statement;
 
 /*
  * <for_init_statement> ::= <expression_statement>
diff --git a/src/mesa/shader/slang/library/slang_shader_syn.h b/src/mesa/shader/slang/library/slang_shader_syn.h
index e1705dfe193..6a382970e1a 100644
--- a/src/mesa/shader/slang/library/slang_shader_syn.h
+++ b/src/mesa/shader/slang/library/slang_shader_syn.h
@@ -566,8 +566,6 @@
 " lbrace .and rbrace;\n"
 "compound_statement_3\n"
 " lbrace .and statement_list .and rbrace;\n"
-"statement_no_new_scope\n"
-" compound_statement_no_new_scope .or simple_statement;\n"
 "compound_statement_no_new_scope\n"
 " compound_statement_no_new_scope_1 .emit OP_BLOCK_BEGIN_NO_NEW_SCOPE .and .true .emit OP_END;\n"
 "compound_statement_no_new_scope_1\n"
@@ -617,7 +615,7 @@
 " expression .and rparen .error RPAREN_EXPECTED .emit OP_END .and semicolon;\n"
 "iteration_statement_3\n"
 " \"for\" .emit OP_FOR .and lparen .error LPAREN_EXPECTED .and for_init_statement .and\n"
-" for_rest_statement .and rparen .error RPAREN_EXPECTED .and statement_no_new_scope;\n"
+" for_rest_statement .and rparen .error RPAREN_EXPECTED .and statement;\n"
 "for_init_statement\n"
 " expression_statement .emit OP_EXPRESSION .or declaration_statement .emit OP_DECLARE;\n"
 "conditionopt\n"
diff --git a/src/mesa/shader/slang/slang_codegen.c b/src/mesa/shader/slang/slang_codegen.c
index ba1c955a1ae..11340d26e21 100644
--- a/src/mesa/shader/slang/slang_codegen.c
+++ b/src/mesa/shader/slang/slang_codegen.c
@@ -228,7 +228,14 @@ _slang_sizeof_type_specifier(const slang_type_specifier *spec)
       break;
    case SLANG_SPEC_STRUCT:
       sz = _slang_field_offset(spec, 0); /* special use */
-      if (sz > 4) {
+      if (sz == 1) {
+         /* 1-float structs are actually troublesome to deal with since they
+          * might get placed at R.x, R.y, R.z or R.z.  Return size=2 to
+          * ensure the object is placed at R.x
+          */
+         sz = 2;
+      }
+      else if (sz > 4) {
          sz = (sz + 3) & ~0x3; /* round up to multiple of four */
       }
       break;
@@ -4232,6 +4239,21 @@ _slang_gen_operation(slang_assemble_ctx * A, slang_operation *oper)
 
 
 /**
+ * Check if the given type specifier is a rectangular texture sampler.
+ */
+static GLboolean
+is_rect_sampler_spec(const slang_type_specifier *spec)
+{
+   while (spec->_array) {
+      spec = spec->_array;
+   }
+   return spec->type == SLANG_SPEC_SAMPLER2DRECT ||
+          spec->type == SLANG_SPEC_SAMPLER2DRECTSHADOW;
+}
+
+
+
+/**
  * Called by compiler when a global variable has been parsed/compiled.
  * Here we examine the variable's type to determine what kind of register
  * storage will be used.
@@ -4254,10 +4276,14 @@ _slang_codegen_global_variable(slang_assemble_ctx *A, slang_variable *var,
    slang_ir_storage *store = NULL;
    int dbg = 0;
    const GLenum datatype = _slang_gltype_from_specifier(&var->type.specifier);
-   const GLint texIndex = sampler_to_texture_index(var->type.specifier.type);
    const GLint size = _slang_sizeof_type_specifier(&var->type.specifier);
    const GLint arrayLen = _slang_array_length(var);
    const GLint totalSize = _slang_array_size(size, arrayLen);
+   GLint texIndex = sampler_to_texture_index(var->type.specifier.type);
+
+   /* check for sampler2D arrays */
+   if (texIndex == -1 && var->type.specifier._array)
+      texIndex = sampler_to_texture_index(var->type.specifier._array->type);
 
    if (texIndex != -1) {
       /* This is a texture sampler variable...
@@ -4271,15 +4297,32 @@ _slang_codegen_global_variable(slang_assemble_ctx *A, slang_variable *var,
       }
 #if FEATURE_es2_glsl /* XXX should use FEATURE_texture_rect */
       /* disallow rect samplers */
-      if (var->type.specifier.type == SLANG_SPEC_SAMPLER2DRECT ||
-          var->type.specifier.type == SLANG_SPEC_SAMPLER2DRECTSHADOW) {
+      if (is_rect_sampler_spec(&var->type.specifier)) {
          slang_info_log_error(A->log, "invalid sampler type for '%s'", varName);
          return GL_FALSE;
       }
+#else
+      (void) is_rect_sampler_spec; /* silence warning */
 #endif
       {
          GLint sampNum = _mesa_add_sampler(prog->Parameters, varName, datatype);
-         store = _slang_new_ir_storage(PROGRAM_SAMPLER, sampNum, texIndex);
+         store = _slang_new_ir_storage_sampler(sampNum, texIndex, totalSize);
+
+         /* If we have a sampler array, then we need to allocate the 
+	  * additional samplers to ensure we don't allocate them elsewhere.
+	  * We can't directly use _mesa_add_sampler() as that checks the
+	  * varName and gets a match, so we call _mesa_add_parameter()
+	  * directly and use the last sampler number from the call above.
+	  */
+	 if (arrayLen > 0) {
+	    GLint a = arrayLen - 1;
+	    GLint i;
+	    for (i = 0; i < a; i++) {
+               GLfloat value = (GLfloat)(i + sampNum + 1);
+               (void) _mesa_add_parameter(prog->Parameters, PROGRAM_SAMPLER,
+                                 varName, 1, datatype, &value, NULL, 0x0);
+	    }
+	 }
       }
       if (dbg) printf("SAMPLER ");
    }
@@ -4463,7 +4506,7 @@ _slang_codegen_global_variable(slang_assemble_ctx *A, slang_variable *var,
       n = _slang_gen_var_decl(A, var, var->initializer);
 
