gallivm/llvmpipe: squash merge of the llvm-context branch

This branch defines a gallivm_state structure which contains the
LLVMBuilderRef, LLVMContextRef, etc.  All data structures built with
this object can be periodically freed during a "garbage collection"
operation.

The gallivm_state object has to be passed to most of the builder
functions where LLVMBuilderRef used to be used.

Conflicts:
	src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
	src/gallium/drivers/llvmpipe/lp_state_setup.c

32 files changed, 692 insertions, 655 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
index e50643790c..518969c320 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c
@@ -43,7 +43,7 @@
 
 
 void
-lp_build_alpha_test(LLVMBuilderRef builder,
+lp_build_alpha_test(struct gallivm_state *gallivm,
                     unsigned func,
                     struct lp_type type,
                     struct lp_build_mask_context *mask,
@@ -54,7 +54,7 @@ lp_build_alpha_test(LLVMBuilderRef builder,
    struct lp_build_context bld;
    LLVMValueRef test;
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
    test = lp_build_cmp(&bld, func, alpha, ref);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h
index 27ca8aad4d..5c9392504f 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h
@@ -43,7 +43,7 @@ struct lp_build_mask_context;
 
 
 void
-lp_build_alpha_test(LLVMBuilderRef builder,
+lp_build_alpha_test(struct gallivm_state *gallivm,
                     unsigned func,
                     struct lp_type type,
                     struct lp_build_mask_context *mask,
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend.h b/src/gallium/drivers/llvmpipe/lp_bld_blend.h
index 5cecec3d7f..f82ae30bb7 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend.h
@@ -30,6 +30,7 @@
 
 
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_init.h"
  
 #include "pipe/p_format.h"
 
@@ -61,7 +62,7 @@ lp_build_blend_func(struct lp_build_context *bld,
 
 
 LLVMValueRef
-lp_build_blend_aos(LLVMBuilderRef builder,
+lp_build_blend_aos(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
@@ -72,7 +73,7 @@ lp_build_blend_aos(LLVMBuilderRef builder,
 
 
 void
-lp_build_blend_soa(LLVMBuilderRef builder,
+lp_build_blend_soa(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
index d1c9b88f9b..c342346a36 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
@@ -301,7 +301,7 @@ lp_build_blend_func(struct lp_build_context *bld,
 
 
 LLVMValueRef
-lp_build_blend_aos(LLVMBuilderRef builder,
+lp_build_blend_aos(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
@@ -322,7 +322,7 @@ lp_build_blend_aos(LLVMBuilderRef builder,
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
-   lp_build_context_init(&bld.base, builder, type);
+   lp_build_context_init(&bld.base, gallivm, type);
    bld.src = src;
    bld.dst = dst;
    bld.const_ = const_;
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
index 30d261e979..4d5bc9642d 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c
@@ -73,6 +73,7 @@
 
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_init.h"
 #include "lp_bld_blend.h"
 
 
@@ -211,7 +212,7 @@ lp_build_blend_factor_complementary(unsigned src_factor, unsigned dst_factor)
  * \param res  the result/output
  */
 void
-lp_build_blend_soa(LLVMBuilderRef builder,
+lp_build_blend_soa(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
                    struct lp_type type,
                    unsigned rt,
@@ -220,6 +221,7 @@ lp_build_blend_soa(LLVMBuilderRef builder,
                    LLVMValueRef con[4],
                    LLVMValueRef res[4])
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_build_blend_soa_context bld;
    unsigned i, j, k;
 
@@ -227,7 +229,7 @@ lp_build_blend_soa(LLVMBuilderRef builder,
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
-   lp_build_context_init(&bld.base, builder, type);
+   lp_build_context_init(&bld.base, gallivm, type);
    for (i = 0; i < 4; ++i) {
       bld.src[i] = src[i];
       bld.dst[i] = dst[i];
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
index 7eb76d4fb3..a1c21fcdaf 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
@@ -107,7 +107,7 @@ lp_build_stencil_test_single(struct lp_build_context *bld,
 
    if (stencil->valuemask != stencilMax) {
       /* compute stencilRef = stencilRef & valuemask */
-      LLVMValueRef valuemask = lp_build_const_int_vec(type, stencil->valuemask);
+      LLVMValueRef valuemask = lp_build_const_int_vec(bld->gallivm, type, stencil->valuemask);
       stencilRef = LLVMBuildAnd(bld->builder, stencilRef, valuemask, "");
       /* compute stencilVals = stencilVals & valuemask */
       stencilVals = LLVMBuildAnd(bld->builder, stencilVals, valuemask, "");
@@ -169,7 +169,7 @@ lp_build_stencil_op_single(struct lp_build_context *bld,
 {
    struct lp_type type = bld->type;
    LLVMValueRef res;
-   LLVMValueRef max = lp_build_const_int_vec(type, 0xff);
+   LLVMValueRef max = lp_build_const_int_vec(bld->gallivm, type, 0xff);
    unsigned stencil_op;
 
    assert(type.sign);
@@ -262,7 +262,8 @@ lp_build_stencil_op(struct lp_build_context *bld,
 
    if (stencil->writemask != 0xff) {
       /* mask &= stencil->writemask */
-      LLVMValueRef writemask = lp_build_const_int_vec(bld->type, stencil->writemask);
+      LLVMValueRef writemask = lp_build_const_int_vec(bld->gallivm, bld->type,
+                                                      stencil->writemask);
       mask = LLVMBuildAnd(bld->builder, mask, writemask, "");
       /* res = (res & mask) | (stencilVals & ~mask) */
       res = lp_build_select_bitwise(bld, writemask, res, stencilVals);
@@ -411,25 +412,27 @@ get_s_shift_and_mask(const struct util_format_description *format_desc,
  * \param counter is a pointer of the uint32 counter.
  */
 void
-lp_build_occlusion_count(LLVMBuilderRef builder,
+lp_build_occlusion_count(struct gallivm_state *gallivm,
                          struct lp_type type,
                          LLVMValueRef maskvalue,
                          LLVMValueRef counter)
 {
-   LLVMValueRef countmask = lp_build_const_int_vec(type, 1);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMContextRef context = gallivm->context;
+   LLVMValueRef countmask = lp_build_const_int_vec(gallivm, type, 1);
    LLVMValueRef countv = LLVMBuildAnd(builder, maskvalue, countmask, "countv");
-   LLVMTypeRef i8v16 = LLVMVectorType(LLVMInt8Type(), 16);
+   LLVMTypeRef i8v16 = LLVMVectorType(LLVMInt8TypeInContext(context), 16);
    LLVMValueRef counti = LLVMBuildBitCast(builder, countv, i8v16, "counti");
    LLVMValueRef maskarray[4] = {
-      LLVMConstInt(LLVMInt32Type(), 0, 0),
-      LLVMConstInt(LLVMInt32Type(), 4, 0),
-      LLVMConstInt(LLVMInt32Type(), 8, 0),
-      LLVMConstInt(LLVMInt32Type(), 12, 0),
+      lp_build_const_int32(gallivm, 0),
+      lp_build_const_int32(gallivm, 4),
+      lp_build_const_int32(gallivm, 8),
+      lp_build_const_int32(gallivm, 12)
    };
    LLVMValueRef shufflemask = LLVMConstVector(maskarray, 4);
    LLVMValueRef shufflev =  LLVMBuildShuffleVector(builder, counti, LLVMGetUndef(i8v16), shufflemask, "shufflev");
-   LLVMValueRef shuffle = LLVMBuildBitCast(builder, shufflev, LLVMInt32Type(), "shuffle");
-   LLVMValueRef count = lp_build_intrinsic_unary(builder, "llvm.ctpop.i32", LLVMInt32Type(), shuffle);
+   LLVMValueRef shuffle = LLVMBuildBitCast(builder, shufflev, LLVMInt32TypeInContext(context), "shuffle");
+   LLVMValueRef count = lp_build_intrinsic_unary(builder, "llvm.ctpop.i32", LLVMInt32TypeInContext(context), shuffle);
    LLVMValueRef orig = LLVMBuildLoad(builder, counter, "orig");
    LLVMValueRef incr = LLVMBuildAdd(builder, orig, count, "incr");
    LLVMBuildStore(builder, incr, counter);
@@ -452,7 +455,7 @@ lp_build_occlusion_count(LLVMBuilderRef builder,
  * \param facing  contains boolean value indicating front/back facing polygon
  */
 void
-lp_build_depth_stencil_test(LLVMBuilderRef builder,
+lp_build_depth_stencil_test(struct gallivm_state *gallivm,
                             const struct pipe_depth_state *depth,
                             const struct pipe_stencil_state stencil[2],
                             struct lp_type z_src_type,
@@ -465,6 +468,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
                             LLVMValueRef *zs_value,
                             boolean do_branch)
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type z_type;
    struct lp_build_context z_bld;
    struct lp_build_context s_bld;
@@ -537,11 +541,11 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
 
