Merge commit 'origin/master' into gallium-msaa

35 files changed, 725 insertions, 362 deletions

diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index f8e65cf6c61..2daed382cf3 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -101,6 +101,7 @@ C_SOURCES = \
 	util/u_blit.c \
 	util/u_blitter.c \
 	util/u_cache.c \
+	util/u_caps.c \
 	util/u_cpu_detect.c \
 	util/u_dl.c \
 	util/u_draw_quad.c \
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index db3a1e73114..a0673df8a8e 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -144,6 +144,7 @@ source = [
     'util/u_blit.c',
     'util/u_blitter.c',
     'util/u_cache.c',
+    'util/u_caps.c',
     'util/u_cpu_detect.c',
     'util/u_debug.c',
     'util/u_debug_memory.c',
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index 030683f60f1..55e0ded1b30 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -314,10 +314,13 @@ void cso_release_all( struct cso_context *ctx )
 }
 
 
+/**
+ * Free the CSO context.  NOTE: the state tracker should have previously called
+ * cso_release_all().
+ */
 void cso_destroy_context( struct cso_context *ctx )
 {
    if (ctx) {
-      /*cso_release_all( ctx );*/
       FREE( ctx );
    }
 }
diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index 703b46b518b..3b2df054c3c 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -24,6 +24,8 @@
 /* generates the draw jit function */
 static void
 draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *var);
+static void
+draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *var);
 
 static void
 init_globals(struct draw_llvm *llvm)
@@ -218,6 +220,7 @@ draw_llvm_prepare(struct draw_llvm *llvm, int num_inputs)
    llvm->vertex_header_ptr_type = create_vertex_header(llvm, num_inputs);
 
    draw_llvm_generate(llvm, variant);
+   draw_llvm_generate_elts(llvm, variant);
 
    return variant;
 }
@@ -696,6 +699,158 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
       lp_disassemble(variant->jit_func);
 }
 
+
+static void
+draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
+{
+   LLVMTypeRef arg_types[7];
+   LLVMTypeRef func_type;
+   LLVMValueRef context_ptr;
+   LLVMBasicBlockRef block;
+   LLVMBuilderRef builder;
+   LLVMValueRef fetch_elts, fetch_count, stride, step, io_itr;
+   LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr;
+   struct draw_context *draw = llvm->draw;
+   unsigned i, j;
+   struct lp_build_context bld;
+   struct lp_build_context bld_int;
+   struct lp_build_loop_state lp_loop;
+   struct lp_type vs_type = lp_type_float_vec(32);
+   const int max_vertices = 4;
+   LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
+   LLVMValueRef fetch_max;
+
+   arg_types[0] = llvm->context_ptr_type;               /* context */
+   arg_types[1] = llvm->vertex_header_ptr_type;         /* vertex_header */
+   arg_types[2] = llvm->buffer_ptr_type;                /* vbuffers */
+   arg_types[3] = LLVMPointerType(LLVMInt32Type(), 0);  /* fetch_elts * */
+   arg_types[4] = LLVMInt32Type();                      /* fetch_count */
+   arg_types[5] = LLVMInt32Type();                      /* stride */
+   arg_types[6] = llvm->vb_ptr_type;                    /* pipe_vertex_buffer's */
+
+   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+
+   variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", func_type);
+   LLVMSetFunctionCallConv(variant->function_elts, LLVMCCallConv);
+   for(i = 0; i < Elements(arg_types); ++i)
+      if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
+         LLVMAddAttribute(LLVMGetParam(variant->function_elts, i), LLVMNoAliasAttribute);
+
+   context_ptr  = LLVMGetParam(variant->function_elts, 0);
+   io_ptr       = LLVMGetParam(variant->function_elts, 1);
+   vbuffers_ptr = LLVMGetParam(variant->function_elts, 2);
+   fetch_elts   = LLVMGetParam(variant->function_elts, 3);
+   fetch_count  = LLVMGetParam(variant->function_elts, 4);
+   stride       = LLVMGetParam(variant->function_elts, 5);
+   vb_ptr       = LLVMGetParam(variant->function_elts, 6);
+
+   lp_build_name(context_ptr, "context");
+   lp_build_name(io_ptr, "io");
+   lp_build_name(vbuffers_ptr, "vbuffers");
+   lp_build_name(fetch_elts, "fetch_elts");
+   lp_build_name(fetch_count, "fetch_count");
+   lp_build_name(stride, "stride");
+   lp_build_name(vb_ptr, "vb");
+
+   /*
+    * Function body
+    */
+
+   block = LLVMAppendBasicBlock(variant->function_elts, "entry");
+   builder = LLVMCreateBuilder();
+   LLVMPositionBuilderAtEnd(builder, block);
+
+   lp_build_context_init(&bld, builder, vs_type);
+   lp_build_context_init(&bld_int, builder, lp_type_int(32));
+
+   step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0);
+
+   fetch_max = LLVMBuildSub(builder, fetch_count,
+                            LLVMConstInt(LLVMInt32Type(), 1, 0),
+                            "fetch_max");
+
+   lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), &lp_loop);
+   {
+      LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
+      LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
+      LLVMValueRef io;
+      const LLVMValueRef (*ptr_aos)[NUM_CHANNELS];
+
+      io_itr = lp_loop.counter;
+      io = LLVMBuildGEP(builder, io_ptr, &io_itr, 1, "");
+#if DEBUG_STORE
+      lp_build_printf(builder, " --- io %d = %p, loop counter %d\n",
+                      io_itr, io, lp_loop.counter);
+#endif
+      for (i = 0; i < NUM_CHANNELS; ++i) {
+         LLVMValueRef true_index = LLVMBuildAdd(
+            builder,
+            lp_loop.counter,
+            LLVMConstInt(LLVMInt32Type(), i, 0), "");
+         LLVMValueRef fetch_ptr;
+
+         /* make sure we're not out of bounds which can happen
+          * if fetch_count % 4 != 0, because on the last iteration
+          * a few of the 4 vertex fetches will be out of bounds */
+         true_index = lp_build_min(&bld_int, true_index, fetch_max);
+
+         fetch_ptr = LLVMBuildGEP(builder, fetch_elts,
+                                  &true_index, 1, "");
+         true_index = LLVMBuildLoad(builder, fetch_ptr, "fetch_elt");
+         for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
+            struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
+            LLVMValueRef vb_index = LLVMConstInt(LLVMInt32Type(),
+                                                 velem->vertex_buffer_index,
+                                                 0);
+            LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr,
+                                           &vb_index, 1, "");
+            generate_fetch(builder, vbuffers_ptr,
+                           &aos_attribs[j][i], velem, vb, true_index);
+         }
+      }
+      convert_to_soa(builder, aos_attribs, inputs,
+                     draw->pt.nr_vertex_elements);
+
+      ptr_aos = (const LLVMValueRef (*)[NUM_CHANNELS]) inputs;
+      generate_vs(llvm,
+                  builder,
+                  outputs,
+                  ptr_aos,
+                  context_ptr);
+
+      convert_to_aos(builder, io, outputs,
+                     draw->vs.vertex_shader->info.num_outputs,
+                     max_vertices);
+   }
+   lp_build_loop_end_cond(builder, fetch_count, step, LLVMIntUGE, &lp_loop);
+
+   LLVMBuildRetVoid(builder);
+
+   LLVMDisposeBuilder(builder);
+
+   /*
+    * Translate the LLVM IR into machine code.
+    */
+#ifdef DEBUG
+   if(LLVMVerifyFunction(variant->function_elts, LLVMPrintMessageAction)) {
+      LLVMDumpValue(variant->function_elts);
+      assert(0);
+   }
+#endif
+
+   LLVMRunFunctionPassManager(llvm->pass, variant->function_elts);
+
+   if (0) {
+      LLVMDumpValue(variant->function_elts);
+      debug_printf("\n");
+   }
+   variant->jit_func_elts = (draw_jit_vert_func_elts)LLVMGetPointerToGlobal(
+      llvm->draw->engine, variant->function_elts);
+
+   if (0)
+      lp_disassemble(variant->jit_func_elts);
+}
+
 void
 draw_llvm_make_variant_key(struct draw_llvm *llvm,
                            struct draw_llvm_variant_key *key)
diff --git a/src/gallium/auxiliary/draw/draw_llvm.h b/src/gallium/auxiliary/draw/draw_llvm.h
index 20f4f3da5f3..58fee7f9d60 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.h
+++ b/src/gallium/auxiliary/draw/draw_llvm.h
@@ -94,6 +94,16 @@ typedef void
                       unsigned stride,
                       struct pipe_vertex_buffer *vertex_buffers);
 
