Merge commit 'origin/master' into gallium-msaa

6 files changed, 301 insertions, 140 deletions

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
index 20ae958714b..f372a48846f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
@@ -1177,9 +1177,34 @@ lp_build_rcp(struct lp_build_context *bld,
    if(LLVMIsConstant(a))
       return LLVMConstFDiv(bld->one, a);
 
-   if(util_cpu_caps.has_sse && type.width == 32 && type.length == 4)
-      /* FIXME: improve precision */
+   if(util_cpu_caps.has_sse && type.width == 32 && type.length == 4) {
+      /*
+       * XXX: Added precision is not always necessary, so only enable this
+       * when we have a better system in place to track minimum precision.
+       */
+
+#if 0
+      /*
+       * Do one Newton-Raphson step to improve precision:
+       *
+       *   x1 = (2 - a * rcp(a)) * rcp(a)
+       */
+
+      LLVMValueRef two = lp_build_const_vec(bld->type, 2.0);
+      LLVMValueRef rcp_a;
+      LLVMValueRef res;
+
+      rcp_a = lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rcp.ps", lp_build_vec_type(type), a);
+
+      res = LLVMBuildMul(bld->builder, a, rcp_a, "");
+      res = LLVMBuildSub(bld->builder, two, res, "");
+      res = LLVMBuildMul(bld->builder, res, rcp_a, "");
+
+      return rcp_a;
+#else
       return lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rcp.ps", lp_build_vec_type(type), a);
+#endif
+   }
 
    return LLVMBuildFDiv(bld->builder, bld->one, a, "");
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_limits.h b/src/gallium/auxiliary/gallivm/lp_bld_limits.h
new file mode 100644
index 00000000000..e095a0abe3a
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/lp_bld_limits.h
@@ -0,0 +1,53 @@
+/**************************************************************************
+ *
+ * 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 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ **************************************************************************/
+
+
+#ifndef LP_BLD_LIMITS_H_
+#define LP_BLD_LIMITS_H_
+
+/*
+ * TGSI translation limits.
+ *
+ * Some are slightly above SM 3.0 requirements to give some wiggle room to
+ * the state trackers.
+ */
+
+#define LP_MAX_TGSI_TEMPS 256
+
+#define LP_MAX_TGSI_ADDRS 16
+
+#define LP_MAX_TGSI_IMMEDIATES 256
+
+/**
+ * Maximum control flow nesting
+ *
+ * SM3.0 requires 24
+ */
+#define LP_MAX_TGSI_NESTING 32
+
+
+#endif /* LP_BLD_LIMITS_H_ */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
index 8ceb20473d5..955b1d21ee5 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
@@ -173,7 +173,10 @@ lp_build_sample_soa(LLVMBuilderRef builder,
                     unsigned unit,
                     unsigned num_coords,
                     const LLVMValueRef *coords,
-                    LLVMValueRef lodbias,
+                    const LLVMValueRef *ddx,
+                    const LLVMValueRef *ddy,
+                    LLVMValueRef lod_bias,
+                    LLVMValueRef explicit_lod,
                     LLVMValueRef *texel);
 
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
index 74dc9e1d812..3f0ea05b795 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
@@ -30,6 +30,7 @@
  * Texture sampling -- SoA.
  *
  * @author Jose Fonseca <[email protected]>
+ * @author Brian Paul <[email protected]>
  */
 
 #include "pipe/p_defines.h"
@@ -325,6 +326,18 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
                                    bld->format_desc,
                                    x, y, z, y_stride, z_stride);
 