    /* Setup build context for Z vals */
-   lp_build_context_init(&z_bld, builder, z_type);
+   lp_build_context_init(&z_bld, gallivm, z_type);
 
    /* Setup build context for stencil vals */
    s_type = lp_type_int_vec(z_type.width);
-   lp_build_context_init(&s_bld, builder, s_type);
+   lp_build_context_init(&s_bld, gallivm, s_type);
 
    /* Load current z/stencil value from z/stencil buffer */
    zs_dst_ptr = LLVMBuildBitCast(builder,
@@ -559,14 +563,14 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
       if (get_z_shift_and_mask(format_desc, &z_shift, &z_width, &z_mask)) {
          if (z_mask != 0xffffffff) {
-            z_bitmask = lp_build_const_int_vec(z_type, z_mask);
+            z_bitmask = lp_build_const_int_vec(gallivm, z_type, z_mask);
          }
 
          /*
           * Align the framebuffer Z 's LSB to the right.
           */
          if (z_shift) {
-            LLVMValueRef shift = lp_build_const_int_vec(z_type, z_shift);
+            LLVMValueRef shift = lp_build_const_int_vec(gallivm, z_type, z_shift);
             z_dst = LLVMBuildLShr(builder, zs_dst, shift, "z_dst");
          } else if (z_bitmask) {
 	    /* TODO: Instead of loading a mask from memory and ANDing, it's
@@ -580,7 +584,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
       if (get_s_shift_and_mask(format_desc, &s_shift, &s_mask)) {
          if (s_shift) {
-            LLVMValueRef shift = lp_build_const_int_vec(s_type, s_shift);
+            LLVMValueRef shift = lp_build_const_int_vec(gallivm, s_type, s_shift);
             stencil_vals = LLVMBuildLShr(builder, zs_dst, shift, "");
             stencil_shift = shift;  /* used below */
          }
@@ -589,7 +593,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          }
 
          if (s_mask != 0xffffffff) {
-            LLVMValueRef mask = lp_build_const_int_vec(s_type, s_mask);
+            LLVMValueRef mask = lp_build_const_int_vec(gallivm, s_type, s_mask);
             stencil_vals = LLVMBuildAnd(builder, stencil_vals, mask, "");
          }
 
@@ -600,12 +604,13 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
    if (stencil[0].enabled) {
 
       if (face) {
-         LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
 
          /* front_facing = face != 0 ? ~0 : 0 */
          front_facing = LLVMBuildICmp(builder, LLVMIntNE, face, zero, "");
          front_facing = LLVMBuildSExt(builder, front_facing,
-                                      LLVMIntType(s_bld.type.length*s_bld.type.width),
+                                      LLVMIntTypeInContext(gallivm->context,
+                                             s_bld.type.length*s_bld.type.width),
                                       "");
          front_facing = LLVMBuildBitCast(builder, front_facing,
                                          s_bld.int_vec_type, "");
@@ -642,7 +647,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
           */
 
          if (!z_type.floating) {
-            z_src = lp_build_clamped_float_to_unsigned_norm(builder,
+            z_src = lp_build_clamped_float_to_unsigned_norm(gallivm,
                                                             z_src_type,
                                                             z_width,
                                                             z_src);
@@ -657,7 +662,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          assert(z_src_type.norm);
          assert(!z_type.floating);
          if (z_src_type.width > z_width) {
-            LLVMValueRef shift = lp_build_const_int_vec(z_src_type,
+            LLVMValueRef shift = lp_build_const_int_vec(gallivm, z_src_type,
                                                         z_src_type.width - z_width);
             z_src = LLVMBuildLShr(builder, z_src, shift, "");
          }
@@ -728,7 +733,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
 
    /* Put Z and ztencil bits in the right place */
    if (z_dst && z_shift) {
-      LLVMValueRef shift = lp_build_const_int_vec(z_type, z_shift);
+      LLVMValueRef shift = lp_build_const_int_vec(gallivm, z_type, z_shift);
       z_dst = LLVMBuildShl(builder, z_dst, shift, "");
    }
    if (stencil_vals && stencil_shift)
@@ -775,7 +780,7 @@ lp_build_depth_write(LLVMBuilderRef builder,
 
 
 void
-lp_build_deferred_depth_write(LLVMBuilderRef builder,
+lp_build_deferred_depth_write(struct gallivm_state *gallivm,
                               struct lp_type z_src_type,
                               const struct util_format_description *format_desc,
                               struct lp_build_mask_context *mask,
@@ -785,11 +790,12 @@ lp_build_deferred_depth_write(LLVMBuilderRef builder,
    struct lp_type z_type;
    struct lp_build_context z_bld;
    LLVMValueRef z_dst;
+   LLVMBuilderRef builder = gallivm->builder;
 