+
+typedef void
+(*draw_jit_vert_func_elts)(struct draw_jit_context *context,
+                           struct vertex_header *io,
+                           const char *vbuffers[PIPE_MAX_ATTRIBS],
+                           const unsigned *fetch_elts,
+                           unsigned fetch_count,
+                           unsigned stride,
+                           struct pipe_vertex_buffer *vertex_buffers);
+
 struct draw_llvm {
    struct draw_context *draw;
 
@@ -122,7 +132,9 @@ struct draw_llvm_variant
 {
    struct draw_llvm_variant_key key;
    LLVMValueRef function;
+   LLVMValueRef function_elts;
    draw_jit_vert_func jit_func;
+   draw_jit_vert_func_elts jit_func_elts;
 
    struct draw_llvm_variant *next;
 };
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 1d9395e06c3..b853f3a89f8 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -38,10 +38,10 @@
 #include "util/u_prim.h"
 
 
-DEBUG_GET_ONCE_BOOL_OPTION(draw_fse, "DRAW_FSE", FALSE);
-DEBUG_GET_ONCE_BOOL_OPTION(draw_no_fse, "DRAW_NO_FSE", FALSE);
+DEBUG_GET_ONCE_BOOL_OPTION(draw_fse, "DRAW_FSE", FALSE)
+DEBUG_GET_ONCE_BOOL_OPTION(draw_no_fse, "DRAW_NO_FSE", FALSE)
 #ifdef HAVE_LLVM
-DEBUG_GET_ONCE_BOOL_OPTION(draw_use_llvm, "DRAW_USE_LLVM", TRUE);
+DEBUG_GET_ONCE_BOOL_OPTION(draw_use_llvm, "DRAW_USE_LLVM", TRUE)
 #endif
 
 static unsigned trim( unsigned count, unsigned first, unsigned incr )
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
index f71271bd915..d2a492f2b4c 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
@@ -167,8 +167,6 @@ static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
 {
    struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
    struct draw_context *draw = fpme->draw;
-   struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
-   struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
    unsigned opt = fpme->opt;
    unsigned alloc_count = align( fetch_count, 4 );
 
@@ -182,35 +180,13 @@ static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
       return;
    }
 
-   /* Fetch into our vertex buffer
-    */
-   draw_pt_fetch_run( fpme->fetch,
-		      fetch_elts,
-		      fetch_count,
-		      (char *)pipeline_verts );
-
-   /* Run the shader, note that this overwrites the data[] parts of
-    * the pipeline verts.  If there is no shader, eg if
-    * bypass_vs_clip_and_viewport, then the inputs == outputs, and are
-    * already in the correct place.*/
-   if (opt & PT_SHADE)
-   {
-      vshader->run_linear(vshader,
-                          (const float (*)[4])pipeline_verts->data,
-                          (      float (*)[4])pipeline_verts->data,
-                          draw->pt.user.vs_constants,
-                          fetch_count,
-                          fpme->vertex_size,
-                          fpme->vertex_size);
-      if (gshader)
-         draw_geometry_shader_run(gshader,
-                                  (const float (*)[4])pipeline_verts->data,
-                                  (      float (*)[4])pipeline_verts->data,
-                                  draw->pt.user.gs_constants,
-                                  fetch_count,
-                                  fpme->vertex_size,
-                                  fpme->vertex_size);
-   }
+   fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
+                                         pipeline_verts,
+                                         (const char **)draw->pt.user.vbuffer,
+                                         fetch_elts,
+                                         fetch_count,
+                                         fpme->vertex_size,
+                                         draw->pt.vertex_buffer );
 
    if (draw_pt_post_vs_run( fpme->post_vs,
 			    pipeline_verts,
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index 0c059573505..c2832eefa2a 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -46,7 +46,7 @@
 #include "tgsi/tgsi_dump.h"
 #include "tgsi/tgsi_exec.h"
 
-DEBUG_GET_ONCE_BOOL_OPTION(gallium_dump_vs, "GALLIUM_DUMP_VS", FALSE);
+DEBUG_GET_ONCE_BOOL_OPTION(gallium_dump_vs, "GALLIUM_DUMP_VS", FALSE)
 
 void
 draw_vs_set_constants(struct draw_context *draw,
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index efead42335f..6c7e94db433 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -169,8 +169,9 @@ draw_create_vs_ppc(struct draw_context *draw,
 struct draw_vs_varient_key;
 struct draw_vertex_shader;
 
-struct draw_vs_varient *draw_vs_varient_aos_sse( struct draw_vertex_shader *vs,
-                                                 const struct draw_vs_varient_key *key );
+struct draw_vs_varient *
+draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs,
+                                const struct draw_vs_varient_key *key );
 
 
 
@@ -188,8 +189,9 @@ struct translate *draw_vs_get_fetch( struct draw_context *draw,
 struct translate *draw_vs_get_emit( struct draw_context *draw,
                                     struct translate_key *key );
 
-struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
-                                                 const struct draw_vs_varient_key *key );
+struct draw_vs_varient *
+draw_vs_create_varient_generic( struct draw_vertex_shader *vs,
+                                const struct draw_vs_varient_key *key );
 
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index e7121f36541..19f49e34c8b 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -2089,13 +2089,21 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 }
 
 
+/** cast wrapper */
+static INLINE struct draw_vs_varient_aos_sse *
+draw_vs_varient_aos_sse(struct draw_vs_varient *varient)
+{
+   return (struct draw_vs_varient_aos_sse *) varient;
+}
+
 
 static void vaos_set_buffer( struct draw_vs_varient *varient,
                              unsigned buf,
                              const void *ptr,
-                             unsigned stride )
+                             unsigned stride,
+                             unsigned max_stride)
 {
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
 
    if (buf < vaos->nr_vb) {
       vaos->buffer[buf].base_ptr = (char *)ptr;
@@ -2112,7 +2120,7 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
                                       unsigned count,
                                       void *output_buffer )
 {
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
    unsigned i;
 
@@ -2136,7 +2144,7 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
                                         unsigned count,
                                         void *output_buffer )
 {
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
    unsigned i;
 
@@ -2165,7 +2173,7 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
 
 static void vaos_destroy( struct draw_vs_varient *varient )
 {
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct draw_vs_varient_aos_sse *vaos = draw_vs_varient_aos_sse(varient);
 
    FREE( vaos->buffer );
 
@@ -2241,13 +2249,14 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
 }
 
 
-struct draw_vs_varient *draw_vs_varient_aos_sse( struct draw_vertex_shader *vs,
-                                                 const struct draw_vs_varient_key *key )
+struct draw_vs_varient *
+draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs,
+                                const struct draw_vs_varient_key *key )
 {
    struct draw_vs_varient *varient = varient_aos_sse( vs, key );
 
    if (varient == NULL) {
-      varient = draw_vs_varient_generic( vs, key );
+      varient = draw_vs_create_varient_generic( vs, key );
    }
 
    return varient;
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index 7deca2b69d9..bc34d390dae 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -203,7 +203,7 @@ draw_create_vs_exec(struct draw_context *draw,
    vs->base.prepare = vs_exec_prepare;
    vs->base.run_linear = vs_exec_run_linear;
    vs->base.delete = vs_exec_delete;
-   vs->base.create_varient = draw_vs_varient_generic;
+   vs->base.create_varient = draw_vs_create_varient_generic;
    vs->machine = draw->vs.machine;
 
    return &vs->base;
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
index d869eecec5e..5df84916c51 100644
--- a/src/gallium/auxiliary/draw/draw_vs_ppc.c
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -125,7 +125,7 @@ vs_ppc_run_linear( struct draw_vertex_shader *base,
        */
       shader->func(inputs_soa, outputs_soa, temps_soa,
 		   (float (*)[4]) shader->base.immediates,
-                   (const float (*)[4])constants[0],
+                   (float (*)[4])constants[0],
                    ppc_builtin_constants);
 