+   if (use_border) {
+      /* If we can sample the border color, it means that texcoords may
+       * lie outside the bounds of the texture image.  We need to do
+       * something to prevent reading out of bounds and causing a segfault.
+       *
+       * Simply AND the texture coords with !use_border.  This will cause
+       * coords which are out of bounds to become zero.  Zero's guaranteed
+       * to be inside the texture image.
+       */
+      offset = lp_build_andc(&bld->uint_coord_bld, offset, use_border);
+   }
+
    lp_build_fetch_rgba_soa(bld->builder,
                            bld->format_desc,
                            bld->texel_type,
@@ -891,17 +904,24 @@ lp_build_minify(struct lp_build_sample_context *bld,
  * \param s  vector of texcoord s values
  * \param t  vector of texcoord t values
  * \param r  vector of texcoord r values
- * \param shader_lod_bias  vector float with the shader lod bias,
+ * \param lod_bias  optional float vector with the shader lod bias
+ * \param explicit_lod  optional float vector with the explicit lod
  * \param width  scalar int texture width
  * \param height  scalar int texture height
  * \param depth  scalar int texture depth
+ *
+ * XXX: The resulting lod is scalar, so ignore all but the first element of
+ * derivatives, lod_bias, etc that are passed by the shader.
  */
 static LLVMValueRef
 lp_build_lod_selector(struct lp_build_sample_context *bld,
                       LLVMValueRef s,
                       LLVMValueRef t,
                       LLVMValueRef r,
-                      LLVMValueRef shader_lod_bias,
+                      const LLVMValueRef *ddx,
+                      const LLVMValueRef *ddy,
+                      LLVMValueRef lod_bias, /* optional */
+                      LLVMValueRef explicit_lod, /* optional */
                       LLVMValueRef width,
                       LLVMValueRef height,
                       LLVMValueRef depth)
@@ -914,7 +934,6 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
       return LLVMConstReal(LLVMFloatType(), bld->static_state->min_lod);
    }
    else {
-      const int dims = texture_dims(bld->static_state->target);
       struct lp_build_context *float_bld = &bld->float_bld;
       LLVMValueRef sampler_lod_bias = LLVMConstReal(LLVMFloatType(),
                                                     bld->static_state->lod_bias);
@@ -922,83 +941,76 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
                                            bld->static_state->min_lod);
       LLVMValueRef max_lod = LLVMConstReal(LLVMFloatType(),
                                            bld->static_state->max_lod);
-
       LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-      LLVMValueRef index1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-      LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-
-      LLVMValueRef s0, s1, s2;
-      LLVMValueRef t0, t1, t2;
-      LLVMValueRef r0, r1, r2;
-      LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
-      LLVMValueRef rho, lod;
-
-      /*
-       * dsdx = abs(s[1] - s[0]);
-       * dsdy = abs(s[2] - s[0]);
-       * dtdx = abs(t[1] - t[0]);
-       * dtdy = abs(t[2] - t[0]);
-       * drdx = abs(r[1] - r[0]);
-       * drdy = abs(r[2] - r[0]);
-       * XXX we're assuming a four-element quad in 2x2 layout here.
-       */
-      s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0");
-      s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1");
-      s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2");
-      dsdx = LLVMBuildSub(bld->builder, s1, s0, "");
-      dsdx = lp_build_abs(float_bld, dsdx);
-      dsdy = LLVMBuildSub(bld->builder, s2, s0, "");
-      dsdy = lp_build_abs(float_bld, dsdy);
-      if (dims > 1) {
-         t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0");
-         t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1");
-         t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2");
-         dtdx = LLVMBuildSub(bld->builder, t1, t0, "");
-         dtdx = lp_build_abs(float_bld, dtdx);
-         dtdy = LLVMBuildSub(bld->builder, t2, t0, "");
-         dtdy = lp_build_abs(float_bld, dtdy);
-         if (dims > 2) {
-            r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0");
-            r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1");
-            r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2");
-            drdx = LLVMBuildSub(bld->builder, r1, r0, "");
-            drdx = lp_build_abs(float_bld, drdx);
-            drdy = LLVMBuildSub(bld->builder, r2, r0, "");
-            drdy = lp_build_abs(float_bld, drdy);
-         }
+      LLVMValueRef lod;
+
+      if (explicit_lod) {
+         lod = LLVMBuildExtractElement(bld->builder, explicit_lod,
+                                       index0, "");
       }
+      else {
+         const int dims = texture_dims(bld->static_state->target);
+         LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
+         LLVMValueRef rho;
+
+         /*
+          * dsdx = abs(s[1] - s[0]);
+          * dsdy = abs(s[2] - s[0]);
+          * dtdx = abs(t[1] - t[0]);
+          * dtdy = abs(t[2] - t[0]);
+          * drdx = abs(r[1] - r[0]);
+          * drdy = abs(r[2] - r[0]);
+          */
+         dsdx = LLVMBuildExtractElement(bld->builder, ddx[0], index0, "dsdx");
+         dsdx = lp_build_abs(float_bld, dsdx);
+         dsdy = LLVMBuildExtractElement(bld->builder, ddy[0], index0, "dsdy");
+         dsdy = lp_build_abs(float_bld, dsdy);
+         if (dims > 1) {
+            dtdx = LLVMBuildExtractElement(bld->builder, ddx[1], index0, "dtdx");
+            dtdx = lp_build_abs(float_bld, dtdx);
+            dtdy = LLVMBuildExtractElement(bld->builder, ddy[1], index0, "dtdy");
+            dtdy = lp_build_abs(float_bld, dtdy);
+            if (dims > 2) {
+               drdx = LLVMBuildExtractElement(bld->builder, ddx[2], index0, "drdx");
+               drdx = lp_build_abs(float_bld, drdx);
+               drdy = LLVMBuildExtractElement(bld->builder, ddy[2], index0, "drdy");
+               drdy = lp_build_abs(float_bld, drdy);
+            }
+         }
 