    /* XXX: pointlessly redo type logic:
     */
    z_type = lp_depth_type(format_desc, z_src_type.width*z_src_type.length);
-   lp_build_context_init(&z_bld, builder, z_type);
+   lp_build_context_init(&z_bld, gallivm, z_type);
 
    zs_dst_ptr = LLVMBuildBitCast(builder, zs_dst_ptr,
                                  LLVMPointerType(z_bld.vec_type, 0), "");
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.h b/src/gallium/drivers/llvmpipe/lp_bld_depth.h
index a54ef3a711..038b136a28 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.h
@@ -51,7 +51,7 @@ lp_depth_type(const struct util_format_description *format_desc,
 
 
 void
-lp_build_depth_stencil_test(LLVMBuilderRef builder,
+lp_build_depth_stencil_test(struct gallivm_state *gallivm,
                             const struct pipe_depth_state *depth,
                             const struct pipe_stencil_state stencil[2],
                             struct lp_type type,
@@ -71,7 +71,7 @@ lp_build_depth_write(LLVMBuilderRef builder,
                      LLVMValueRef zs_value);
 
 void
-lp_build_deferred_depth_write(LLVMBuilderRef builder,
+lp_build_deferred_depth_write(struct gallivm_state *gallivm,
                               struct lp_type z_src_type,
                               const struct util_format_description *format_desc,
                               struct lp_build_mask_context *mask,
@@ -79,7 +79,7 @@ lp_build_deferred_depth_write(LLVMBuilderRef builder,
                               LLVMValueRef zs_value);
 
 void
-lp_build_occlusion_count(LLVMBuilderRef builder,
+lp_build_occlusion_count(struct gallivm_state *gallivm,
                          struct lp_type type,
                          LLVMValueRef maskvalue,
                          LLVMValueRef counter);
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.c b/src/gallium/drivers/llvmpipe/lp_bld_interp.c
index c9da8900d0..e61c3b86a6 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_interp.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.c
@@ -128,12 +128,13 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
 {
    struct lp_build_context *coeff_bld = &bld->coeff_bld;
    LLVMBuilderRef builder = coeff_bld->builder;
+   struct gallivm_state *gallivm = coeff_bld->gallivm;
    LLVMValueRef zero = LLVMConstNull(coeff_bld->elem_type);
    LLVMValueRef one = LLVMConstReal(coeff_bld->elem_type, 1.0);
-   LLVMValueRef i0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef i1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef i2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-   LLVMValueRef i3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   LLVMValueRef i0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef i1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef i2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef i3 = lp_build_const_int32(gallivm, 3);
    unsigned attrib;
    unsigned chan;
 
@@ -144,7 +145,8 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
       const unsigned interp = bld->interp[attrib];
       for (chan = 0; chan < NUM_CHANNELS; ++chan) {
          if (mask & (1 << chan)) {
-            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), attrib*NUM_CHANNELS + chan, 0);
+            LLVMValueRef index = lp_build_const_int32(gallivm,
+                                      attrib * NUM_CHANNELS + chan);
             LLVMValueRef a0 = zero;
             LLVMValueRef dadx = zero;
             LLVMValueRef dady = zero;
@@ -231,7 +233,7 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
              * a = {a, a, a, a}
              */
 
-            a = lp_build_broadcast(builder, coeff_bld->vec_type, a);
+            a = lp_build_broadcast(gallivm, coeff_bld->vec_type, a);
 
             /*
              * Compute the attrib values on the upper-left corner of each quad.
@@ -273,12 +275,13 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
  */
 static void
 attribs_update(struct lp_build_interp_soa_context *bld,
+               struct gallivm_state *gallivm,
                int quad_index,
                int start,
                int end)
 {
    struct lp_build_context *coeff_bld = &bld->coeff_bld;
-   LLVMValueRef shuffle = lp_build_const_int_vec(coeff_bld->type, quad_index);
+   LLVMValueRef shuffle = lp_build_const_int_vec(gallivm, coeff_bld->type, quad_index);
    LLVMValueRef oow = NULL;
    unsigned attrib;
    unsigned chan;
@@ -392,6 +395,7 @@ pos_init(struct lp_build_interp_soa_context *bld,
  */
 void
 lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
+                         struct gallivm_state *gallivm,
                          unsigned num_inputs,
                          const struct lp_shader_input *inputs,
                          LLVMBuilderRef builder,
@@ -417,7 +421,7 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
    /* XXX: we don't support interpolating into any other types */
    assert(memcmp(&coeff_type, &type, sizeof coeff_type) == 0);
 
-   lp_build_context_init(&bld->coeff_bld, builder, coeff_type);
+   lp_build_context_init(&bld->coeff_bld, gallivm, coeff_type);
 
    /* For convenience */
    bld->pos = bld->attribs[0];
@@ -453,19 +457,21 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
  */
 void
 lp_build_interp_soa_update_inputs(struct lp_build_interp_soa_context *bld,
+                                  struct gallivm_state *gallivm,
                                   int quad_index)
 {
    assert(quad_index < 4);
 
-   attribs_update(bld, quad_index, 1, bld->num_attribs);
+   attribs_update(bld, gallivm, quad_index, 1, bld->num_attribs);
 }
 
 void
 lp_build_interp_soa_update_pos(struct lp_build_interp_soa_context *bld,
+                                  struct gallivm_state *gallivm,
                                   int quad_index)
 {
    assert(quad_index < 4);
 
-   attribs_update(bld, quad_index, 0, 1);
+   attribs_update(bld, gallivm, quad_index, 0, 1);
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.h b/src/gallium/drivers/llvmpipe/lp_bld_interp.h
index a7ebdd1bfa..b58b2dc115 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_interp.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.h
@@ -102,6 +102,7 @@ struct lp_build_interp_soa_context
 
 void
 lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
+                         struct gallivm_state *gallivm,
                          unsigned num_inputs,
                          const struct lp_shader_input *inputs,
                          LLVMBuilderRef builder,
@@ -114,11 +115,13 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
 
 void
 lp_build_interp_soa_update_inputs(struct lp_build_interp_soa_context *bld,
-                           int quad_index);
+                                  struct gallivm_state *gallivm,
+                                  int quad_index);
 
 void
 lp_build_interp_soa_update_pos(struct lp_build_interp_soa_context *bld,
-                           int quad_index);
+                               struct gallivm_state *gallivm,
+                               int quad_index);
 
 
 #endif /* LP_BLD_INTERP_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index 763432ed71..2de20d6e9a 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -50,6 +50,46 @@
 DEBUG_GET_ONCE_BOOL_OPTION(lp_no_rast, "LP_NO_RAST", FALSE)
 
 
+/** shared by all contexts */
+unsigned llvmpipe_variant_count;
+
+
+/**
+ * This function is called by the gallivm "garbage collector" when
+ * the LLVM global data structures are freed.  We must free all LLVM-related
+ * data.  Specifically, all JIT'd shader variants.
+ */
+static void
+garbage_collect_callback(void *cb_data)
+{
+   struct llvmpipe_context *lp = (struct llvmpipe_context *) cb_data;
+   struct lp_fs_variant_list_item *li;
+
+   /* Free all the context's shader variants */
+   li = first_elem(&lp->fs_variants_list);
+   while (!at_end(&lp->fs_variants_list, li)) {
+      struct lp_fs_variant_list_item *next = next_elem(li);
+      llvmpipe_remove_shader_variant(lp, li->base);
+      li = next;
+   }
+
+   /* Free all the context's primitive setup variants */
+   lp_delete_setup_variants(lp);
+
+   /* release references to setup variants, shaders */
+   lp_setup_set_setup_variant(lp->setup, NULL);
+   lp_setup_set_fs_variant(lp->setup, NULL);
+   lp_setup_reset(lp->setup);
+
+   /* This type will be recreated upon demand */
+   lp->jit_context_ptr_type = NULL;
+
+   /* mark all state as dirty to ensure new shaders are jit'd, etc. */
+   lp->dirty = ~0;
+}
+
+
+
 static void llvmpipe_destroy( struct pipe_context *pipe )
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context( pipe );
@@ -57,6 +97,9 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
 
    lp_print_counters();
 