       /* emit GPU instructions */
-      success = _slang_emit_code(n, A->vartable, A->program, GL_FALSE, A->log);
+      success = _slang_emit_code(n, A->vartable, A->program, A->pragmas, GL_FALSE, A->log);
 
       _slang_free_ir_tree(n);
    }
@@ -4573,7 +4616,7 @@ _slang_codegen_function(slang_assemble_ctx * A, slang_function * fun)
 #endif
 
    /* Emit program instructions */
-   success = _slang_emit_code(n, A->vartable, A->program, GL_TRUE, A->log);
+   success = _slang_emit_code(n, A->vartable, A->program, A->pragmas, GL_TRUE, A->log);
    _slang_free_ir_tree(n);
 
    /* free codegen context */
diff --git a/src/mesa/shader/slang/slang_codegen.h b/src/mesa/shader/slang/slang_codegen.h
index 9489033d7b8..f2daa034e4f 100644
--- a/src/mesa/shader/slang/slang_codegen.h
+++ b/src/mesa/shader/slang/slang_codegen.h
@@ -36,6 +36,7 @@ typedef struct slang_assemble_ctx_
    slang_atom_pool *atoms;
    slang_name_space space;
    struct gl_program *program;
+   struct gl_sl_pragmas *pragmas;
    slang_var_table *vartable;
    slang_info_log *log;
    struct slang_label_ *curFuncEndLabel;
diff --git a/src/mesa/shader/slang/slang_compile.c b/src/mesa/shader/slang/slang_compile.c
index 0a527f32206..818b90b7a86 100644
--- a/src/mesa/shader/slang/slang_compile.c
+++ b/src/mesa/shader/slang/slang_compile.c
@@ -144,6 +144,7 @@ typedef struct slang_output_ctx_
    slang_function_scope *funs;
    slang_struct_scope *structs;
    struct gl_program *program;
+   struct gl_sl_pragmas *pragmas;
    slang_var_table *vartable;
    GLuint default_precision[TYPE_SPECIFIER_COUNT];
    GLboolean allow_precision;
@@ -1005,7 +1006,7 @@ parse_statement(slang_parse_ctx * C, slang_output_ctx * O,
       /* parse child statements, do not create new variable scope */
       oper->type = SLANG_OPER_BLOCK_NO_NEW_SCOPE;
       while (*C->I != OP_END)
-         if (!parse_child_operation(C, O, oper, 1))
+         if (!parse_child_operation(C, O, oper, GL_TRUE))
             RETURN0;
       C->I++;
       break;
@@ -1017,7 +1018,7 @@ parse_statement(slang_parse_ctx * C, slang_output_ctx * O,
          oper->type = SLANG_OPER_BLOCK_NEW_SCOPE;
          o.vars = oper->locals;
          while (*C->I != OP_END)
-            if (!parse_child_operation(C, &o, oper, 1))
+            if (!parse_child_operation(C, &o, oper, GL_TRUE))
                RETURN0;
          C->I++;
       }
@@ -1074,7 +1075,7 @@ parse_statement(slang_parse_ctx * C, slang_output_ctx * O,
       if (oper->a_id == SLANG_ATOM_NULL)
          RETURN0;
       while (*C->I != OP_END) {
-         if (!parse_child_operation(C, O, oper, 0))
+         if (!parse_child_operation(C, O, oper, GL_FALSE))
             RETURN0;
       }
       C->I++;
@@ -1090,21 +1091,21 @@ parse_statement(slang_parse_ctx * C, slang_output_ctx * O,
       break;
    case OP_RETURN:
       oper->type = SLANG_OPER_RETURN;
-      if (!parse_child_operation(C, O, oper, 0))
+      if (!parse_child_operation(C, O, oper, GL_FALSE))
          RETURN0;
       break;
    case OP_EXPRESSION:
       oper->type = SLANG_OPER_EXPRESSION;
-      if (!parse_child_operation(C, O, oper, 0))
+      if (!parse_child_operation(C, O, oper, GL_FALSE))
          RETURN0;
       break;
    case OP_IF:
       oper->type = SLANG_OPER_IF;
-      if (!parse_child_operation(C, O, oper, 0))
+      if (!parse_child_operation(C, O, oper, GL_FALSE))
          RETURN0;
-      if (!parse_child_operation(C, O, oper, 1))
+      if (!parse_child_operation(C, O, oper, GL_TRUE))
          RETURN0;
-      if (!parse_child_operation(C, O, oper, 1))
+      if (!parse_child_operation(C, O, oper, GL_TRUE))
          RETURN0;
       break;
    case OP_WHILE:
@@ -1113,17 +1114,17 @@ parse_statement(slang_parse_ctx * C, slang_output_ctx * O,
 