       /* convert (up to) four output verts from SoA back to AoS format */
@@ -190,7 +190,7 @@ draw_create_vs_ppc(struct draw_context *draw,
       vs->base.create_varient = draw_vs_varient_aos_ppc;
    else
 #endif
-      vs->base.create_varient = draw_vs_varient_generic;
+      vs->base.create_varient = draw_vs_create_varient_generic;
    vs->base.prepare = vs_ppc_prepare;
    vs->base.run_linear = vs_ppc_run_linear;
    vs->base.delete = vs_ppc_delete;
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 54e6423388f..14c95082a9d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -165,9 +165,9 @@ draw_create_vs_sse(struct draw_context *draw,
 
    vs->base.draw = draw;
    if (1)
-      vs->base.create_varient = draw_vs_varient_aos_sse;
+      vs->base.create_varient = draw_vs_create_varient_aos_sse;
    else
-      vs->base.create_varient = draw_vs_varient_generic;
+      vs->base.create_varient = draw_vs_create_varient_generic;
    vs->base.prepare = vs_sse_prepare;
    vs->base.run_linear = vs_sse_run_linear;
    vs->base.delete = vs_sse_delete;
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 0abd827b615..6eb26927f27 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -263,8 +263,9 @@ static void vsvg_destroy( struct draw_vs_varient *varient )
 }
 
 
-struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
-                                                 const struct draw_vs_varient_key *key )
+struct draw_vs_varient *
+draw_vs_create_varient_generic( struct draw_vertex_shader *vs,
+                                const struct draw_vs_varient_key *key )
 {
    unsigned i;
    struct translate_key fetch, emit;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
index 5f9d4c62d1c..20ae958714b 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
@@ -1210,6 +1210,14 @@ LLVMValueRef
 lp_build_cos(struct lp_build_context *bld,
               LLVMValueRef a)
 {
+#ifdef PIPE_OS_WINDOWS
+   /*
+    * FIXME: X86 backend translates llvm.cos.v4f32 to 4 calls to CRT's cosf()
+    * which is neither efficient nor does the CRT linkage work on Windows
+    * causing segmentation fault. So simply disable the code for now.
+    */
+   return bld->one;
+#else
    const struct lp_type type = bld->type;
    LLVMTypeRef vec_type = lp_build_vec_type(type);
    char intrinsic[32];
@@ -1220,6 +1228,7 @@ lp_build_cos(struct lp_build_context *bld,
    util_snprintf(intrinsic, sizeof intrinsic, "llvm.cos.v%uf%u", type.length, type.width);
 
    return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a);
+#endif
 }
 
 
@@ -1230,6 +1239,14 @@ LLVMValueRef
 lp_build_sin(struct lp_build_context *bld,
               LLVMValueRef a)
 {
+#ifdef PIPE_OS_WINDOWS
+   /*
+    * FIXME: X86 backend translates llvm.sin.v4f32 to 4 calls to CRT's sinf()
+    * which is neither efficient nor does the CRT linkage work on Windows
+    * causing segmentation fault. So simply disable the code for now.
+    */
+   return bld->zero;
+#else
    const struct lp_type type = bld->type;
    LLVMTypeRef vec_type = lp_build_vec_type(type);
    char intrinsic[32];
@@ -1240,6 +1257,7 @@ lp_build_sin(struct lp_build_context *bld,
    util_snprintf(intrinsic, sizeof intrinsic, "llvm.sin.v%uf%u", type.length, type.width);
 
    return lp_build_intrinsic_unary(bld->builder, intrinsic, vec_type, a);
+#endif
 }
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.c b/src/gallium/auxiliary/gallivm/lp_bld_flow.c
index e60ab4f6ba1..8f15b1d287d 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_flow.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.c
@@ -792,3 +792,78 @@ lp_build_endif(struct lp_build_if_state *ctx)
    /* Resume building code at end of the ifthen->merge_block */
    LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
 }
+
+
+/**
+ * Allocate a scalar (or vector) variable.
+ *
+ * Although not strictly part of control flow, control flow has deep impact in
+ * how variables should be allocated.
+ *
+ * The mem2reg optimization pass is the recommended way to dealing with mutable
+ * variables, and SSA. It looks for allocas and if it can handle them, it
+ * promotes them, but only looks for alloca instructions in the entry block of
+ * the function. Being in the entry block guarantees that the alloca is only
+ * executed once, which makes analysis simpler.
+ *
+ * See also:
+ * - http://www.llvm.org/docs/tutorial/OCamlLangImpl7.html#memory
+ */
+LLVMValueRef
+lp_build_alloca(LLVMBuilderRef builder,
+                LLVMTypeRef type,
+                const char *name)
+{
+   LLVMBasicBlockRef current_block = LLVMGetInsertBlock(builder);
+   LLVMValueRef function = LLVMGetBasicBlockParent(current_block);
+   LLVMBasicBlockRef first_block = LLVMGetEntryBasicBlock(function);
+   LLVMValueRef first_instr = LLVMGetFirstInstruction(first_block);
+   LLVMBuilderRef first_builder = LLVMCreateBuilder();
+   LLVMValueRef res;
+
+   LLVMPositionBuilderAtEnd(first_builder, first_block);
+   LLVMPositionBuilderBefore(first_builder, first_instr);
+
+   res = LLVMBuildAlloca(first_builder, type, name);
+
+   LLVMDisposeBuilder(first_builder);
+
+   return res;
+}
+
+
+/**
+ * Allocate an array of scalars/vectors.
+ *
+ * mem2reg pass is not capable of promoting structs or arrays to registers, but
+ * we still put it in the first block anyway as failure to put allocas in the
+ * first block may prevent the X86 backend from successfully align the stack as
+ * required.
+ *
+ * Also the scalarrepl pass is supossedly more powerful and can promote
+ * arrays in many cases.
+ *
+ * See also:
+ * - http://www.llvm.org/docs/tutorial/OCamlLangImpl7.html#memory
+ */
+LLVMValueRef
+lp_build_array_alloca(LLVMBuilderRef builder,
+                      LLVMTypeRef type,
+                      LLVMValueRef count,
+                      const char *name)
+{
+   LLVMBasicBlockRef current_block = LLVMGetInsertBlock(builder);
+   LLVMValueRef function = LLVMGetBasicBlockParent(current_block);
+   LLVMBasicBlockRef first_block = LLVMGetEntryBasicBlock(function);
+   LLVMValueRef first_instr = LLVMGetFirstInstruction(first_block);
+   LLVMBuilderRef first_builder = LLVMCreateBuilder();
+   LLVMValueRef res;
+
+   LLVMPositionBuilderBefore(first_builder, first_instr);
+
+   res = LLVMBuildArrayAlloca(first_builder, type, count, name);
+
+   LLVMDisposeBuilder(first_builder);
+
+   return res;
+}
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.h b/src/gallium/auxiliary/gallivm/lp_bld_flow.h
index 745838570c8..fffb493a93b 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_flow.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.h
@@ -156,5 +156,15 @@ lp_build_endif(struct lp_build_if_state *ctx);
 LLVMBasicBlockRef
 lp_build_insert_new_block(LLVMBuilderRef builder, const char *name);
 
+LLVMValueRef
+lp_build_alloca(LLVMBuilderRef builder,
+                LLVMTypeRef type,
+                const char *name);
+
+LLVMValueRef
+lp_build_array_alloca(LLVMBuilderRef builder,
+                      LLVMTypeRef type,
+                      LLVMValueRef count,
+                      const char *name);
 
 #endif /* !LP_BLD_FLOW_H */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c
index a5a019fa92a..6257e9a4047 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format_aos.c
@@ -40,6 +40,7 @@
 