+   gallivm_remove_garbage_collector_callback(garbage_collect_callback,
+                                             llvmpipe);
+
    /* This will also destroy llvmpipe->setup:
     */
    if (llvmpipe->draw)
@@ -82,7 +125,7 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
       }
    }
 
-   lp_delete_setup_variants(llvmpipe);
+   gallivm_destroy(llvmpipe->gallivm);
 
    align_free( llvmpipe );
 }
@@ -110,8 +153,10 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    memset(llvmpipe, 0, sizeof *llvmpipe);
 
    make_empty_list(&llvmpipe->fs_variants_list);
+
    make_empty_list(&llvmpipe->setup_variants_list);
 
+
    llvmpipe->pipe.winsys = screen->winsys;
    llvmpipe->pipe.screen = screen;
    llvmpipe->pipe.priv = priv;
@@ -136,10 +181,12 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe_init_context_resource_funcs( &llvmpipe->pipe );
    llvmpipe_init_surface_functions(llvmpipe);
 
+   llvmpipe->gallivm = gallivm_create();
+
    /*
     * Create drawing context and plug our rendering stage into it.
     */
-   llvmpipe->draw = draw_create(&llvmpipe->pipe);
+   llvmpipe->draw = draw_create_gallivm(&llvmpipe->pipe, llvmpipe->gallivm);
    if (!llvmpipe->draw)
       goto fail;
 
@@ -173,6 +220,9 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
 
    lp_reset_counters();
 
+   gallivm_register_garbage_collector_callback(garbage_collect_callback,
+                                               llvmpipe);
+
    return &llvmpipe->pipe;
 
  fail:
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index a35e09e8b4..503f09d810 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -126,14 +126,27 @@ struct llvmpipe_context {
    unsigned tex_timestamp;
    boolean no_rast;
 
+   /** List of all fragment shader variants */
    struct lp_fs_variant_list_item fs_variants_list;
    unsigned nr_fs_variants;
 
+   /** JIT code generation */
+   struct gallivm_state *gallivm;
+   LLVMTypeRef jit_context_ptr_type;
+
    struct lp_setup_variant_list_item setup_variants_list;
    unsigned nr_setup_variants;
 };
 
 
+/**
+ * Fragment and setup variant count, used to trigger garbage collection.
+ * This is global since all variants in all contexts will be free when
+ * we do garbage collection.
+ */
+extern unsigned llvmpipe_variant_count;
+
+
 struct pipe_context *
 llvmpipe_create_context( struct pipe_screen *screen, void *priv );
 
diff --git a/src/gallium/drivers/llvmpipe/lp_flush.c b/src/gallium/drivers/llvmpipe/lp_flush.c
index e2c723b7a8..e8d00cf516 100644
--- a/src/gallium/drivers/llvmpipe/lp_flush.c
+++ b/src/gallium/drivers/llvmpipe/lp_flush.c
@@ -56,6 +56,13 @@ llvmpipe_flush( struct pipe_context *pipe,
    /* ask the setup module to flush */
    lp_setup_flush(llvmpipe->setup, flags, fence, reason);
 
+
+   if (llvmpipe_variant_count > 1000) {
+      /* time to do a garbage collection */
+      gallivm_garbage_collect(llvmpipe->gallivm);
+      llvmpipe_variant_count = 0;
+   }
+
    /* Enable to dump BMPs of the color/depth buffers each frame */
    if (0) {
       if (flags & PIPE_FLUSH_FRAME) {
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index c540f9b362..a775990f92 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -33,82 +33,85 @@
  */
 
 
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "util/u_memory.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_debug.h"
-#include "lp_screen.h"
 #include "gallivm/lp_bld_intr.h"
+#include "lp_context.h"
+#include "lp_screen.h"
 #include "lp_jit.h"
 
 
 static void
-lp_jit_init_globals(struct llvmpipe_screen *screen)
+lp_jit_create_types(struct llvmpipe_context *lp)
 {
+   struct gallivm_state *gallivm = lp->gallivm;
+   LLVMContextRef lc = gallivm->context;
    LLVMTypeRef texture_type;
 
    /* struct lp_jit_texture */
    {
       LLVMTypeRef elem_types[LP_JIT_TEXTURE_NUM_FIELDS];
 
-      elem_types[LP_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_WIDTH]  =
+      elem_types[LP_JIT_TEXTURE_HEIGHT] =
+      elem_types[LP_JIT_TEXTURE_DEPTH] =
+      elem_types[LP_JIT_TEXTURE_LAST_LEVEL] =  LLVMInt32TypeInContext(lc);
       elem_types[LP_JIT_TEXTURE_ROW_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), LP_MAX_TEXTURE_LEVELS);
       elem_types[LP_JIT_TEXTURE_IMG_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), LP_MAX_TEXTURE_LEVELS);
+         LLVMArrayType(LLVMInt32TypeInContext(lc), LP_MAX_TEXTURE_LEVELS);
       elem_types[LP_JIT_TEXTURE_DATA] =
-         LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
+         LLVMArrayType(LLVMPointerType(LLVMInt8TypeInContext(lc), 0),
                        LP_MAX_TEXTURE_LEVELS);
-      elem_types[LP_JIT_TEXTURE_MIN_LOD] = LLVMFloatType();
-      elem_types[LP_JIT_TEXTURE_MAX_LOD] = LLVMFloatType();
-      elem_types[LP_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType();
+      elem_types[LP_JIT_TEXTURE_MIN_LOD] =
+      elem_types[LP_JIT_TEXTURE_MAX_LOD] =
+      elem_types[LP_JIT_TEXTURE_LOD_BIAS] = LLVMFloatTypeInContext(lc);
       elem_types[LP_JIT_TEXTURE_BORDER_COLOR] = 
-         LLVMArrayType(LLVMFloatType(), 4);
+         LLVMArrayType(LLVMFloatTypeInContext(lc), 4);
 
-      texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+      texture_type = LLVMStructTypeInContext(lc, elem_types,
+                                             Elements(elem_types), 0);
+
+      LLVMInvalidateStructLayout(gallivm->target, texture_type);
 
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, width,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_WIDTH);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, height,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_HEIGHT);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, depth,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_DEPTH);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, last_level,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_LAST_LEVEL);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, row_stride,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_ROW_STRIDE);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, img_stride,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_IMG_STRIDE);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, data,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_DATA);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, min_lod,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_MIN_LOD);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, max_lod,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_MAX_LOD);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, lod_bias,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_LOD_BIAS);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, border_color,
-                             screen->target, texture_type,
+                             gallivm->target, texture_type,
                              LP_JIT_TEXTURE_BORDER_COLOR);
 