          oper->type = SLANG_OPER_WHILE;
          o.vars = oper->locals;
-         if (!parse_child_operation(C, &o, oper, 1))
+         if (!parse_child_operation(C, &o, oper, GL_TRUE))
             RETURN0;
-         if (!parse_child_operation(C, &o, oper, 1))
+         if (!parse_child_operation(C, &o, oper, GL_TRUE))
             RETURN0;
       }
       break;
    case OP_DO:
       oper->type = SLANG_OPER_DO;
-      if (!parse_child_operation(C, O, oper, 1))
+      if (!parse_child_operation(C, O, oper, GL_TRUE))
          RETURN0;
-      if (!parse_child_operation(C, O, oper, 0))
+      if (!parse_child_operation(C, O, oper, GL_FALSE))
          RETURN0;
       break;
    case OP_FOR:
@@ -1132,13 +1133,13 @@ parse_statement(slang_parse_ctx * C, slang_output_ctx * O,
 
          oper->type = SLANG_OPER_FOR;
          o.vars = oper->locals;
-         if (!parse_child_operation(C, &o, oper, 1))
+         if (!parse_child_operation(C, &o, oper, GL_TRUE))
             RETURN0;
-         if (!parse_child_operation(C, &o, oper, 1))
+         if (!parse_child_operation(C, &o, oper, GL_TRUE))
             RETURN0;
-         if (!parse_child_operation(C, &o, oper, 0))
+         if (!parse_child_operation(C, &o, oper, GL_FALSE))
             RETURN0;
-         if (!parse_child_operation(C, &o, oper, 1))
+         if (!parse_child_operation(C, &o, oper, GL_TRUE))
             RETURN0;
       }
       break;
@@ -1429,7 +1430,7 @@ parse_expression(slang_parse_ctx * C, slang_output_ctx * O,
          C->I++;
 
          while (*C->I != OP_END)
-            if (!parse_child_operation(C, O, op, 0))
+            if (!parse_child_operation(C, O, op, GL_FALSE))
                RETURN0;
          C->I++;
 #if 0
@@ -1470,9 +1471,9 @@ parse_expression(slang_parse_ctx * C, slang_output_ctx * O,
                   RETURN0;
                }
                else {
+                  /* parse the array constructor size */
                   slang_operation array_size;
                   array_constructor = GL_TRUE;
-                  /* parse the array constructor size */
                   slang_operation_construct(&array_size);
                   if (!parse_expression(C, O, &array_size)) {
                      slang_operation_destruct(&array_size);
@@ -1494,7 +1495,7 @@ parse_expression(slang_parse_ctx * C, slang_output_ctx * O,
                RETURN0;
             }
             while (*C->I != OP_END)
-               if (!parse_child_operation(C, O, op, 0))
+               if (!parse_child_operation(C, O, op, GL_FALSE))
                   RETURN0;
             C->I++;
 
@@ -2059,6 +2060,7 @@ parse_init_declarator(slang_parse_ctx * C, slang_output_ctx * O,
       A.space.structs = O->structs;
       A.space.vars = O->vars;
       A.program = O->program;
+      A.pragmas = O->pragmas;
       A.vartable = O->vartable;
       A.log = C->L;
       A.curFuncEndLabel = NULL;
@@ -2349,6 +2351,7 @@ parse_code_unit(slang_parse_ctx * C, slang_code_unit * unit,
    o.structs = &unit->structs;
    o.vars = &unit->vars;
    o.program = shader ? shader->Program : NULL;
+   o.pragmas = shader ? &shader->Pragmas : NULL;
    o.vartable = _slang_new_var_table(maxRegs);
    _slang_push_var_table(o.vartable);
 
@@ -2417,6 +2420,7 @@ parse_code_unit(slang_parse_ctx * C, slang_code_unit * unit,
       A.space.structs = o.structs;
       A.space.vars = o.vars;
       A.program = o.program;
+      A.pragmas = &shader->Pragmas;
       A.vartable = o.vartable;
       A.log = C->L;
 
@@ -2475,7 +2479,8 @@ compile_with_grammar(grammar id, const char *source, slang_code_unit * unit,
                      slang_unit_type type, slang_info_log * infolog,
                      slang_code_unit * builtin,
                      struct gl_shader *shader,
-                     const struct gl_extensions *extensions)
+                     const struct gl_extensions *extensions,
+                     struct gl_sl_pragmas *pragmas)
 {
    byte *prod;
    GLuint size, start, version;
@@ -2504,7 +2509,7 @@ compile_with_grammar(grammar id, const char *source, slang_code_unit * unit,
    /* Now preprocess the source string. */
    slang_string_init(&preprocessed);
    if (!_slang_preprocess_directives(&preprocessed, &source[start],
-                                     infolog, extensions)) {
+                                     infolog, extensions, pragmas)) {
       slang_string_free(&preprocessed);
       slang_info_log_error(infolog, "failed to preprocess the source.");
       return GL_FALSE;
@@ -2578,7 +2583,8 @@ static GLboolean
 compile_object(grammar * id, const char *source, slang_code_object * object,
                slang_unit_type type, slang_info_log * infolog,
                struct gl_shader *shader,
-               const struct gl_extensions *extensions)
+               const struct gl_extensions *extensions,
+               struct gl_sl_pragmas *pragmas)
 {
    slang_code_unit *builtins = NULL;
    GLuint base_version = 110;
@@ -2677,7 +2683,7 @@ compile_object(grammar * id, const char *source, slang_code_object * object,
 