-      /* Compute rho = max of all partial derivatives scaled by texture size.
-       * XXX this could be vectorized somewhat
-       */
-      rho = LLVMBuildMul(bld->builder,
-                         lp_build_max(float_bld, dsdx, dsdy),
-                         lp_build_int_to_float(float_bld, width), "");
-      if (dims > 1) {
-         LLVMValueRef max;
-         max = LLVMBuildMul(bld->builder,
-                            lp_build_max(float_bld, dtdx, dtdy),
-                            lp_build_int_to_float(float_bld, height), "");
-         rho = lp_build_max(float_bld, rho, max);
-         if (dims > 2) {
+         /* Compute rho = max of all partial derivatives scaled by texture size.
+          * XXX this could be vectorized somewhat
+          */
+         rho = LLVMBuildMul(bld->builder,
+                            lp_build_max(float_bld, dsdx, dsdy),
+                            lp_build_int_to_float(float_bld, width), "");
+         if (dims > 1) {
+            LLVMValueRef max;
             max = LLVMBuildMul(bld->builder,
-                               lp_build_max(float_bld, drdx, drdy),
-                               lp_build_int_to_float(float_bld, depth), "");
+                               lp_build_max(float_bld, dtdx, dtdy),
+                               lp_build_int_to_float(float_bld, height), "");
             rho = lp_build_max(float_bld, rho, max);
+            if (dims > 2) {
+               max = LLVMBuildMul(bld->builder,
+                                  lp_build_max(float_bld, drdx, drdy),
+                                  lp_build_int_to_float(float_bld, depth), "");
+               rho = lp_build_max(float_bld, rho, max);
+            }
          }
-      }
 
-      /* compute lod = log2(rho) */
-      lod = lp_build_log2(float_bld, rho);
+         /* compute lod = log2(rho) */
+         lod = lp_build_log2(float_bld, rho);
 
-      /* add sampler lod bias */
-      lod = LLVMBuildAdd(bld->builder, lod, sampler_lod_bias, "sampler LOD bias");
+         /* add shader lod bias */
+         if (lod_bias) {
+            lod_bias = LLVMBuildExtractElement(bld->builder, lod_bias,
+                                               index0, "");
+            lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "shader_lod_bias");
+         }
+      }
 
-      /* add shader lod bias */
-      /* XXX for now we take only the first element since our lod is scalar */
-      shader_lod_bias = LLVMBuildExtractElement(bld->builder, shader_lod_bias,
-                                                LLVMConstInt(LLVMInt32Type(), 0, 0), "");
-      lod = LLVMBuildAdd(bld->builder, lod, shader_lod_bias, "shader LOD bias");
+      /* add sampler lod bias */
+      lod = LLVMBuildAdd(bld->builder, lod, sampler_lod_bias, "sampler_lod_bias");
 
       /* clamp lod */
       lod = lp_build_clamp(float_bld, lod, min_lod, max_lod);
@@ -1584,7 +1596,10 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
                         LLVMValueRef s,
                         LLVMValueRef t,
                         LLVMValueRef r,
-                        LLVMValueRef lodbias,
+                        const LLVMValueRef *ddx,
+                        const LLVMValueRef *ddy,
+                        LLVMValueRef lod_bias, /* optional */
+                        LLVMValueRef explicit_lod, /* optional */
                         LLVMValueRef width,
                         LLVMValueRef height,
                         LLVMValueRef depth,
@@ -1622,7 +1637,9 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
       /* Need to compute lod either to choose mipmap levels or to
        * distinguish between minification/magnification with one mipmap level.
        */
-      lod = lp_build_lod_selector(bld, s, t, r, lodbias, width, height, depth);
+      lod = lp_build_lod_selector(bld, s, t, r, ddx, ddy,
+                                  lod_bias, explicit_lod,
+                                  width, height, depth);
    }
 