 #include "lp_bld_init.h"
 #include "lp_bld_type.h"
+#include "lp_bld_flow.h"
 #include "lp_bld_format.h"
 
 
@@ -370,11 +371,7 @@ lp_build_fetch_rgba_aos(LLVMBuilderRef builder,
          LLVMAddGlobalMapping(lp_build_engine, function, format_desc->fetch_rgba_float);
       }
 
-      /*
-       * XXX: this should better go to the first block in the function
-       */
-
-      tmp = LLVMBuildAlloca(builder, LLVMVectorType(LLVMFloatType(), 4), "");
+      tmp = lp_build_alloca(builder, LLVMVectorType(LLVMFloatType(), 4), "");
 
       /*
        * Invoke format_desc->fetch_rgba_float() for each pixel and insert the result
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
index a3b69701162..d13fa1a5d04 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
@@ -472,18 +472,6 @@ lp_build_select_aos(struct lp_build_context *bld,
    }
 }
 
-LLVMValueRef
-lp_build_alloca(struct lp_build_context *bld)
-{
-   const struct lp_type type = bld->type;
-
-   if (type.length > 1) { /*vector*/
-      return LLVMBuildAlloca(bld->builder, lp_build_vec_type(type), "");
-   } else { /*scalar*/
-      return LLVMBuildAlloca(bld->builder, lp_build_elem_type(type), "");
-   }
-}
-
 
 /** Return (a & ~b) */
 LLVMValueRef
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.h b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
index 00a8c750196..29f9fc3b205 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
@@ -76,9 +76,6 @@ lp_build_select_aos(struct lp_build_context *bld,
                     LLVMValueRef b,
                     const boolean cond[4]);
 
-LLVMValueRef
-lp_build_alloca(struct lp_build_context *bld);
-
 
 LLVMValueRef
 lp_build_andc(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
index 54ef921678d..74dc9e1d812 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
@@ -1830,6 +1830,11 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld,
    LLVMValueRef unswizzled[4];
    LLVMValueRef stride;
 
+   assert(bld->static_state->target == PIPE_TEXTURE_2D);
+   assert(bld->static_state->min_img_filter == PIPE_TEX_FILTER_LINEAR);
+   assert(bld->static_state->mag_img_filter == PIPE_TEX_FILTER_LINEAR);
+   assert(bld->static_state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE);
+
    lp_build_context_init(&i32, builder, lp_type_int_vec(32));
    lp_build_context_init(&h16, builder, lp_type_ufixed(16));
    lp_build_context_init(&u8n, builder, lp_type_unorm(8));
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index fac90c60064..d3c769e28b8 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -744,22 +744,11 @@ emit_declaration(
    struct lp_build_tgsi_soa_context *bld,
    const struct tgsi_full_declaration *decl)
 {
+   LLVMTypeRef vec_type = lp_build_vec_type(bld->base.type);
+
    unsigned first = decl->Range.First;
    unsigned last = decl->Range.Last;
    unsigned idx, i;
-   LLVMBasicBlockRef current_block =
-      LLVMGetInsertBlock(bld->base.builder);
-   LLVMBasicBlockRef first_block =
-      LLVMGetEntryBasicBlock(
-         LLVMGetBasicBlockParent(current_block));
-   LLVMValueRef first_inst =
-      LLVMGetFirstInstruction(first_block);
-
-   /* we want alloca's to be the first instruction
-    * in the function so we need to rewind the builder
-    * to the very beginning */
-   LLVMPositionBuilderBefore(bld->base.builder,
-                             first_inst);
 
    for (idx = first; idx <= last; ++idx) {
       switch (decl->Declaration.File) {
@@ -767,23 +756,25 @@ emit_declaration(
          if (bld->has_indirect_addressing) {
             LLVMValueRef val = LLVMConstInt(LLVMInt32Type(),
                                             last*4 + 4, 0);
-            bld->temps_array = LLVMBuildArrayAlloca(bld->base.builder,
-                                                    lp_build_vec_type(bld->base.type),
-                                                    val, "");
+            bld->temps_array = lp_build_array_alloca(bld->base.builder,
+                                                     vec_type, val, "");
          } else {
             for (i = 0; i < NUM_CHANNELS; i++)
-               bld->temps[idx][i] = lp_build_alloca(&bld->base);
+               bld->temps[idx][i] = lp_build_alloca(bld->base.builder,
+                                                    vec_type, "");
          }
          break;
 
       case TGSI_FILE_OUTPUT:
          for (i = 0; i < NUM_CHANNELS; i++)
-            bld->outputs[idx][i] = lp_build_alloca(&bld->base);
+            bld->outputs[idx][i] = lp_build_alloca(bld->base.builder,
+                                                   vec_type, "");
          break;
 
       case TGSI_FILE_ADDRESS:
          for (i = 0; i < NUM_CHANNELS; i++)
-            bld->addr[idx][i] = lp_build_alloca(&bld->base);
+            bld->addr[idx][i] = lp_build_alloca(bld->base.builder,
+                                                vec_type, "");
          break;
 
       default:
@@ -792,8 +783,6 @@ emit_declaration(
       }
    }
 
-   LLVMPositionBuilderAtEnd(bld->base.builder,
-                            current_block);
    return TRUE;
 }
 
@@ -1600,18 +1589,10 @@ emit_instruction(
       lp_exec_mask_cond_push(&bld->exec_mask, tmp0);
       break;
 
-   case TGSI_OPCODE_BGNFOR:
-      /* fall through */
    case TGSI_OPCODE_BGNLOOP:
       lp_exec_bgnloop(&bld->exec_mask);
       break;
 
-   case TGSI_OPCODE_REP:
-      /* deprecated */
-      assert(0);
-      return FALSE;
-      break;
-
    case TGSI_OPCODE_ELSE:
       lp_exec_mask_cond_invert(&bld->exec_mask);
       break;
@@ -1620,18 +1601,10 @@ emit_instruction(
       lp_exec_mask_cond_pop(&bld->exec_mask);
       break;
 
-   case TGSI_OPCODE_ENDFOR:
-      /* fall-through */
    case TGSI_OPCODE_ENDLOOP:
       lp_exec_endloop(&bld->exec_mask);
       break;
 
-   case TGSI_OPCODE_ENDREP:
-      /* deprecated */
-      assert(0);
-      return FALSE;
-      break;
-
    case TGSI_OPCODE_PUSHA:
       /* deprecated? */
       assert(0);
diff --git a/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt b/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt
index 080fd4c7310..5d9eed92580 100644
--- a/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt
+++ b/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt
@@ -661,25 +661,6 @@ TGSI Instruction Specification
   TBD
 
 
-1.9.8  BGNFOR - Begin a For-Loop
-
-  dst.x = floor(src.x)
-  dst.y = floor(src.y)
-  dst.z = floor(src.z)
-
-  if (dst.y <= 0)
-    pc = [matching ENDFOR] + 1
-  endif
-
-  Note: The destination must be a loop register.
-        The source must be a constant register.
-
-
-1.9.9  REP - Repeat
-
-  TBD
-
-
 1.9.10  ELSE - Else
 
   TBD
@@ -690,23 +671,6 @@ TGSI Instruction Specification
   TBD
 
 
-1.9.12  ENDFOR - End a For-Loop
-
-  dst.x = dst.x + dst.z
-  dst.y = dst.y - 1.0
-
-  if (dst.y > 0)
-    pc = [matching BGNFOR instruction] + 1
-  endif
-
-  Note: The destination must be a loop register.
-
-
-1.9.13  ENDREP - End Repeat
-
-  TBD
-
-
 1.10  GL_NV_vertex_program3
 ---------------------------
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index 57031419f8e..83000200189 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -586,7 +586,6 @@ iter_instruction(
    /* update indentation */
    if (inst->Instruction.Opcode == TGSI_OPCODE_IF ||
        inst->Instruction.Opcode == TGSI_OPCODE_ELSE ||
-       inst->Instruction.Opcode == TGSI_OPCODE_BGNFOR ||
        inst->Instruction.Opcode == TGSI_OPCODE_BGNLOOP) {
       ctx->indentation += indent_spaces;
    }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 11045e4ba2f..82eac05dc4d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -3186,14 +3186,6 @@ exec_instruction(
       *pc = -1;
       break;
 