       LP_CHECK_STRUCT_SIZE(struct lp_jit_texture,
-                           screen->target, texture_type);
+                           gallivm->target, texture_type);
 
-      LLVMAddTypeName(screen->module, "texture", texture_type);
+      LLVMAddTypeName(gallivm->module, "texture", texture_type);
    }
 
    /* struct lp_jit_context */
@@ -116,44 +119,47 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
       LLVMTypeRef elem_types[LP_JIT_CTX_COUNT];
       LLVMTypeRef context_type;
 
-      elem_types[LP_JIT_CTX_CONSTANTS] = LLVMPointerType(LLVMFloatType(), 0);
-      elem_types[LP_JIT_CTX_ALPHA_REF] = LLVMFloatType();
-      elem_types[LP_JIT_CTX_STENCIL_REF_FRONT] = LLVMInt32Type();
-      elem_types[LP_JIT_CTX_STENCIL_REF_BACK] = LLVMInt32Type();
-      elem_types[LP_JIT_CTX_BLEND_COLOR] = LLVMPointerType(LLVMInt8Type(), 0);
+      elem_types[LP_JIT_CTX_CONSTANTS] = LLVMPointerType(LLVMFloatTypeInContext(lc), 0);
+      elem_types[LP_JIT_CTX_ALPHA_REF] = LLVMFloatTypeInContext(lc);
+      elem_types[LP_JIT_CTX_STENCIL_REF_FRONT] =
+      elem_types[LP_JIT_CTX_STENCIL_REF_BACK] = LLVMInt32TypeInContext(lc);
+      elem_types[LP_JIT_CTX_BLEND_COLOR] = LLVMPointerType(LLVMInt8TypeInContext(lc), 0);
       elem_types[LP_JIT_CTX_TEXTURES] = LLVMArrayType(texture_type,
                                                       PIPE_MAX_SAMPLERS);
 
-      context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+      context_type = LLVMStructTypeInContext(lc, elem_types,
+                                             Elements(elem_types), 0);
+
+      LLVMInvalidateStructLayout(gallivm->target, context_type);
 
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, constants,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_CONSTANTS);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, alpha_ref_value,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_ALPHA_REF);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, stencil_ref_front,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_STENCIL_REF_FRONT);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, stencil_ref_back,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_STENCIL_REF_BACK);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, blend_color,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_BLEND_COLOR);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, textures,
-                             screen->target, context_type,
+                             gallivm->target, context_type,
                              LP_JIT_CTX_TEXTURES);
       LP_CHECK_STRUCT_SIZE(struct lp_jit_context,
-                           screen->target, context_type);
+                           gallivm->target, context_type);
 
-      LLVMAddTypeName(screen->module, "context", context_type);
+      LLVMAddTypeName(gallivm->module, "context", context_type);
 
-      screen->context_ptr_type = LLVMPointerType(context_type, 0);
+      lp->jit_context_ptr_type = LLVMPointerType(context_type, 0);
    }
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
-      LLVMDumpModule(screen->module);
+      LLVMDumpModule(gallivm->module);
    }
 }
 
@@ -161,8 +167,7 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
 void
 lp_jit_screen_cleanup(struct llvmpipe_screen *screen)
 {
-   if(screen->pass)
-      LLVMDisposePassManager(screen->pass);
+   /* nothing */
 }
 
 
@@ -170,30 +175,14 @@ void
 lp_jit_screen_init(struct llvmpipe_screen *screen)
 {
    lp_build_init();
+}
 
-   screen->module = lp_build_module;
-   screen->provider = lp_build_provider;
-   screen->engine = lp_build_engine;
-   screen->target = lp_build_target;
-
-   screen->pass = LLVMCreateFunctionPassManager(screen->provider);
-   LLVMAddTargetData(screen->target, screen->pass);
-
-   if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
-      /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-       * but there are more on SVN. */
-      /* TODO: Add more passes */
-      LLVMAddCFGSimplificationPass(screen->pass);
-      LLVMAddPromoteMemoryToRegisterPass(screen->pass);
-      LLVMAddConstantPropagationPass(screen->pass);
-      LLVMAddInstructionCombiningPass(screen->pass);
-      LLVMAddGVNPass(screen->pass);
-   } else {
-      /* We need at least this pass to prevent the backends to fail in
-       * unexpected ways.
-       */
-      LLVMAddPromoteMemoryToRegisterPass(screen->pass);
-   }
 
-   lp_jit_init_globals(screen);
+LLVMTypeRef
+lp_jit_get_context_type(struct llvmpipe_context *lp)
+{
+   if (!lp->jit_context_ptr_type)
+      lp_jit_create_types(lp);
+
+   return lp->jit_context_ptr_type;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index 114f21f2d1..a6763dce17 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -120,23 +120,23 @@ enum {
 };
 
 
-#define lp_jit_context_constants(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_CONSTANTS, "constants")
+#define lp_jit_context_constants(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_CONSTANTS, "constants")
 
-#define lp_jit_context_alpha_ref_value(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_ALPHA_REF, "alpha_ref_value")
+#define lp_jit_context_alpha_ref_value(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_ALPHA_REF, "alpha_ref_value")
 
-#define lp_jit_context_stencil_ref_front_value(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_STENCIL_REF_FRONT, "stencil_ref_front")
+#define lp_jit_context_stencil_ref_front_value(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_STENCIL_REF_FRONT, "stencil_ref_front")
 
-#define lp_jit_context_stencil_ref_back_value(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_STENCIL_REF_BACK, "stencil_ref_back")
+#define lp_jit_context_stencil_ref_back_value(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_STENCIL_REF_BACK, "stencil_ref_back")
 
-#define lp_jit_context_blend_color(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, LP_JIT_CTX_BLEND_COLOR, "blend_color")
+#define lp_jit_context_blend_color(_gallivm, _ptr) \
+   lp_build_struct_get(_gallivm, _ptr, LP_JIT_CTX_BLEND_COLOR, "blend_color")
 
-#define lp_jit_context_textures(_builder, _ptr) \
-   lp_build_struct_get_ptr(_builder, _ptr, LP_JIT_CTX_TEXTURES, "textures")
+#define lp_jit_context_textures(_gallivm, _ptr) \
+   lp_build_struct_get_ptr(_gallivm, _ptr, LP_JIT_CTX_TEXTURES, "textures")
 
 
 
@@ -162,4 +162,8 @@ void
 lp_jit_screen_init(struct llvmpipe_screen *screen);
 
 
+LLVMTypeRef
+lp_jit_get_context_type(struct llvmpipe_context *lp);
+
+
 #endif /* LP_JIT_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index decf3bd449..dd6e6d566b 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -47,6 +47,7 @@
 #ifdef DEBUG
 int jit_line = 0;
 const struct lp_rast_state *jit_state = NULL;
+const struct lp_rasterizer_task *jit_task = NULL;
 #endif
 