    /* compile the actual shader - pass-in built-in library for external shader */
    return compile_with_grammar(*id, source, &object->unit, type, infolog,
-                               builtins, shader, extensions);
+                               builtins, shader, extensions, pragmas);
 }
 
 
@@ -2701,7 +2707,7 @@ compile_shader(GLcontext *ctx, slang_code_object * object,
    _slang_code_object_ctr(object);
 
    success = compile_object(&id, shader->Source, object, type, infolog, shader,
-                            &ctx->Extensions);
+                            &ctx->Extensions, &shader->Pragmas);
    if (id != 0)
       grammar_destroy(id);
    if (!success)
diff --git a/src/mesa/shader/slang/slang_emit.c b/src/mesa/shader/slang/slang_emit.c
index d3b4e64b78d..ea446fa5d49 100644
--- a/src/mesa/shader/slang/slang_emit.c
+++ b/src/mesa/shader/slang/slang_emit.c
@@ -1271,6 +1271,20 @@ emit_tex(slang_emit_info *emitInfo, slang_ir_node *n)
       opcode = OPCODE_TXP;
    }
 
+   if (n->Children[0]->Opcode == IR_ELEMENT) {
+      /* array is the sampler (a uniform which'll indicate the texture unit) */
+      assert(n->Children[0]->Children[0]->Store);
+      assert(n->Children[0]->Children[0]->Store->File == PROGRAM_SAMPLER);
+
+      emit(emitInfo, n->Children[0]);
+
+      n->Children[0]->Var = n->Children[0]->Children[0]->Var;
+   } else {
+      /* this is the sampler (a uniform which'll indicate the texture unit) */
+      assert(n->Children[0]->Store);
+      assert(n->Children[0]->Store->File == PROGRAM_SAMPLER);
+   }
+
    /* emit code for the texcoord operand */
    (void) emit(emitInfo, n->Children[1]);
 
@@ -1286,17 +1300,10 @@ emit_tex(slang_emit_info *emitInfo, slang_ir_node *n)
                            NULL,
                            NULL);
 
-   /* Child[0] is the sampler (a uniform which'll indicate the texture unit) */
-   assert(n->Children[0]->Store);
-   assert(n->Children[0]->Store->File == PROGRAM_SAMPLER);
-   /* Store->Index is the sampler index */
+   /* Store->Index is the uniform/sampler index */
    assert(n->Children[0]->Store->Index >= 0);
-   /* Store->Size is the texture target */
-   assert(n->Children[0]->Store->Size >= TEXTURE_1D_INDEX);
-   assert(n->Children[0]->Store->Size <= TEXTURE_RECT_INDEX);
-
-   inst->TexSrcTarget = n->Children[0]->Store->Size;
-   inst->TexSrcUnit = n->Children[0]->Store->Index; /* i.e. uniform's index */
+   inst->TexSrcUnit = n->Children[0]->Store->Index;
+   inst->TexSrcTarget = n->Children[0]->Store->TexTarget;
 
    /* mark the sampler as being used */
    _mesa_use_uniform(emitInfo->prog->Parameters,
@@ -1342,9 +1349,10 @@ emit_copy(slang_emit_info *emitInfo, slang_ir_node *n)
 
    if (n->Store->File == PROGRAM_SAMPLER) {
       /* no code generated for sampler assignments,
-       * just copy the sampler index at compile time.
+       * just copy the sampler index/target at compile time.
        */
       n->Store->Index = n->Children[1]->Store->Index;
+      n->Store->TexTarget = n->Children[1]->Store->TexTarget;
       return NULL;
    }
 
@@ -2370,10 +2378,20 @@ _slang_resolve_subroutines(slang_emit_info *emitInfo)
 
 
 
-
+/**
+ * Convert the IR tree into GPU instructions.
+ * \param n  root of IR tree
+ * \param vt  variable table
+ * \param prog  program to put GPU instructions into
+ * \param pragmas  controls codegen options
+ * \param withEnd  if true, emit END opcode at end
+ * \param log  log for emitting errors/warnings/info
+ */
 GLboolean
 _slang_emit_code(slang_ir_node *n, slang_var_table *vt,
-                 struct gl_program *prog, GLboolean withEnd,
+                 struct gl_program *prog,
+                 const struct gl_sl_pragmas *pragmas,
+                 GLboolean withEnd,
                  slang_info_log *log)
 {
    GET_CURRENT_CONTEXT(ctx);
@@ -2390,7 +2408,7 @@ _slang_emit_code(slang_ir_node *n, slang_var_table *vt,
 
    emitInfo.EmitHighLevelInstructions = ctx->Shader.EmitHighLevelInstructions;
    emitInfo.EmitCondCodes = ctx->Shader.EmitCondCodes;
-   emitInfo.EmitComments = ctx->Shader.EmitComments;
+   emitInfo.EmitComments = ctx->Shader.EmitComments || pragmas->Debug;
    emitInfo.EmitBeginEndSub = GL_TRUE;
 
    if (!emitInfo.EmitCondCodes) {
diff --git a/src/mesa/shader/slang/slang_emit.h b/src/mesa/shader/slang/slang_emit.h
index 59fb2fa890d..8ff52bf605d 100644
--- a/src/mesa/shader/slang/slang_emit.h
+++ b/src/mesa/shader/slang/slang_emit.h
@@ -46,7 +46,9 @@ _slang_var_swizzle(GLint size, GLint comp);
 
 extern GLboolean
 _slang_emit_code(slang_ir_node *n, slang_var_table *vartable,
-                 struct gl_program *prog, GLboolean withEnd,
+                 struct gl_program *prog,
+                 const struct gl_sl_pragmas *pragmas,
+                 GLboolean withEnd,
                  slang_info_log *log);
 
 
diff --git a/src/mesa/shader/slang/slang_ir.c b/src/mesa/shader/slang/slang_ir.c
index c33c80da999..e4c6e0ea516 100644
--- a/src/mesa/shader/slang/slang_ir.c
+++ b/src/mesa/shader/slang/slang_ir.c
@@ -1,8 +1,8 @@
 /*
  * Mesa 3-D graphics library
- * Version:  7.1
  *
  * Copyright (C) 2005-2008  Brian Paul   All Rights Reserved.
+ * Copyright (C) 2009  VMware, Inc.   All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -216,6 +216,26 @@ _slang_new_ir_storage_indirect(enum register_file file,
 }
 