-   case TGSI_OPCODE_REP:
-      assert (0);
-      break;
-
-   case TGSI_OPCODE_ENDREP:
-       assert (0);
-       break;
-
    case TGSI_OPCODE_PUSHA:
       assert (0);
       break;
@@ -3258,29 +3250,6 @@ exec_instruction(
       emit_primitive(mach);
       break;
 
-   case TGSI_OPCODE_BGNFOR:
-      assert(mach->LoopCounterStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
-      for (chan_index = 0; chan_index < 3; chan_index++) {
-         FETCH( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[chan_index], 0, chan_index );
-      }
-      ++mach->LoopCounterStackTop;
-      STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 0, CHAN_X);
-      /* update LoopMask */
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) {
-         mach->LoopMask &= ~0x1;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) {
-         mach->LoopMask &= ~0x2;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) {
-         mach->LoopMask &= ~0x4;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) {
-         mach->LoopMask &= ~0x8;
-      }
-      /* TODO: if mach->LoopMask == 0, jump to end of loop */
-      UPDATE_EXEC_MASK(mach);
-      /* fall-through (for now) */
    case TGSI_OPCODE_BGNLOOP:
       /* push LoopMask and ContMasks */
       assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
@@ -3295,56 +3264,6 @@ exec_instruction(
       mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP;
       break;
 
-   case TGSI_OPCODE_ENDFOR:
-      assert(mach->LoopCounterStackTop > 0);
-      micro_sub(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], 
-                &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y],
-                &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]);
-      /* update LoopMask */
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) {
-         mach->LoopMask &= ~0x1;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) {
-         mach->LoopMask &= ~0x2;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) {
-         mach->LoopMask &= ~0x4;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) {
-         mach->LoopMask &= ~0x8;
-      }
-      micro_add(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 
-                &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 
-                &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Z]);
-      assert(mach->LoopLabelStackTop > 0);
-      inst = mach->Instructions + mach->LoopLabelStack[mach->LoopLabelStackTop - 1];
-      STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_X], 0, CHAN_X);
-      /* Restore ContMask, but don't pop */
-      assert(mach->ContStackTop > 0);
-      mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
-      UPDATE_EXEC_MASK(mach);
-      if (mach->ExecMask) {
-         /* repeat loop: jump to instruction just past BGNLOOP */
-         assert(mach->LoopLabelStackTop > 0);
-         *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
-      }
-      else {
-         /* exit loop: pop LoopMask */
-         assert(mach->LoopStackTop > 0);
-         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
-         /* pop ContMask */
-         assert(mach->ContStackTop > 0);
-         mach->ContMask = mach->ContStack[--mach->ContStackTop];
-         assert(mach->LoopLabelStackTop > 0);
-         --mach->LoopLabelStackTop;
-         assert(mach->LoopCounterStackTop > 0);
-         --mach->LoopCounterStackTop;
-
-         mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
-      }
-      UPDATE_EXEC_MASK(mach);
-      break;
-      
    case TGSI_OPCODE_ENDLOOP:
       /* Restore ContMask, but don't pop */
       assert(mach->ContStackTop > 0);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
index de0e09cdbae..cfa2f631bd8 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -106,12 +106,12 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 1, 2, 1, 0, 0, 0, "TXL", TGSI_OPCODE_TXL },
    { 0, 0, 0, 0, 0, 0, "BRK", TGSI_OPCODE_BRK },
    { 0, 1, 0, 1, 0, 1, "IF", TGSI_OPCODE_IF },
-   { 1, 1, 0, 0, 0, 1, "BGNFOR", TGSI_OPCODE_BGNFOR },
-   { 0, 1, 0, 0, 0, 1, "REP", TGSI_OPCODE_REP },
+   { 1, 1, 0, 0, 0, 1, "", 75 },      /* removed */
+   { 0, 1, 0, 0, 0, 1, "", 76 },      /* removed */
    { 0, 0, 0, 1, 1, 1, "ELSE", TGSI_OPCODE_ELSE },
    { 0, 0, 0, 0, 1, 0, "ENDIF", TGSI_OPCODE_ENDIF },
-   { 1, 0, 0, 0, 1, 0, "ENDFOR", TGSI_OPCODE_ENDFOR },
-   { 0, 0, 0, 0, 1, 0, "ENDREP", TGSI_OPCODE_ENDREP },
+   { 1, 0, 0, 0, 1, 0, "", 79 },      /* removed */
+   { 0, 0, 0, 0, 1, 0, "", 80 },      /* removed */
    { 0, 1, 0, 0, 0, 0, "PUSHA", TGSI_OPCODE_PUSHA },
    { 1, 0, 0, 0, 0, 0, "POPA", TGSI_OPCODE_POPA },
    { 1, 1, 0, 0, 0, 0, "CEIL", TGSI_OPCODE_CEIL },
diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
index e4af15c156f..e472947507d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
@@ -111,12 +111,8 @@ OP12(DP2)
 OP12_TEX(TXL)
 OP00(BRK)
 OP01_LBL(IF)
-OP11(BGNFOR)
-OP01(REP)
 OP00_LBL(ELSE)
 OP00(ENDIF)
-OP10(ENDFOR)
-OP00(ENDREP)
 OP01(PUSHA)
 OP10(POPA)
 OP11(CEIL)
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index 371f690b295..76b7564cc36 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -346,25 +346,6 @@ iter_instruction(
       }
    }
 
-   switch (inst->Instruction.Opcode) {
-   case TGSI_OPCODE_BGNFOR:
-   case TGSI_OPCODE_ENDFOR:
-      if (inst->Dst[0].Register.File != TGSI_FILE_LOOP ||
-          inst->Dst[0].Register.Index != 0) {
-         report_error(ctx, "Destination register must be LOOP[0]");
-      }
-      break;
-   }
-
-   switch (inst->Instruction.Opcode) {
-   case TGSI_OPCODE_BGNFOR:
-      if (inst->Src[0].Register.File != TGSI_FILE_CONSTANT &&
-          inst->Src[0].Register.File != TGSI_FILE_IMMEDIATE) {
-         report_error(ctx, "Source register file must be either CONST or IMM");
-      }
-      break;
-   }
-
    ctx->num_instructions++;
 
    return TRUE;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index a85cc4659e0..1071298b497 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -2533,14 +2533,6 @@ emit_instruction(
       return 0;
       break;
 
-   case TGSI_OPCODE_BGNFOR:
-      return 0;
-      break;
-
-   case TGSI_OPCODE_REP:
-      return 0;
-      break;
-
    case TGSI_OPCODE_ELSE:
       return 0;
       break;
@@ -2549,14 +2541,6 @@ emit_instruction(
       return 0;
       break;
 