 
@@ -362,7 +363,7 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
          depth = lp_rast_get_depth_block_pointer(task, tile_x + x, tile_y + y);
 
          /* run shader on 4x4 block */
-         BEGIN_JIT_CALL(state);
+         BEGIN_JIT_CALL(state, task);
          variant->jit_function[RAST_WHOLE]( &state->jit_context,
                                             tile_x + x, tile_y + y,
                                             inputs->frontfacing,
@@ -443,7 +444,7 @@ lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
    assert(lp_check_alignment(state->jit_context.blend_color, 16));
 
    /* run shader on 4x4 block */
-   BEGIN_JIT_CALL(state);
+   BEGIN_JIT_CALL(state, task);
    variant->jit_function[RAST_EDGE_TEST](&state->jit_context,
                                          x, y,
                                          inputs->frontfacing,
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
index b30408f097..6864aeea78 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
@@ -45,13 +45,16 @@
  */
 #ifdef DEBUG
 
+struct lp_rasterizer_task;
 extern int jit_line;
 extern const struct lp_rast_state *jit_state;
+extern const struct lp_rasterizer_task *jit_task;
 
-#define BEGIN_JIT_CALL(state) \
+#define BEGIN_JIT_CALL(state, task)                  \
    do { \
       jit_line = __LINE__; \
       jit_state = state; \
+      jit_task = task; \
    } while (0)
 
 #define END_JIT_CALL() \
@@ -62,7 +65,7 @@ extern const struct lp_rast_state *jit_state;
 
 #else
 
-#define BEGIN_JIT_CALL(X)
+#define BEGIN_JIT_CALL(X, Y)
 #define END_JIT_CALL()
 
 #endif
@@ -258,7 +261,7 @@ lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
    depth = lp_rast_get_depth_block_pointer(task, x, y);
 
    /* run shader on 4x4 block */
-   BEGIN_JIT_CALL(state);
+   BEGIN_JIT_CALL(state, task);
    variant->jit_function[RAST_WHOLE]( &state->jit_context,
                                       x, y,
                                       inputs->frontfacing,
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.h b/src/gallium/drivers/llvmpipe/lp_screen.h
index 731526dfab..7f69a11a6e 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.h
+++ b/src/gallium/drivers/llvmpipe/lp_screen.h
@@ -34,12 +34,10 @@
 #ifndef LP_SCREEN_H
 #define LP_SCREEN_H
 
-#include "gallivm/lp_bld.h"
-#include <llvm-c/ExecutionEngine.h>
-
-#include "os/os_thread.h"
 #include "pipe/p_screen.h"
 #include "pipe/p_defines.h"
+#include "os/os_thread.h"
+#include "gallivm/lp_bld.h"
 
 
 struct sw_winsys;
@@ -51,14 +49,6 @@ struct llvmpipe_screen
 
    struct sw_winsys *winsys;
 
-   LLVMModuleRef module;
-   LLVMExecutionEngineRef engine;
-   LLVMModuleProviderRef provider;
-   LLVMTargetDataRef target;
-   LLVMPassManagerRef pass;
-
-   LLVMTypeRef context_ptr_type;
-
    unsigned num_threads;
 
    /* Increments whenever textures are modified.  Contexts can track this.
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c
index a173e71aba..db04c84efb 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup.c
@@ -114,7 +114,7 @@ first_point( struct lp_setup_context *setup,
    setup->point( setup, v0 );
 }
 
-static void lp_setup_reset( struct lp_setup_context *setup )
+void lp_setup_reset( struct lp_setup_context *setup )
 {
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
@@ -912,6 +912,12 @@ lp_setup_update_state( struct lp_setup_context *setup,
          llvmpipe_update_derived(lp);
       }
 
+      if (lp->setup->dirty) {
+         llvmpipe_update_setup(lp);
+      }
+
+      assert(setup->setup.variant);
+
       /* Will probably need to move this somewhere else, just need  
        * to know about vertex shader point size attribute.
        */
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.h b/src/gallium/drivers/llvmpipe/lp_setup.h
index ebb18f8134..0d6e161a21 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_setup.h
@@ -45,6 +45,9 @@ struct lp_jit_context;
 struct llvmpipe_query;
 struct pipe_fence_handle;
 struct lp_setup_variant;
+struct lp_setup_context;
+
+void lp_setup_reset( struct lp_setup_context *setup );
 
 struct lp_setup_context *
 lp_setup_create( struct pipe_context *pipe,
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
index 9c1f0fe793..384242f81d 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
@@ -141,6 +141,8 @@ lp_setup_draw_elements(struct vbuf_render *vbr, const ushort *indices, uint nr)
    const boolean flatshade_first = setup->flatshade_first;
    unsigned i;
 
+   assert(setup->setup.variant);
+
    if (!lp_setup_update_state(setup, TRUE))
       return;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 48971510f2..2c4943a69f 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -102,10 +102,10 @@
 #include <llvm-c/BitWriter.h>
 
 
+/** Fragment shader number (for debugging) */
 static unsigned fs_no = 0;
 
 
-
 /**
  * Expand the relevent bits of mask_input to a 4-dword mask for the 
  * four pixels in a 2x2 quad.  This will set the four elements of the
@@ -115,13 +115,14 @@ static unsigned fs_no = 0;
  * \param mask_input  bitwise mask for the whole 4x4 stamp
  */
 static LLVMValueRef
-generate_quad_mask(LLVMBuilderRef builder,
+generate_quad_mask(struct gallivm_state *gallivm,
                    struct lp_type fs_type,
                    unsigned quad,
                    LLVMValueRef mask_input) /* int32 */
 {
+   LLVMBuilderRef builder = gallivm->builder;
    struct lp_type mask_type;
-   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
    LLVMValueRef bits[4];
    LLVMValueRef mask;
    int shift;
@@ -136,7 +137,6 @@ generate_quad_mask(LLVMBuilderRef builder,
    /*
     * mask_input >>= (quad * 4)
     */
-   
    switch (quad) {
    case 0:
       shift = 0;
@@ -163,8 +163,9 @@ generate_quad_mask(LLVMBuilderRef builder,
    /*
     * mask = { mask_input & (1 << i), for i in [0,3] }
     */
-
-   mask = lp_build_broadcast(builder, lp_build_vec_type(mask_type), mask_input);
+   mask = lp_build_broadcast(gallivm,
+                             lp_build_vec_type(gallivm, mask_type),
+                             mask_input);
 
    bits[0] = LLVMConstInt(i32t, 1 << 0, 0);
    bits[1] = LLVMConstInt(i32t, 1 << 1, 0);
@@ -176,11 +177,10 @@ generate_quad_mask(LLVMBuilderRef builder,
    /*
     * mask = mask != 0 ? ~0 : 0
     */
-
-   mask = lp_build_compare(builder,
+   mask = lp_build_compare(gallivm,
                            mask_type, PIPE_FUNC_NOTEQUAL,
                            mask,
-                           lp_build_const_int_vec(mask_type, 0));
+                           lp_build_const_int_vec(gallivm, mask_type, 0));
 
    return mask;
 }
@@ -213,7 +213,8 @@ find_output_by_semantic( const struct tgsi_shader_info *info,
  * \param partial_mask  if 1, do mask_input testing
  */
 static void
-generate_fs(struct lp_fragment_shader *shader,
+generate_fs(struct gallivm_state *gallivm,
+            struct lp_fragment_shader *shader,
             const struct lp_fragment_shader_variant_key *key,
             LLVMBuilderRef builder,
             struct lp_type type,
@@ -278,42 +279,42 @@ generate_fs(struct lp_fragment_shader *shader,
 
    assert(i < 4);
 