    /*
@@ -2049,6 +2066,24 @@ lp_build_sample_compare(struct lp_build_sample_context *bld,
 
 
 /**
+ * Just set texels to white instead of actually sampling the texture.
+ * For debugging.
+ */
+static void
+lp_build_sample_nop(struct lp_build_sample_context *bld,
+                    LLVMValueRef *texel)
+{
+   struct lp_build_context *texel_bld = &bld->texel_bld;
+   unsigned chan;
+
+   for (chan = 0; chan < 4; chan++) {
+      /*lp_bld_mov(texel_bld, texel, texel_bld->one);*/
+      texel[chan] = texel_bld->one;
+   }  
+}
+
+
+/**
  * Build texture sampling code.
  * 'texel' will return a vector of four LLVMValueRefs corresponding to
  * R, G, B, A.
@@ -2062,7 +2097,10 @@ lp_build_sample_soa(LLVMBuilderRef builder,
                     unsigned unit,
                     unsigned num_coords,
                     const LLVMValueRef *coords,
-                    LLVMValueRef lodbias,
+                    const LLVMValueRef *ddx,
+                    const LLVMValueRef *ddy,
+                    LLVMValueRef lod_bias, /* optional */
+                    LLVMValueRef explicit_lod, /* optional */
                     LLVMValueRef *texel)
 {
    struct lp_build_sample_context bld;
@@ -2113,19 +2151,24 @@ lp_build_sample_soa(LLVMBuilderRef builder,
    height_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, height);
    depth_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, depth);
 
-   if (util_format_is_rgba8_variant(bld.format_desc) &&
-       static_state->target == PIPE_TEXTURE_2D &&
-       static_state->min_img_filter == PIPE_TEX_FILTER_LINEAR &&
-       static_state->mag_img_filter == PIPE_TEX_FILTER_LINEAR &&
-       static_state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE &&
-       is_simple_wrap_mode(static_state->wrap_s) &&
-       is_simple_wrap_mode(static_state->wrap_t)) {
+   if (0) {
+      /* For debug: no-op texture sampling */
+      lp_build_sample_nop(&bld, texel);
+   }
+   else if (util_format_is_rgba8_variant(bld.format_desc) &&
+            static_state->target == PIPE_TEXTURE_2D &&
+            static_state->min_img_filter == PIPE_TEX_FILTER_LINEAR &&
+            static_state->mag_img_filter == PIPE_TEX_FILTER_LINEAR &&
+            static_state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE &&
+            is_simple_wrap_mode(static_state->wrap_s) &&
+            is_simple_wrap_mode(static_state->wrap_t)) {
       /* special case */
       lp_build_sample_2d_linear_aos(&bld, s, t, width_vec, height_vec,
                                     row_stride_array, data_array, texel);
    }
    else {
-      lp_build_sample_general(&bld, unit, s, t, r, lodbias,
+      lp_build_sample_general(&bld, unit, s, t, r, ddx, ddy,
+                              lod_bias, explicit_lod,
                               width, height, depth,
                               width_vec, height_vec, depth_vec,
                               row_stride_array, img_stride_array,
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
index 2eac5da6c69..5ce1385c480 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
@@ -65,7 +65,10 @@ struct lp_build_sampler_soa
                         unsigned unit,
                         unsigned num_coords,
                         const LLVMValueRef *coords,
-                        LLVMValueRef lodbias,
+                        const LLVMValueRef *ddx,
+                        const LLVMValueRef *ddy,
+                        LLVMValueRef lod_bias, /* optional */
+                        LLVMValueRef explicit_lod, /* optional */
                         LLVMValueRef *texel);
 };
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index d3c769e28b8..0b1a28a7ab7 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -54,13 +54,10 @@
 #include "lp_bld_swizzle.h"
 #include "lp_bld_flow.h"
 #include "lp_bld_tgsi.h"
+#include "lp_bld_limits.h"
 #include "lp_bld_debug.h"
 