-   case TGSI_OPCODE_ENDFOR:
-      return 0;
-      break;
-
-   case TGSI_OPCODE_ENDREP:
-      return 0;
-      break;
-
    case TGSI_OPCODE_PUSHA:
       return 0;
       break;
diff --git a/src/gallium/auxiliary/util/u_caps.c b/src/gallium/auxiliary/util/u_caps.c
new file mode 100644
index 00000000000..048bd5c34da
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_caps.c
@@ -0,0 +1,244 @@
+/**************************************************************************
+ *
+ * Copyright 2010 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"), 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 VMWARE 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 "pipe/p_screen.h"
+#include "util/u_format.h"
+#include "util/u_debug.h"
+#include "u_caps.h"
+
+/**
+ * Iterates over a list of caps checks as defined in u_caps.h. Should
+ * all checks pass returns TRUE and out is set to the last element of
+ * the list (TERMINATE). Should any check fail returns FALSE and set
+ * out to the index of the start of the first failing check.
+ */
+boolean
+util_check_caps_out(struct pipe_screen *screen, const unsigned *list, int *out)
+{
+   int i, tmpi;
+   float tmpf;
+
+   for (i = 0; list[i];) {
+      switch(list[i++]) {
+      case UTIL_CAPS_CHECK_CAP:
+         if (!screen->get_param(screen, list[i++])) {
+            *out = i - 2;
+            return FALSE;
+         }
+         break;
+      case UTIL_CAPS_CHECK_INT:
+         tmpi = screen->get_param(screen, list[i++]);
+         if (tmpi < (int)list[i++]) {
+            *out = i - 3;
+            return FALSE;
+         }
+         break;
+      case UTIL_CAPS_CHECK_FLOAT:
+         tmpf = screen->get_paramf(screen, list[i++]);
+         if (tmpf < (float)list[i++]) {
+            *out = i - 3;
+            return FALSE;
+         }
+         break;
+      case UTIL_CAPS_CHECK_FORMAT:
+         if (!screen->is_format_supported(screen,
+                                          list[i++],
+                                          PIPE_TEXTURE_2D,
+                                          PIPE_BIND_SAMPLER_VIEW,
+                                          0)) {
+            *out = i - 2;
+            return FALSE;
+         }
+      case UTIL_CAPS_CHECK_UNIMPLEMENTED:
+         *out = i - 1;
+         return FALSE;
+      default:
+         assert(!"Unsupported check");
+         return FALSE;
+      }
+   }
+
+   *out = i;
+   return TRUE;
+}
+
+/**
+ * Iterates over a list of caps checks as defined in u_caps.h.
+ * Returns TRUE if all caps checks pass returns FALSE otherwise.
+ */
+boolean
+util_check_caps(struct pipe_screen *screen, const unsigned *list)
+{
+   int out;
+   return util_check_caps_out(screen, list, &out);
+}
+
+
+/*
+ * Below follows some demo lists.
+ *
+ * None of these lists are exhausting lists of what is
+ * actually needed to support said API and more here for
+ * as example on how to uses the above functions. Especially
+ * for DX10 and DX11 where Gallium is missing features.
+ */
+
+/* DX 9_1 */
+static unsigned caps_dx_9_1[] = {
+   UTIL_CHECK_INT(MAX_RENDER_TARGETS, 1),
+   UTIL_CHECK_INT(MAX_TEXTURE_2D_LEVELS, 12),    /* 2048 */
+   UTIL_CHECK_INT(MAX_TEXTURE_3D_LEVELS, 9),     /* 256 */
+   UTIL_CHECK_INT(MAX_TEXTURE_CUBE_LEVELS, 10),  /* 512 */
+   UTIL_CHECK_FLOAT(MAX_TEXTURE_ANISOTROPY, 2),
+   UTIL_CHECK_TERMINATE
+};
+
+/* DX 9_2 */
+static unsigned caps_dx_9_2[] = {
+   UTIL_CHECK_CAP(OCCLUSION_QUERY),
+   UTIL_CHECK_CAP(BLEND_EQUATION_SEPARATE),
+   UTIL_CHECK_INT(MAX_RENDER_TARGETS, 1),
+   UTIL_CHECK_INT(MAX_TEXTURE_2D_LEVELS, 12),    /* 2048 */
+   UTIL_CHECK_INT(MAX_TEXTURE_3D_LEVELS, 9),     /* 256 */
+   UTIL_CHECK_INT(MAX_TEXTURE_CUBE_LEVELS, 10),  /* 512 */
+   UTIL_CHECK_FLOAT(MAX_TEXTURE_ANISOTROPY, 16),
+   UTIL_CHECK_TERMINATE
+};
+
+/* DX 9_3 */
+static unsigned caps_dx_9_3[] = {
+   UTIL_CHECK_CAP(SM3),
+ //UTIL_CHECK_CAP(INSTANCING),
+   UTIL_CHECK_CAP(OCCLUSION_QUERY),
+   UTIL_CHECK_INT(MAX_RENDER_TARGETS, 4),
+   UTIL_CHECK_INT(MAX_TEXTURE_2D_LEVELS, 13),    /* 4096 */
+   UTIL_CHECK_INT(MAX_TEXTURE_3D_LEVELS, 9),     /* 256 */
+   UTIL_CHECK_INT(MAX_TEXTURE_CUBE_LEVELS, 10),  /* 512 */
+   UTIL_CHECK_FLOAT(MAX_TEXTURE_ANISOTROPY, 16),
+   UTIL_CHECK_TERMINATE
+};
+
+/* DX 10 */
+static unsigned caps_dx_10[] = {
+   UTIL_CHECK_CAP(SM3),
+ //UTIL_CHECK_CAP(INSTANCING),
+   UTIL_CHECK_CAP(OCCLUSION_QUERY),
+   UTIL_CHECK_INT(MAX_RENDER_TARGETS, 8),
+   UTIL_CHECK_INT(MAX_TEXTURE_2D_LEVELS, 14),    /* 8192 */
+   UTIL_CHECK_INT(MAX_TEXTURE_3D_LEVELS, 12),    /* 2048 */
+   UTIL_CHECK_INT(MAX_TEXTURE_CUBE_LEVELS, 14),  /* 8192 */
+   UTIL_CHECK_FLOAT(MAX_TEXTURE_ANISOTROPY, 16),
+   UTIL_CHECK_UNIMPLEMENTED, /* XXX Unimplemented features in Gallium */
+   UTIL_CHECK_TERMINATE
+};
+
+/* DX11 */
+static unsigned caps_dx_11[] = {
+   UTIL_CHECK_CAP(SM3),
+ //UTIL_CHECK_CAP(INSTANCING),
+   UTIL_CHECK_CAP(OCCLUSION_QUERY),
+   UTIL_CHECK_INT(MAX_RENDER_TARGETS, 8),
+   UTIL_CHECK_INT(MAX_TEXTURE_2D_LEVELS, 14),    /* 16384 */
+   UTIL_CHECK_INT(MAX_TEXTURE_3D_LEVELS, 12),    /* 2048 */
+   UTIL_CHECK_INT(MAX_TEXTURE_CUBE_LEVELS, 14),  /* 16384 */
+   UTIL_CHECK_FLOAT(MAX_TEXTURE_ANISOTROPY, 16),
+   UTIL_CHECK_FORMAT(B8G8R8A8_UNORM),
+   UTIL_CHECK_UNIMPLEMENTED, /* XXX Unimplemented features in Gallium */
+   UTIL_CHECK_TERMINATE
+};
+
+/* OpenGL 2.1 */
+static unsigned caps_opengl_2_1[] = {
+   UTIL_CHECK_CAP(GLSL),
+   UTIL_CHECK_CAP(OCCLUSION_QUERY),
+   UTIL_CHECK_CAP(TWO_SIDED_STENCIL),
+   UTIL_CHECK_CAP(BLEND_EQUATION_SEPARATE),
+   UTIL_CHECK_INT(MAX_RENDER_TARGETS, 2),
+   UTIL_CHECK_TERMINATE
+};
+
+/* OpenGL 3.0 */
+/* UTIL_CHECK_INT(MAX_RENDER_TARGETS, 8), */
+
+
+/**
+ * Demo function which checks against theoretical caps needed for different APIs.
+ */
+void util_caps_demo_print(struct pipe_screen *screen)
+{
+   struct {
+      char* name;
+      unsigned *list;
+   } list[] = {
+      {"DX 9.1", caps_dx_9_1},
+      {"DX 9.2", caps_dx_9_2},
+      {"DX 9.3", caps_dx_9_3},
+      {"DX 10", caps_dx_10},
+      {"DX 11", caps_dx_11},
+      {"OpenGL 2.1", caps_opengl_2_1},
+/*    {"OpenGL 3.0", caps_opengl_3_0},*/
+      {NULL, NULL}
+   };
+   int i, out = 0;
+
+   for (i = 0; list[i].name; i++) {
+      if (util_check_caps_out(screen, list[i].list, &out)) {
+         debug_printf("%s: %s yes\n", __FUNCTION__, list[i].name);
+         continue;
+      }
+      switch (list[i].list[out]) {
+      case UTIL_CAPS_CHECK_CAP:
+         debug_printf("%s: %s no (cap %u not supported)\n", __FUNCTION__,
+                      list[i].name,
+                      list[i].list[out + 1]);
+         break;
+      case UTIL_CAPS_CHECK_INT:
+         debug_printf("%s: %s no (cap %u less then %u)\n", __FUNCTION__,
+                      list[i].name,
+                      list[i].list[out + 1],
+                      list[i].list[out + 2]);
+         break;
+      case UTIL_CAPS_CHECK_FLOAT:
+         debug_printf("%s: %s no (cap %u less then %f)\n", __FUNCTION__,
+                      list[i].name,
+                      list[i].list[out + 1],
+                      (double)(int)list[i].list[out + 2]);
+         break;
+      case UTIL_CAPS_CHECK_FORMAT:
+         debug_printf("%s: %s no (format %s not supported)\n", __FUNCTION__,
+                      list[i].name,
+                      util_format_name(list[i].list[out + 1]) + 12);
+         break;
+      case UTIL_CAPS_CHECK_UNIMPLEMENTED:
+         debug_printf("%s: %s no (not implemented in gallium or state tracker)\n",
+                      __FUNCTION__, list[i].name);
+         break;
+      default:
+            assert(!"Unsupported check");
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/util/u_caps.h b/src/gallium/auxiliary/util/u_caps.h
new file mode 100644
index 00000000000..b1074f9eb21
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_caps.h
@@ -0,0 +1,67 @@
+/**************************************************************************
+ *
+ * Copyright 2010 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"), 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 VMWARE 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 U_CAPS_H
+#define U_CAPS_H
+
+#include "pipe/p_compiler.h"
+
+struct pipe_screen;
+
+enum u_caps_check_enum {
+   UTIL_CAPS_CHECK_TERMINATE = 0,
+   UTIL_CAPS_CHECK_CAP,
+   UTIL_CAPS_CHECK_INT,
+   UTIL_CAPS_CHECK_FLOAT,
+   UTIL_CAPS_CHECK_FORMAT,
+   UTIL_CAPS_CHECK_UNIMPLEMENTED,
+};
+
+#define UTIL_CHECK_CAP(cap) \
+   UTIL_CAPS_CHECK_CAP, PIPE_CAP_##cap
+
+#define UTIL_CHECK_INT(cap, higher) \
+   UTIL_CAPS_CHECK_INT, PIPE_CAP_##cap, (unsigned)(higher)
+
+/* Floats currently lose precision */
+#define UTIL_CHECK_FLOAT(cap, higher) \
+   UTIL_CAPS_CHECK_FLOAT, PIPE_CAP_##cap, (unsigned)(int)(higher)
+
+#define UTIL_CHECK_FORMAT(format) \
+   UTIL_CAPS_CHECK_FORMAT, PIPE_FORMAT_##format
+
+#define UTIL_CHECK_UNIMPLEMENTED \
+   UTIL_CAPS_CHECK_UNIMPLEMENTED
+
+#define UTIL_CHECK_TERMINATE \
+   UTIL_CAPS_CHECK_TERMINATE
+
+boolean util_check_caps(struct pipe_screen *screen, const unsigned *list);
+boolean util_check_caps_out(struct pipe_screen *screen, const unsigned *list, int *out);
+void util_caps_demo_print(struct pipe_screen *screen);
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c
index c134f13e908..2ce643e90cd 100644
--- a/src/gallium/auxiliary/util/u_dump_state.c
+++ b/src/gallium/auxiliary/util/u_dump_state.c
@@ -656,12 +656,12 @@ util_dump_transfer(struct os_stream *stream, const struct pipe_transfer *state)
    util_dump_struct_begin(stream, "pipe_transfer");
 