-   stencil_refs[0] = lp_jit_context_stencil_ref_front_value(builder, context_ptr);
-   stencil_refs[1] = lp_jit_context_stencil_ref_back_value(builder, context_ptr);
+   stencil_refs[0] = lp_jit_context_stencil_ref_front_value(gallivm, context_ptr);
+   stencil_refs[1] = lp_jit_context_stencil_ref_back_value(gallivm, context_ptr);
 
-   vec_type = lp_build_vec_type(type);
+   vec_type = lp_build_vec_type(gallivm, type);
 
-   consts_ptr = lp_jit_context_constants(builder, context_ptr);
+   consts_ptr = lp_jit_context_constants(gallivm, context_ptr);
 
    memset(outputs, 0, sizeof outputs);
 
    /* Declare the color and z variables */
    for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
       for(chan = 0; chan < NUM_CHANNELS; ++chan) {
-	 color[cbuf][chan] = lp_build_alloca(builder, vec_type, "color");
+	 color[cbuf][chan] = lp_build_alloca(gallivm, vec_type, "color");
       }
    }
 
    /* do triangle edge testing */
    if (partial_mask) {
-      *pmask = generate_quad_mask(builder, type,
+      *pmask = generate_quad_mask(gallivm, type,
                                   i, mask_input);
    }
    else {
-      *pmask = lp_build_const_int_vec(type, ~0);
+      *pmask = lp_build_const_int_vec(gallivm, type, ~0);
    }
 
    /* 'mask' will control execution based on quad's pixel alive/killed state */
-   lp_build_mask_begin(&mask, builder, type, *pmask);
+   lp_build_mask_begin(&mask, gallivm, type, *pmask);
 
    if (!(depth_mode & EARLY_DEPTH_TEST) && !simple_shader)
       lp_build_mask_check(&mask);
 
-   lp_build_interp_soa_update_pos(interp, i);
+   lp_build_interp_soa_update_pos(interp, gallivm, i);
    z = interp->pos[2];
 
    if (depth_mode & EARLY_DEPTH_TEST) {
-      lp_build_depth_stencil_test(builder,
+      lp_build_depth_stencil_test(gallivm,
                                   &key->depth,
                                   key->stencil,
                                   type,
@@ -330,14 +331,13 @@ generate_fs(struct lp_fragment_shader *shader,
       }
    }
 
-   lp_build_interp_soa_update_inputs(interp, i);
+   lp_build_interp_soa_update_inputs(interp, gallivm, i);
    
    /* Build the actual shader */
-   lp_build_tgsi_soa(builder, tokens, type, &mask,
+   lp_build_tgsi_soa(gallivm, tokens, type, &mask,
                      consts_ptr, interp->pos, interp->inputs,
                      outputs, sampler, &shader->info.base);
 
-
    /* Alpha test */
    if (key->alpha.enabled) {
       int color0 = find_output_by_semantic(&shader->info.base,
@@ -348,10 +348,10 @@ generate_fs(struct lp_fragment_shader *shader,
          LLVMValueRef alpha = LLVMBuildLoad(builder, outputs[color0][3], "alpha");
          LLVMValueRef alpha_ref_value;
 
-         alpha_ref_value = lp_jit_context_alpha_ref_value(builder, context_ptr);
-         alpha_ref_value = lp_build_broadcast(builder, vec_type, alpha_ref_value);
+         alpha_ref_value = lp_jit_context_alpha_ref_value(gallivm, context_ptr);
+         alpha_ref_value = lp_build_broadcast(gallivm, vec_type, alpha_ref_value);
 
-         lp_build_alpha_test(builder, key->alpha.func, type,
+         lp_build_alpha_test(gallivm, key->alpha.func, type,
                              &mask, alpha, alpha_ref_value,
                              (depth_mode & LATE_DEPTH_TEST) != 0);
       }
@@ -367,7 +367,7 @@ generate_fs(struct lp_fragment_shader *shader,
          z = LLVMBuildLoad(builder, outputs[pos0][2], "output.z");
       }
 
-      lp_build_depth_stencil_test(builder,
+      lp_build_depth_stencil_test(gallivm,
                                   &key->depth,
                                   key->stencil,
                                   type,
@@ -390,7 +390,7 @@ generate_fs(struct lp_fragment_shader *shader,
        * depth value, update from zs_value with the new mask value and
        * write that out.
        */
-      lp_build_deferred_depth_write(builder,
+      lp_build_deferred_depth_write(gallivm,
                                     type,
                                     zs_format_desc,
                                     &mask,
@@ -420,7 +420,7 @@ generate_fs(struct lp_fragment_shader *shader,
    }
 
    if (counter)
-      lp_build_occlusion_count(builder, type,
+      lp_build_occlusion_count(gallivm, type,
                                lp_build_mask_value(&mask), counter);
 
    *pmask = lp_build_mask_end(&mask);
@@ -437,7 +437,8 @@ generate_fs(struct lp_fragment_shader *shader,
  * \param dst_ptr  the destination color buffer pointer
  */
 static void
-generate_blend(const struct pipe_blend_state *blend,
+generate_blend(struct gallivm_state *gallivm,
+               const struct pipe_blend_state *blend,
                unsigned rt,
                LLVMBuilderRef builder,
                struct lp_type type,
@@ -456,21 +457,21 @@ generate_blend(const struct pipe_blend_state *blend,
    LLVMValueRef res[4];
    unsigned chan;
 
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&bld, gallivm, type);
 
-   lp_build_mask_begin(&mask_ctx, builder, type, mask);
+   lp_build_mask_begin(&mask_ctx, gallivm, type, mask);
    if (do_branch)
       lp_build_mask_check(&mask_ctx);
 
-   vec_type = lp_build_vec_type(type);
+   vec_type = lp_build_vec_type(gallivm, type);
 
-   const_ptr = lp_jit_context_blend_color(builder, context_ptr);
+   const_ptr = lp_jit_context_blend_color(gallivm, context_ptr);
    const_ptr = LLVMBuildBitCast(builder, const_ptr,
                                 LLVMPointerType(vec_type, 0), "");
 
    /* load constant blend color and colors from the dest color buffer */
    for(chan = 0; chan < 4; ++chan) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, chan);
       con[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), "");
 
       dst[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), "");
@@ -480,12 +481,12 @@ generate_blend(const struct pipe_blend_state *blend,
    }
 
    /* do blend */
-   lp_build_blend_soa(builder, blend, type, rt, src, dst, con, res);
+   lp_build_blend_soa(gallivm, blend, type, rt, src, dst, con, res);
 