 
-#define LP_MAX_TEMPS 256
-#define LP_MAX_IMMEDIATES 256
-
-
 #define FOR_EACH_CHANNEL( CHAN )\
    for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)
 
@@ -84,7 +81,6 @@
 #define QUAD_BOTTOM_LEFT  2
 #define QUAD_BOTTOM_RIGHT 3
 
-#define LP_TGSI_MAX_NESTING 16
 
 struct lp_exec_mask {
    struct lp_build_context *bld;
@@ -93,19 +89,19 @@ struct lp_exec_mask {
 
    LLVMTypeRef int_vec_type;
 
-   LLVMValueRef cond_stack[LP_TGSI_MAX_NESTING];
+   LLVMValueRef cond_stack[LP_MAX_TGSI_NESTING];
    int cond_stack_size;
    LLVMValueRef cond_mask;
 
-   LLVMValueRef break_stack[LP_TGSI_MAX_NESTING];
+   LLVMValueRef break_stack[LP_MAX_TGSI_NESTING];
    int break_stack_size;
    LLVMValueRef break_mask;
 
-   LLVMValueRef cont_stack[LP_TGSI_MAX_NESTING];
+   LLVMValueRef cont_stack[LP_MAX_TGSI_NESTING];
    int cont_stack_size;
    LLVMValueRef cont_mask;
 
-   LLVMBasicBlockRef loop_stack[LP_TGSI_MAX_NESTING];
+   LLVMBasicBlockRef loop_stack[LP_MAX_TGSI_NESTING];
    int loop_stack_size;
    LLVMBasicBlockRef loop_block;
 
@@ -124,9 +120,9 @@ struct lp_build_tgsi_soa_context
 
    struct lp_build_sampler_soa *sampler;
 
-   LLVMValueRef immediates[LP_MAX_IMMEDIATES][NUM_CHANNELS];
-   LLVMValueRef temps[LP_MAX_TEMPS][NUM_CHANNELS];
-   LLVMValueRef addr[LP_MAX_TEMPS][NUM_CHANNELS];
+   LLVMValueRef immediates[LP_MAX_TGSI_IMMEDIATES][NUM_CHANNELS];
+   LLVMValueRef temps[LP_MAX_TGSI_TEMPS][NUM_CHANNELS];
+   LLVMValueRef addr[LP_MAX_TGSI_ADDRS][NUM_CHANNELS];
 
    /* we allocate an array of temps if we have indirect
     * addressing and then the temps above is unused */
@@ -198,6 +194,7 @@ static void lp_exec_mask_update(struct lp_exec_mask *mask)
 static void lp_exec_mask_cond_push(struct lp_exec_mask *mask,
                                    LLVMValueRef val)
 {
+   assert(mask->cond_stack_size < LP_MAX_TGSI_NESTING);
    mask->cond_stack[mask->cond_stack_size++] = mask->cond_mask;
    mask->cond_mask = LLVMBuildBitCast(mask->bld->builder, val,
                                       mask->int_vec_type, "");
@@ -239,6 +236,10 @@ static void lp_exec_bgnloop(struct lp_exec_mask *mask)
    if (mask->cond_stack_size == 0)
       mask->cond_mask = LLVMConstAllOnes(mask->int_vec_type);
 
+   assert(mask->break_stack_size < LP_MAX_TGSI_NESTING);
+   assert(mask->cont_stack_size < LP_MAX_TGSI_NESTING);
+   assert(mask->break_stack_size < LP_MAX_TGSI_NESTING);
+
    mask->break_stack[mask->break_stack_size++] = mask->break_mask;
    mask->cont_stack[mask->cont_stack_size++] = mask->cont_mask;
    mask->loop_stack[mask->loop_stack_size++] = mask->loop_block;
@@ -255,16 +256,9 @@ static void lp_exec_break(struct lp_exec_mask *mask)
                                          mask->exec_mask,
                                          "break");
 