    util_dump_member(stream, ptr, state, resource);
-//   util_dump_member(stream, uint, state, box);
+   /*util_dump_member(stream, uint, state, box);*/
 
    util_dump_member(stream, uint, state, stride);
    util_dump_member(stream, uint, state, slice_stride);
 
-//   util_dump_member(stream, ptr, state, data);
+   /*util_dump_member(stream, ptr, state, data);*/
 
    util_dump_struct_end(stream);
 }
diff --git a/src/gallium/auxiliary/util/u_format.h b/src/gallium/auxiliary/util/u_format.h
index 605b13bd114..fb6ade5c06b 100644
--- a/src/gallium/auxiliary/util/u_format.h
+++ b/src/gallium/auxiliary/util/u_format.h
@@ -332,7 +332,7 @@ util_format_name(enum pipe_format format)
 
    assert(desc);
    if (!desc) {
-      return "???";
+      return "PIPE_FORMAT_???";
    }
 
    return desc->name;
diff --git a/src/gallium/auxiliary/util/u_format_s3tc.c b/src/gallium/auxiliary/util/u_format_s3tc.c
index 66edb597fc8..5b279b8fe26 100644
--- a/src/gallium/auxiliary/util/u_format_s3tc.c
+++ b/src/gallium/auxiliary/util/u_format_s3tc.c
@@ -240,13 +240,14 @@ util_format_dxtn_rgb_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
                                         util_format_dxtn_fetch_t fetch,
                                         unsigned block_size)
 {
+   const unsigned bw = 4, bh = 4, comps = 4;
    unsigned x, y, i, j;
-   for(y = 0; y < height; y += 4) {
+   for(y = 0; y < height; y += bh) {
       const uint8_t *src = src_row;
-      for(x = 0; x < width; x += 4) {
-         for(j = 0; j < 4; ++j) {
-            for(i = 0; i < 4; ++i) {
-               uint8_t *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+      for(x = 0; x < width; x += bw) {
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+               uint8_t *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*comps;
                fetch(0, src, i, j, dst);
             }
          }
@@ -379,212 +380,197 @@ util_format_dxt5_rgba_unpack_rgba_float(float *dst_row, unsigned dst_stride,
 
 void
 util_format_dxt1_rgb_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
-                                      const uint8_t *src_row, unsigned src_stride,
+                                      const uint8_t *src, unsigned src_stride,
                                       unsigned width, unsigned height)
 {
+   const unsigned bw = 4, bh = 4, bytes_per_block = 8;
    unsigned x, y, i, j, k;
-   for(y = 0; y < height; y += 4) {
-      const uint8_t *src = src_row;
+   for(y = 0; y < height; y += bh) {
       uint8_t *dst = dst_row;
-      for(x = 0; x < width; x += 4) {
-         uint8_t tmp[4][4][3];
-         for(j = 0; j < 4; ++j) {
-            for(i = 0; i < 4; ++i) {
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp[4][4][3];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
                for(k = 0; k < 3; ++k) {
-                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + (x + i)*4 + k];
                }
             }
          }
-         util_format_dxtn_pack(3, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGB, dst, dst_stride);
-         src += 4*4;
-         dst += 8;
+         util_format_dxtn_pack(3, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGB, dst, 0);
+         dst += bytes_per_block;
       }
-      src_row += src_stride;
-      dst_row += 4*dst_stride/sizeof(*dst_row);
+      dst_row += dst_stride / sizeof(*dst_row);
    }
 }
 
 void
 util_format_dxt1_rgba_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
-                                       const uint8_t *src_row, unsigned src_stride,
+                                       const uint8_t *src, unsigned src_stride,
                                        unsigned width, unsigned height)
 {
+   const unsigned bw = 4, bh = 4, comps = 4, bytes_per_block = 8;
    unsigned x, y, i, j, k;
-   for(y = 0; y < height; y += 4) {
-      const uint8_t *src = src_row;
+   for(y = 0; y < height; y += bh) {
       uint8_t *dst = dst_row;
-      for(x = 0; x < width; x += 4) {
-         uint8_t tmp[4][4][4];
-         for(j = 0; j < 4; ++j) {
-            for(i = 0; i < 4; ++i) {
-               for(k = 0; k < 4; ++k) {
-                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp[4][4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+               for(k = 0; k < comps; ++k) {
+                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + (x + i)*comps + k];
                }
             }
          }
-         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGBA, dst, dst_stride);
-         src += 4*4;
-         dst += 8;
+         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGBA, dst, 0);
+         dst += bytes_per_block;
       }
-      src_row += src_stride;
-      dst_row += 4*dst_stride/sizeof(*dst_row);
+      dst_row += dst_stride / sizeof(*dst_row);
    }
 }
 