 
+/**
+ * Allocate IR storage for a texture sampler.
+ * \param sampNum  the sampler number/index
+ * \param texTarget  one of TEXTURE_x_INDEX values
+ * \param size  number of samplers (in case of sampler array)
+ */
+slang_ir_storage *
+_slang_new_ir_storage_sampler(GLint sampNum, GLuint texTarget, GLint size)
+{
+   slang_ir_storage *st;
+   assert(texTarget < NUM_TEXTURE_TARGETS);
+   st = _slang_new_ir_storage(PROGRAM_SAMPLER, sampNum, size);
+   if (st) {
+      st->TexTarget = texTarget;
+   }
+   return st;
+}
+
+
+
 /* XXX temporary function */
 void
 _slang_copy_ir_storage(slang_ir_storage *dst, const slang_ir_storage *src)
diff --git a/src/mesa/shader/slang/slang_ir.h b/src/mesa/shader/slang/slang_ir.h
index a258e92e06f..644269d491c 100644
--- a/src/mesa/shader/slang/slang_ir.h
+++ b/src/mesa/shader/slang/slang_ir.h
@@ -1,8 +1,8 @@
 /*
  * Mesa 3-D graphics library
- * Version:  6.5.3
  *
- * Copyright (C) 2005-2007  Brian Paul   All Rights Reserved.
+ * Copyright (C) 2005-2008  Brian Paul   All Rights Reserved.
+ * Copyright (C) 2009  VMware, Inc.   All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -168,8 +168,8 @@ typedef enum
 struct slang_ir_storage_
 {
    enum register_file File;  /**< PROGRAM_TEMPORARY, PROGRAM_INPUT, etc */
-   GLint Index;  /**< -1 means unallocated */
-   GLint Size;  /**< number of floats */
+   GLint Index;    /**< -1 means unallocated */
+   GLint Size;     /**< number of floats or ints */
    GLuint Swizzle; /**< Swizzle AND writemask info */
    GLint RefCount; /**< Used during IR tree delete */
 
@@ -179,6 +179,7 @@ struct slang_ir_storage_
    enum register_file IndirectFile;
    GLint IndirectIndex;
    GLuint IndirectSwizzle;
+   GLuint TexTarget;  /**< If File==PROGRAM_SAMPLER, one of TEXTURE_x_INDEX */
 
    /** If Parent is non-null, Index is relative to parent.
     * The other fields are ignored.
@@ -254,6 +255,10 @@ _slang_new_ir_storage_indirect(enum register_file file,
                                GLint indirectIndex,
                                GLuint indirectSwizzle);
 
+extern slang_ir_storage *
+_slang_new_ir_storage_sampler(GLint sampNum, GLuint texTarget, GLint size);
+
+
 extern void
 _slang_copy_ir_storage(slang_ir_storage *dst, const slang_ir_storage *src);
 
diff --git a/src/mesa/shader/slang/slang_link.c b/src/mesa/shader/slang/slang_link.c
index a68cafc0865..b5862bda822 100644
--- a/src/mesa/shader/slang/slang_link.c
+++ b/src/mesa/shader/slang/slang_link.c
@@ -260,8 +260,8 @@ link_uniform_vars(GLcontext *ctx,
          GLuint newSampNum = *numSamplers;
          if (newSampNum >= ctx->Const.MaxTextureImageUnits) {
             char s[100];
-            sprintf(s, "Too many texture samplers (%u, max is %u)",
-                    newSampNum, ctx->Const.MaxTextureImageUnits);
+            _mesa_sprintf(s, "Too many texture samplers (%u, max is %u)",
+                          newSampNum, ctx->Const.MaxTextureImageUnits);
             link_error(shProg, s);
             return GL_FALSE;
          }
@@ -282,12 +282,14 @@ link_uniform_vars(GLcontext *ctx,
    for (i = 0; i < prog->NumInstructions; i++) {
       struct prog_instruction *inst = prog->Instructions + i;
       if (_mesa_is_tex_instruction(inst->Opcode)) {
-         /*
+         const GLint oldSampNum = inst->TexSrcUnit;
+
+#if 0
          printf("====== remap sampler from %d to %d\n",
-                inst->Sampler, map[ inst->Sampler ]);
-         */
+                inst->TexSrcUnit, samplerMap[ inst->TexSrcUnit ]);
+#endif
+
          /* here, texUnit is really samplerUnit */
-         const GLint oldSampNum = inst->TexSrcUnit;
          if (oldSampNum < Elements(samplerMap)) {
             const GLuint newSampNum = samplerMap[oldSampNum];
             inst->TexSrcUnit = newSampNum;
diff --git a/src/mesa/shader/slang/slang_preprocess.c b/src/mesa/shader/slang/slang_preprocess.c
index 76e757cb381..cd79c8b94a6 100644
--- a/src/mesa/shader/slang/slang_preprocess.c
+++ b/src/mesa/shader/slang/slang_preprocess.c
@@ -51,6 +51,9 @@ grammar_error_to_log (slang_info_log *log)
    GLint pos;
 
    grammar_get_last_error ((byte *) (buf), sizeof (buf), &pos);
+   if (buf[0] == 0) {
+      _mesa_snprintf(buf, sizeof(buf), "Preprocessor error");
+   }
    slang_info_log_error (log, buf);
 }
 