-   /* mask->break_stack_size > 1 implies that we encountered a break
-    * statemant already and if that's the case we want to make sure
-    * our mask is a combination of the previous break and the current
-    * execution mask */
-   if (mask->break_stack_size > 1) {
-      mask->break_mask = LLVMBuildAnd(mask->bld->builder,
-                                      mask->break_mask,
-                                      exec_mask, "break_full");
-   } else
-      mask->break_mask = exec_mask;
+   mask->break_mask = LLVMBuildAnd(mask->bld->builder,
+                                   mask->break_mask,
+                                   exec_mask, "break_full");
 
    lp_exec_mask_update(mask);
 }
@@ -275,12 +269,9 @@ static void lp_exec_continue(struct lp_exec_mask *mask)
                                          mask->exec_mask,
                                          "");
 
-   if (mask->cont_stack_size > 1) {
-      mask->cont_mask = LLVMBuildAnd(mask->bld->builder,
-                                     mask->cont_mask,
-                                     exec_mask, "");
-   } else
-      mask->cont_mask = exec_mask;
+   mask->cont_mask = LLVMBuildAnd(mask->bld->builder,
+                                  mask->cont_mask,
+                                  exec_mask, "");
 
    lp_exec_mask_update(mask);
 }
@@ -589,7 +580,6 @@ emit_store(
 
    case TGSI_FILE_PREDICATE:
       /* FIXME */
-      assert(0);
       break;
 
    default:
@@ -602,21 +592,37 @@ emit_store(
  * High-level instruction translators.
  */
 
+enum tex_modifier {
+   TEX_MODIFIER_NONE = 0,
+   TEX_MODIFIER_PROJECTED,
+   TEX_MODIFIER_LOD_BIAS,
+   TEX_MODIFIER_EXPLICIT_LOD,
+   TEX_MODIFIER_EXPLICIT_DERIV
+};
 
 static void
 emit_tex( struct lp_build_tgsi_soa_context *bld,
           const struct tgsi_full_instruction *inst,
-          boolean apply_lodbias,
-          boolean projected,
+          enum tex_modifier modifier,
           LLVMValueRef *texel)
 {
-   const uint unit = inst->Src[1].Register.Index;
-   LLVMValueRef lodbias;
+   unsigned unit;
+   LLVMValueRef lod_bias, explicit_lod;
    LLVMValueRef oow = NULL;
    LLVMValueRef coords[3];
+   LLVMValueRef ddx[3];
+   LLVMValueRef ddy[3];
    unsigned num_coords;
    unsigned i;
 
+   if (!bld->sampler) {
+      _debug_printf("warning: found texture instruction but no sampler generator supplied\n");
+      for (i = 0; i < 4; i++) {
+         texel[i] = bld->base.undef;
+      }
+      return;
+   }
+
    switch (inst->Texture.Texture) {
    case TGSI_TEXTURE_1D:
       num_coords = 1;
@@ -637,29 +643,57 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
       return;
    }
 
-   if(apply_lodbias)
-      lodbias = emit_fetch( bld, inst, 0, 3 );
-   else
-      lodbias = bld->base.zero;
+   if (modifier == TEX_MODIFIER_LOD_BIAS) {
+      lod_bias = emit_fetch( bld, inst, 0, 3 );
+      explicit_lod = NULL;
+   }
+   else if (modifier == TEX_MODIFIER_EXPLICIT_LOD) {
+      lod_bias = NULL;
+      explicit_lod = emit_fetch( bld, inst, 0, 3 );
+   }
+   else {
+      lod_bias = NULL;
+      explicit_lod = NULL;
+   }
 
-   if (projected) {
+   if (modifier == TEX_MODIFIER_PROJECTED) {
       oow = emit_fetch( bld, inst, 0, 3 );
       oow = lp_build_rcp(&bld->base, oow);
    }
 
    for (i = 0; i < num_coords; i++) {
       coords[i] = emit_fetch( bld, inst, 0, i );
-      if (projected)
+      if (modifier == TEX_MODIFIER_PROJECTED)
          coords[i] = lp_build_mul(&bld->base, coords[i], oow);
    }
    for (i = num_coords; i < 3; i++) {
       coords[i] = bld->base.undef;
    }
 