 void
 util_format_dxt3_rgba_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
-                                       const uint8_t *src_row, unsigned src_stride,
+                                       const uint8_t *src, unsigned src_stride,
                                        unsigned width, unsigned height)
 {
+   const unsigned bw = 4, bh = 4, comps = 4, bytes_per_block = 16;
    unsigned x, y, i, j, k;
-   for(y = 0; y < height; y += 4) {
-      const uint8_t *src = src_row;
+   for(y = 0; y < height; y += bh) {
       uint8_t *dst = dst_row;
-      for(x = 0; x < width; x += 4) {
-         uint8_t tmp[4][4][4];
-         for(j = 0; j < 4; ++j) {
-            for(i = 0; i < 4; ++i) {
-               for(k = 0; k < 4; ++k) {
-                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp[4][4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+               for(k = 0; k < comps; ++k) {
+                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + (x + i)*comps + k];
                }
             }
          }
-         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT3_RGBA, dst, dst_stride);
-         src += 4*4;
-         dst += 16;
+         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT3_RGBA, dst, 0);
+         dst += bytes_per_block;
       }
-      src_row += src_stride;
-      dst_row += 4*dst_stride/sizeof(*dst_row);
+      dst_row += dst_stride / sizeof(*dst_row);
    }
 }
 
 void
 util_format_dxt5_rgba_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
-                                       const uint8_t *src_row, unsigned src_stride,
+                                       const uint8_t *src, unsigned src_stride,
                                        unsigned width, unsigned height)
 {
+   const unsigned bw = 4, bh = 4, comps = 4, bytes_per_block = 16;
    unsigned x, y, i, j, k;
-   for(y = 0; y < height; y += 4) {
-      const uint8_t *src = src_row;
+
+   for(y = 0; y < height; y += bh) {
       uint8_t *dst = dst_row;
-      for(x = 0; x < width; x += 4) {
-         uint8_t tmp[4][4][4];
-         for(j = 0; j < 4; ++j) {
-            for(i = 0; i < 4; ++i) {
-               for(k = 0; k < 4; ++k) {
-                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp[4][4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+               for(k = 0; k < comps; ++k) {
+                  tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + (x + i)*comps + k];
                }
             }
          }
-         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT5_RGBA, dst, dst_stride);
-         src += 4*4;
-         dst += 16;
+         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT5_RGBA, dst, 0);
+         dst += bytes_per_block;
       }
-      src_row += src_stride;
-      dst_row += 4*dst_stride/sizeof(*dst_row);
+      dst_row += dst_stride / sizeof(*dst_row);
    }
 }
 
 void
 util_format_dxt1_rgb_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride,
-                                     const float *src_row, unsigned src_stride,
+                                     const float *src, unsigned src_stride,
                                      unsigned width, unsigned height)
 {
    unsigned x, y, i, j, k;
    for(y = 0; y < height; y += 4) {
-      const float *src = src_row;
       uint8_t *dst = dst_row;
       for(x = 0; x < width; x += 4) {
          uint8_t tmp[4][4][3];
          for(j = 0; j < 4; ++j) {
             for(i = 0; i < 4; ++i) {
                for(k = 0; k < 3; ++k) {
-                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + (x+i)*4 + k]);
                }
             }
          }
-         util_format_dxtn_pack(3, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGB, dst, dst_stride);
-         src += 4*4;
+         util_format_dxtn_pack(3, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGB, dst, 0);
          dst += 8;
       }
-      src_row += src_stride;
       dst_row += 4*dst_stride/sizeof(*dst_row);
    }
 }
 
 void
 util_format_dxt1_rgba_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride,
-                                      const float *src_row, unsigned src_stride,
+                                      const float *src, unsigned src_stride,
                                       unsigned width, unsigned height)
 {
    unsigned x, y, i, j, k;
    for(y = 0; y < height; y += 4) {
-      const float *src = src_row;
       uint8_t *dst = dst_row;
       for(x = 0; x < width; x += 4) {
          uint8_t tmp[4][4][4];
          for(j = 0; j < 4; ++j) {
             for(i = 0; i < 4; ++i) {
                for(k = 0; k < 4; ++k) {
-                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + (x+i)*4 + k]);
                }
             }
          }
-         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGBA, dst, dst_stride);
-         src += 4*4;
+         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGBA, dst, 0);
          dst += 8;
       }
-      src_row += src_stride;
       dst_row += 4*dst_stride/sizeof(*dst_row);
    }
 }
 
 void
-util_format_dxt3_rgba_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
+util_format_dxt3_rgba_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride,
+                                      const float *src, unsigned src_stride,
+                                      unsigned width, unsigned height)
 {
    unsigned x, y, i, j, k;
    for(y = 0; y < height; y += 4) {
-      const float *src = src_row;
       uint8_t *dst = dst_row;
       for(x = 0; x < width; x += 4) {
          uint8_t tmp[4][4][4];
          for(j = 0; j < 4; ++j) {
             for(i = 0; i < 4; ++i) {
                for(k = 0; k < 4; ++k) {
-                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + (x+i)*4 + k]);
                }
             }
          }
-         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT3_RGBA, dst, dst_stride);
-         src += 4*4;
+         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT3_RGBA, dst, 0);
          dst += 16;
       }
-      src_row += src_stride;
       dst_row += 4*dst_stride/sizeof(*dst_row);
    }
 }
 
 void
-util_format_dxt5_rgba_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
+util_format_dxt5_rgba_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride,
+                                      const float *src, unsigned src_stride,
+                                      unsigned width, unsigned height)
 {
    unsigned x, y, i, j, k;
    for(y = 0; y < height; y += 4) {
-      const float *src = src_row;
       uint8_t *dst = dst_row;
       for(x = 0; x < width; x += 4) {
          uint8_t tmp[4][4][4];
          for(j = 0; j < 4; ++j) {
             for(i = 0; i < 4; ++i) {
                for(k = 0; k < 4; ++k) {
-                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + (x+i)*4 + k]);
                }
             }
          }
-         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT5_RGBA, dst, dst_stride);
-         src += 4*4;
+         util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT5_RGBA, dst, 0);
          dst += 16;
       }
-      src_row += src_stride;
       dst_row += 4*dst_stride/sizeof(*dst_row);
    }
 }
diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c
index fe327c302b7..f7aa1403d08 100644
--- a/src/gallium/auxiliary/util/u_tile.c
+++ b/src/gallium/auxiliary/util/u_tile.c
@@ -544,7 +544,7 @@ pipe_put_tile_z(struct pipe_context *pipe,
    case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
       {
          uint *pDest = (uint *) (map + y * pt->stride + x*4);
-         //assert((pt->usage & PIPE_TRANSFER_READ_WRITE) == PIPE_TRANSFER_READ_WRITE);
+         /*assert((pt->usage & PIPE_TRANSFER_READ_WRITE) == PIPE_TRANSFER_READ_WRITE);*/
          for (i = 0; i < h; i++) {
             for (j = 0; j < w; j++) {
                /* convert 32-bit Z to 24-bit Z, preserve stencil */
@@ -571,7 +571,7 @@ pipe_put_tile_z(struct pipe_context *pipe,
    case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
       {
          uint *pDest = (uint *) (map + y * pt->stride + x*4);
-         //assert((pt->usage & PIPE_TRANSFER_READ_WRITE) == PIPE_TRANSFER_READ_WRITE);
+         /*assert((pt->usage & PIPE_TRANSFER_READ_WRITE) == PIPE_TRANSFER_READ_WRITE);*/
          for (i = 0; i < h; i++) {
             for (j = 0; j < w; j++) {
                /* convert 32-bit Z to 24-bit Z, preserve stencil */