@@ -528,6 +531,56 @@ pp_ext_set(pp_ext *self, const char *name, GLboolean enable)
 }
 
 
+static void
+pp_pragmas_init(struct gl_sl_pragmas *pragmas)
+{
+   pragmas->Optimize = GL_TRUE;
+   pragmas->Debug = GL_FALSE;
+}
+
+
+/**
+ * Called in response to #pragma.  For example, "#pragma debug(on)" would
+ * call this function as pp_pragma("debug", "on").
+ * \return GL_TRUE if pragma is valid, GL_FALSE if invalid
+ */
+static GLboolean
+pp_pragma(struct gl_sl_pragmas *pragmas, const char *pragma, const char *param)
+{
+#if 0
+   printf("#pragma %s %s\n", pragma, param);
+#endif
+   if (_mesa_strcmp(pragma, "optimize") == 0) {
+      if (!param)
+         return GL_FALSE; /* missing required param */
+      if (_mesa_strcmp(param, "on") == 0) {
+         pragmas->Optimize = GL_TRUE;
+      }
+      else if (_mesa_strcmp(param, "off") == 0) {
+         pragmas->Optimize = GL_FALSE;
+      }
+      else {
+         return GL_FALSE; /* invalid param */
+      }
+   }
+   else if (_mesa_strcmp(pragma, "debug") == 0) {
+      if (!param)
+         return GL_FALSE; /* missing required param */
+      if (_mesa_strcmp(param, "on") == 0) {
+         pragmas->Debug = GL_TRUE;
+      }
+      else if (_mesa_strcmp(param, "off") == 0) {
+         pragmas->Debug = GL_FALSE;
+      }
+      else {
+         return GL_FALSE; /* invalid param */
+      }
+   }
+   /* all other pragmas are silently ignored */
+   return GL_TRUE;
+}
+
+
 /**
  * The state of preprocessor: current line, file and version number, list
  * of all defined macros and the #if/#endif context.
@@ -862,11 +915,16 @@ parse_if (slang_string *output, const byte *prod, GLuint *pi, GLint *result, pp_
 #define BEHAVIOR_WARN    3
 #define BEHAVIOR_DISABLE 4
 
+#define PRAGMA_NO_PARAM  0
+#define PRAGMA_PARAM     1
+
+
 static GLboolean
 preprocess_source (slang_string *output, const char *source,
                    grammar pid, grammar eid,
                    slang_info_log *elog,
-                   const struct gl_extensions *extensions)
+                   const struct gl_extensions *extensions,
+                   struct gl_sl_pragmas *pragmas)
 {
    static const char *predefined[] = {
       "__FILE__",
@@ -888,6 +946,7 @@ preprocess_source (slang_string *output, const char *source,
    }
 
    pp_state_init (&state, elog, extensions);
+   pp_pragmas_init (pragmas);
 
    /* add the predefined symbols to the symbol table */
    for (i = 0; predefined[i]; i++) {
@@ -940,9 +999,11 @@ preprocess_source (slang_string *output, const char *source,
       else {
          const char *id;
          GLuint idlen;
+         GLubyte token;
 
          i++;
-         switch (prod[i++]) {
+         token = prod[i++];
+         switch (token) {
 
          case TOKEN_END:
             /* End of source string.
@@ -1159,6 +1220,25 @@ preprocess_source (slang_string *output, const char *source,
             }
             break;
 
+         case TOKEN_PRAGMA:
+            {
+               GLint have_param;
+               const char *pragma, *param;
+
+               pragma = (const char *) (&prod[i]);
+               i += _mesa_strlen(pragma) + 1;
+               have_param = (prod[i++] == PRAGMA_PARAM);
+               if (have_param) {
+                  param = (const char *) (&prod[i]);
+                  i += _mesa_strlen(param) + 1;
+               }
+               else {
+                  param = NULL;
+               }
+               pp_pragma(pragmas, pragma, param);
+            }
+            break;
+
          case TOKEN_LINE:
             id = (const char *) (&prod[i]);
             i += _mesa_strlen (id) + 1;
@@ -1223,7 +1303,8 @@ GLboolean
 _slang_preprocess_directives(slang_string *output,
                              const char *input,
                              slang_info_log *elog,
-                             const struct gl_extensions *extensions)
+                             const struct gl_extensions *extensions,
+                             struct gl_sl_pragmas *pragmas)
 {
    grammar pid, eid;
    GLboolean success;
@@ -1239,7 +1320,7 @@ _slang_preprocess_directives(slang_string *output,
       grammar_destroy (pid);
       return GL_FALSE;
    }
-   success = preprocess_source (output, input, pid, eid, elog, extensions);
+   success = preprocess_source (output, input, pid, eid, elog, extensions, pragmas);
    grammar_destroy (eid);
    grammar_destroy (pid);
    return success;
diff --git a/src/mesa/shader/slang/slang_preprocess.h b/src/mesa/shader/slang/slang_preprocess.h
index dd996a6314b..f344820daef 100644
--- a/src/mesa/shader/slang/slang_preprocess.h
+++ b/src/mesa/shader/slang/slang_preprocess.h
@@ -1,8 +1,8 @@
 /*
  * Mesa 3-D graphics library
- * Version:  6.5.3
  *
- * Copyright (C) 2005-2007  Brian Paul   All Rights Reserved.
+ * Copyright (C) 2005-2008  Brian Paul   All Rights Reserved.
+ * Copyright (C) 2009 VMware, Inc.   All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -35,6 +35,7 @@ _slang_preprocess_version (const char *, GLuint *, GLuint *, slang_info_log *);
 extern GLboolean
 _slang_preprocess_directives(slang_string *output, const char *input,
                              slang_info_log *,
-                             const struct gl_extensions *extensions);
+                             const struct gl_extensions *extensions,
+                             struct gl_sl_pragmas *pragmas);
 