+   if (modifier == TEX_MODIFIER_EXPLICIT_DERIV) {
+      for (i = 0; i < num_coords; i++) {
+         ddx[i] = emit_fetch( bld, inst, 1, i );
+         ddy[i] = emit_fetch( bld, inst, 2, i );
+      }
+      unit = inst->Src[3].Register.Index;
+   }  else {
+      for (i = 0; i < num_coords; i++) {
+         ddx[i] = emit_ddx( bld, coords[i] );
+         ddy[i] = emit_ddy( bld, coords[i] );
+      }
+      unit = inst->Src[1].Register.Index;
+   }
+   for (i = num_coords; i < 3; i++) {
+      ddx[i] = bld->base.undef;
+      ddy[i] = bld->base.undef;
+   }
+
    bld->sampler->emit_fetch_texel(bld->sampler,
                                   bld->base.builder,
                                   bld->base.type,
-                                  unit, num_coords, coords, lodbias,
+                                  unit, num_coords, coords,
+                                  ddx, ddy,
+                                  lod_bias, explicit_lod,
                                   texel);
 }
 
@@ -739,7 +773,7 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld,
    lp_build_mask_update(bld->mask, mask);
 }
 
-static int
+static void
 emit_declaration(
    struct lp_build_tgsi_soa_context *bld,
    const struct tgsi_full_declaration *decl)
@@ -753,6 +787,7 @@ emit_declaration(
    for (idx = first; idx <= last; ++idx) {
       switch (decl->Declaration.File) {
       case TGSI_FILE_TEMPORARY:
+         assert(idx < LP_MAX_TGSI_TEMPS);
          if (bld->has_indirect_addressing) {
             LLVMValueRef val = LLVMConstInt(LLVMInt32Type(),
                                             last*4 + 4, 0);
@@ -772,18 +807,21 @@ emit_declaration(
          break;
 
       case TGSI_FILE_ADDRESS:
+         assert(idx < LP_MAX_TGSI_ADDRS);
          for (i = 0; i < NUM_CHANNELS; i++)
             bld->addr[idx][i] = lp_build_alloca(bld->base.builder,
                                                 vec_type, "");
          break;
 
+      case TGSI_FILE_PREDICATE:
+         _debug_printf("warning: predicate registers not yet implemented\n");
+         break;
+
       default:
          /* don't need to declare other vars */
          break;
       }
    }
-
-   return TRUE;
 }
 
 
@@ -1359,12 +1397,11 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TEX:
-      emit_tex( bld, inst, FALSE, FALSE, dst0 );
+      emit_tex( bld, inst, TEX_MODIFIER_NONE, dst0 );
       break;
 
    case TGSI_OPCODE_TXD:
-      /* FIXME */
-      return FALSE;
+      emit_tex( bld, inst, TEX_MODIFIER_EXPLICIT_DERIV, dst0 );
       break;
 
    case TGSI_OPCODE_UP2H:
@@ -1466,7 +1503,7 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TXB:
-      emit_tex( bld, inst, TRUE, FALSE, dst0 );
+      emit_tex( bld, inst, TEX_MODIFIER_LOD_BIAS, dst0 );
       break;
 
    case TGSI_OPCODE_NRM:
@@ -1571,11 +1608,11 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TXL:
-      emit_tex( bld, inst, TRUE, FALSE, dst0 );
+      emit_tex( bld, inst, TEX_MODIFIER_EXPLICIT_LOD, dst0 );
       break;
 
    case TGSI_OPCODE_TXP:
-      emit_tex( bld, inst, FALSE, TRUE, dst0 );
+      emit_tex( bld, inst, TEX_MODIFIER_PROJECTED, dst0 );
       break;
 
    case TGSI_OPCODE_BRK:
@@ -1765,10 +1802,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
       switch( parse.FullToken.Token.Type ) {
       case TGSI_TOKEN_TYPE_DECLARATION:
          /* Inputs already interpolated */
-         {
-            if (!emit_declaration( &bld, &parse.FullToken.FullDeclaration ))
-               _debug_printf("warning: failed to define LLVM variable\n");
-         }
+         emit_declaration( &bld, &parse.FullToken.FullDeclaration );
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
@@ -1787,7 +1821,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
          {
             const uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
             assert(size <= 4);
-            assert(num_immediates < LP_MAX_IMMEDIATES);
+            assert(num_immediates < LP_MAX_TGSI_IMMEDIATES);
             for( i = 0; i < size; ++i )
                bld.immediates[num_immediates][i] =
                   lp_build_const_vec(type, parse.FullToken.FullImmediate.u[i].Float);