 #endif /* SLANG_PREPROCESS_H */
diff --git a/src/mesa/shader/slang/slang_typeinfo.c b/src/mesa/shader/slang/slang_typeinfo.c
index a5bcde404f6..1ef43f58c02 100644
--- a/src/mesa/shader/slang/slang_typeinfo.c
+++ b/src/mesa/shader/slang/slang_typeinfo.c
@@ -23,7 +23,7 @@
  */
 
 /**
- * \file slang_assemble_typeinfo.c
+ * \file slang_typeinfo.c
  * slang type info
  * \author Michal Krol
  */
diff --git a/src/mesa/state_tracker/st_cb_fbo.c b/src/mesa/state_tracker/st_cb_fbo.c
index eece7dee112..0c69e166233 100644
--- a/src/mesa/state_tracker/st_cb_fbo.c
+++ b/src/mesa/state_tracker/st_cb_fbo.c
@@ -354,11 +354,13 @@ st_render_texture(GLcontext *ctx,
 {
    struct st_renderbuffer *strb;
    struct gl_renderbuffer *rb;
-   struct pipe_texture *pt;
+   struct pipe_texture *pt = st_get_texobj_texture(att->Texture);
    struct st_texture_object *stObj;
    const struct gl_texture_image *texImage =
       att->Texture->Image[att->CubeMapFace][att->TextureLevel];
 
+   if (!pt) 
+      return;
 
    assert(!att->Renderbuffer);
 
@@ -387,8 +389,6 @@ st_render_texture(GLcontext *ctx,
    rb->Height = texImage->Height2;
    /*printf("***** render to texture level %d: %d x %d\n", att->TextureLevel, rb->Width, rb->Height);*/
 
-   pt = st_get_texobj_texture(att->Texture);
-   assert(pt);
    /*printf("***** pipe texture %d x %d\n", pt->width[0], pt->height[0]);*/
 
    pipe_texture_reference( &strb->texture, pt );
diff --git a/src/mesa/state_tracker/st_gen_mipmap.c b/src/mesa/state_tracker/st_gen_mipmap.c
index a15faf732ca..9e5c35072a1 100644
--- a/src/mesa/state_tracker/st_gen_mipmap.c
+++ b/src/mesa/state_tracker/st_gen_mipmap.c
@@ -160,9 +160,14 @@ st_generate_mipmap(GLcontext *ctx, GLenum target,
    struct st_context *st = ctx->st;
    struct pipe_texture *pt = st_get_texobj_texture(texObj);
    const uint baseLevel = texObj->BaseLevel;
-   const uint lastLevel = pt->last_level;
+   uint lastLevel;
    uint dstLevel;
 
+   if (!pt)
+      return;
+
+   lastLevel = pt->last_level;
+
    if (!st_render_mipmap(st, target, pt, baseLevel, lastLevel)) {
       fallback_generate_mipmap(ctx, target, texObj);
    }
diff --git a/src/mesa/state_tracker/wgl/stw_device.c b/src/mesa/state_tracker/wgl/stw_device.c
index e2a17d83ac6..129b24ce777 100644
--- a/src/mesa/state_tracker/wgl/stw_device.c
+++ b/src/mesa/state_tracker/wgl/stw_device.c
@@ -28,6 +28,8 @@
 #include <windows.h>
 
 #include "pipe/p_debug.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_screen.h"
 
 #include "stw_device.h"
 #include "stw_winsys.h"
@@ -37,6 +39,23 @@
 struct stw_device *stw_dev = NULL;
 
 
+/**
+ * XXX: Dispatch pipe_winsys::flush_front_buffer to our 
+ * stw_winsys::flush_front_buffer.
+ */
+static void 
+st_flush_frontbuffer(struct pipe_winsys *ws,
+                     struct pipe_surface *surf,
+                     void *context_private )
+{
+   const struct stw_winsys *stw_winsys = stw_dev->stw_winsys;
+   struct pipe_winsys *winsys = stw_dev->screen->winsys;
+   HDC hdc = (HDC)context_private;
+   
+   stw_winsys->flush_frontbuffer(winsys, surf, hdc);
+}
+
+
 boolean
 st_init(const struct stw_winsys *stw_winsys)
 {
@@ -53,6 +72,9 @@ st_init(const struct stw_winsys *stw_winsys)
    if(!stw_dev->screen)
       goto error1;
 
+   /* XXX: pipe_winsys::flush_frontbuffer should go away */
+   stw_dev->screen->winsys->flush_frontbuffer = st_flush_frontbuffer;
+   
    pixelformat_init();
 
    return TRUE;
diff --git a/src/mesa/state_tracker/wgl/stw_wgl_context.c b/src/mesa/state_tracker/wgl/stw_wgl_context.c
index 0c13c6b68ab..890d97fd72a 100644
--- a/src/mesa/state_tracker/wgl/stw_wgl_context.c
+++ b/src/mesa/state_tracker/wgl/stw_wgl_context.c
@@ -111,6 +111,9 @@ wglCreateContext(
       FREE( ctx );
       return NULL;
    }
+   
+   assert(!pipe->priv);
+   pipe->priv = hdc;
 
    ctx->st = st_create_context( pipe, visual, NULL );
    if (ctx->st == NULL) {
@@ -265,6 +268,8 @@ wglMakeCurrent(
    if (ctx && fb) {
       st_make_current( ctx->st, fb->stfb, fb->stfb );
       framebuffer_resize( fb, width, height );
+      ctx->hdc = hdc;
+      ctx->st->pipe->priv = hdc;
    }
    else {
       /* Detach */
diff --git a/src/libXvMC/Makefile b/src/xvmc/Makefile
index 7badcfd2645..7badcfd2645 100644
--- a/src/libXvMC/Makefile
+++ b/src/xvmc/Makefile
diff --git a/src/libXvMC/attributes.c b/src/xvmc/attributes.c
index 674524b8b82..674524b8b82 100644
--- a/src/libXvMC/attributes.c
+++ b/src/xvmc/attributes.c
diff --git a/src/libXvMC/block.c b/src/xvmc/block.c
index b38a89be092..b38a89be092 100644
--- a/src/libXvMC/block.c
+++ b/src/xvmc/block.c
diff --git a/src/libXvMC/context.c b/src/xvmc/context.c
index 273f6580292..273f6580292 100644
--- a/src/libXvMC/context.c
+++ b/src/xvmc/context.c
diff --git a/src/libXvMC/subpicture.c b/src/xvmc/subpicture.c
index c8f70c90d0c..c8f70c90d0c 100644
--- a/src/libXvMC/subpicture.c
+++ b/src/xvmc/subpicture.c
diff --git a/src/libXvMC/surface.c b/src/xvmc/surface.c
index 67c179e66d3..7c5f45bd346 100644
--- a/src/libXvMC/surface.c
+++ b/src/xvmc/surface.c
@@ -73,14 +73,13 @@ Status XvMCCreateSurface(Display *display, XvMCContext *context, XvMCSurface *su
 
 	assert(display == vlGetNativeDisplay(vlGetDisplay(vlContextGetScreen(vl_ctx))));
 
-	vlCreateSurface
-	(
-		vlContextGetScreen(vl_ctx),
-		context->width,
-		context->height,
-		vlGetPictureFormat(vl_ctx),
-		&vl_sfc
-	);
+	if (vlCreateSurface(vlContextGetScreen(vl_ctx),
+	                    context->width, context->height,
+	                    vlGetPictureFormat(vl_ctx),
+	                    &vl_sfc))
+	{
+		return BadAlloc;
+	}
 
 	vlBindToContext(vl_sfc, vl_ctx);
 
diff --git a/src/libXvMC/tests/.gitignore b/src/xvmc/tests/.gitignore
index e1d2f9023df..e1d2f9023df 100644
--- a/src/libXvMC/tests/.gitignore
+++ b/src/xvmc/tests/.gitignore
diff --git a/src/libXvMC/tests/Makefile b/src/xvmc/tests/Makefile
index de095161d29..de095161d29 100644
--- a/src/libXvMC/tests/Makefile
+++ b/src/xvmc/tests/Makefile
diff --git a/src/libXvMC/tests/test_blocks.c b/src/xvmc/tests/test_blocks.c
index dc80adfa652..dc80adfa652 100644
--- a/src/libXvMC/tests/test_blocks.c
+++ b/src/xvmc/tests/test_blocks.c
diff --git a/src/libXvMC/tests/test_context.c b/src/xvmc/tests/test_context.c
index 53f7449cd0b..53f7449cd0b 100644
--- a/src/libXvMC/tests/test_context.c
+++ b/src/xvmc/tests/test_context.c
diff --git a/src/libXvMC/tests/test_rendering.c b/src/xvmc/tests/test_rendering.c
index 1e7467a3aa6..1e7467a3aa6 100644
--- a/src/libXvMC/tests/test_rendering.c
+++ b/src/xvmc/tests/test_rendering.c
diff --git a/src/libXvMC/tests/test_surface.c b/src/xvmc/tests/test_surface.c
index 25ebdcc4fcf..25ebdcc4fcf 100644
--- a/src/libXvMC/tests/test_surface.c
+++ b/src/xvmc/tests/test_surface.c
diff --git a/src/libXvMC/tests/testlib.c b/src/xvmc/tests/testlib.c
index 59a03ca813f..59a03ca813f 100644
--- a/src/libXvMC/tests/testlib.c
+++ b/src/xvmc/tests/testlib.c
diff --git a/src/libXvMC/tests/testlib.h b/src/xvmc/tests/testlib.h
index af71ad74e17..af71ad74e17 100644
--- a/src/libXvMC/tests/testlib.h
+++ b/src/xvmc/tests/testlib.h
diff --git a/src/libXvMC/tests/xvmc_bench.c b/src/xvmc/tests/xvmc_bench.c
index d5a39ecf17d..d5a39ecf17d 100644
--- a/src/libXvMC/tests/xvmc_bench.c
+++ b/src/xvmc/tests/xvmc_bench.c