    /* store results to color buffer */
    for(chan = 0; chan < 4; ++chan) {
       if(blend->rt[rt].colormask & (1 << chan)) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
+         LLVMValueRef index = lp_build_const_int32(gallivm, chan);
          lp_build_name(res[chan], "res.%c", "rgba"[chan]);
          res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
          LLVMBuildStore(builder, res[chan], LLVMBuildGEP(builder, dst_ptr, &index, 1, ""));
@@ -503,11 +504,12 @@ generate_blend(const struct pipe_blend_state *blend,
  * 2x2 pixels.
  */
 static void
-generate_fragment(struct llvmpipe_screen *screen,
+generate_fragment(struct llvmpipe_context *lp,
                   struct lp_fragment_shader *shader,
                   struct lp_fragment_shader_variant *variant,
                   unsigned partial_mask)
 {
+   struct gallivm_state *gallivm = lp->gallivm;
    const struct lp_fragment_shader_variant_key *key = &variant->key;
    struct lp_shader_input inputs[PIPE_MAX_SHADER_INPUTS];
    char func_name[256];
@@ -518,6 +520,8 @@ generate_fragment(struct llvmpipe_screen *screen,
    LLVMTypeRef blend_vec_type;
    LLVMTypeRef arg_types[11];
    LLVMTypeRef func_type;
+   LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
+   LLVMTypeRef int8_type = LLVMInt8TypeInContext(gallivm->context);
    LLVMValueRef context_ptr;
    LLVMValueRef x;
    LLVMValueRef y;
@@ -579,29 +583,30 @@ generate_fragment(struct llvmpipe_screen *screen,
     * lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
     */
 
-   fs_elem_type = lp_build_elem_type(fs_type);
-   fs_int_vec_type = lp_build_int_vec_type(fs_type);
+   fs_elem_type = lp_build_elem_type(gallivm, fs_type);
+   fs_int_vec_type = lp_build_int_vec_type(gallivm, fs_type);
 
-   blend_vec_type = lp_build_vec_type(blend_type);
+   blend_vec_type = lp_build_vec_type(gallivm, blend_type);
 
    util_snprintf(func_name, sizeof(func_name), "fs%u_variant%u_%s", 
 		 shader->no, variant->no, partial_mask ? "partial" : "whole");
 
-   arg_types[0] = screen->context_ptr_type;            /* context */
-   arg_types[1] = LLVMInt32Type();                     /* x */
-   arg_types[2] = LLVMInt32Type();                     /* y */
-   arg_types[3] = LLVMInt32Type();                     /* facing */
+   arg_types[0] = lp_jit_get_context_type(lp);         /* context */
+   arg_types[1] = int32_type;                          /* x */
+   arg_types[2] = int32_type;                          /* y */
+   arg_types[3] = int32_type;                          /* facing */
    arg_types[4] = LLVMPointerType(fs_elem_type, 0);    /* a0 */
    arg_types[5] = LLVMPointerType(fs_elem_type, 0);    /* dadx */
    arg_types[6] = LLVMPointerType(fs_elem_type, 0);    /* dady */
    arg_types[7] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0);  /* color */
-   arg_types[8] = LLVMPointerType(LLVMInt8Type(), 0);  /* depth */
-   arg_types[9] = LLVMInt32Type();                     /* mask_input */
-   arg_types[10] = LLVMPointerType(LLVMInt32Type(), 0);/* counter */
+   arg_types[8] = LLVMPointerType(int8_type, 0);       /* depth */
+   arg_types[9] = int32_type;                          /* mask_input */
+   arg_types[10] = LLVMPointerType(int32_type, 0);     /* counter */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
+                                arg_types, Elements(arg_types), 0);
 
-   function = LLVMAddFunction(screen->module, func_name, func_type);
+   function = LLVMAddFunction(gallivm->module, func_name, func_type);
    LLVMSetFunctionCallConv(function, LLVMCCallConv);
 
    variant->function[partial_mask] = function;
@@ -643,8 +648,9 @@ generate_fragment(struct llvmpipe_screen *screen,
     * Function body
     */
 
-   block = LLVMAppendBasicBlock(function, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(gallivm->context, function, "entry");
+   builder = gallivm->builder;
+   assert(builder);
    LLVMPositionBuilderAtEnd(builder, block);
 
    /*
@@ -653,6 +659,7 @@ generate_fragment(struct llvmpipe_screen *screen,
     * already included in the shader key.
     */
    lp_build_interp_soa_init(&interp, 
+                            gallivm,
                             shader->info.base.num_inputs,
                             inputs,
                             builder, fs_type,
@@ -666,7 +673,7 @@ generate_fragment(struct llvmpipe_screen *screen,
    zs_format_desc = util_format_description(key->zsbuf_format);
 
    for(i = 0; i < num_fs; ++i) {
-      LLVMValueRef depth_offset = LLVMConstInt(LLVMInt32Type(),
+      LLVMValueRef depth_offset = LLVMConstInt(int32_type,
                                                i*fs_type.length*zs_format_desc->block.bits/8,
                                                0);
       LLVMValueRef out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS];
@@ -674,7 +681,8 @@ generate_fragment(struct llvmpipe_screen *screen,
 
       depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &depth_offset, 1, "");
 
-      generate_fs(shader, key,
+      generate_fs(gallivm,
+                  shader, key,
                   builder,
                   fs_type,
                   context_ptr,
@@ -700,7 +708,7 @@ generate_fragment(struct llvmpipe_screen *screen,
     */
    for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
       LLVMValueRef color_ptr;
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), cbuf, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, cbuf);
       LLVMValueRef blend_in_color[NUM_CHANNELS];
       unsigned rt;
 
@@ -715,7 +723,7 @@ generate_fragment(struct llvmpipe_screen *screen,
                LLVMBuildLoad(builder, fs_out_color[cbuf][chan][i], "fs_color_vals");
          }
 
-	 lp_build_conv(builder, fs_type, blend_type,
+	 lp_build_conv(gallivm, fs_type, blend_type,
                        fs_color_vals,
                        num_fs,
 		       &blend_in_color[chan], 1);
@@ -724,11 +732,11 @@ generate_fragment(struct llvmpipe_screen *screen,
       }
 
       if (partial_mask || !variant->opaque) {
-         lp_build_conv_mask(builder, fs_type, blend_type,
+         lp_build_conv_mask(lp->gallivm, fs_type, blend_type,
                             fs_mask, num_fs,
                             &blend_mask, 1);
       } else {
-         blend_mask = lp_build_const_int_vec(blend_type, ~0);
+         blend_mask = lp_build_const_int_vec(lp->gallivm, blend_type, ~0);
       }
 
       color_ptr = LLVMBuildLoad(builder, 
@@ -749,7 +757,8 @@ generate_fragment(struct llvmpipe_screen *screen,
                               !key->alpha.enabled &&
                               !shader->info.base.uses_kill);
 
-         generate_blend(&key->blend,
+         generate_blend(lp->gallivm,
+                        &key->blend,
                         rt,
                         builder,
                         blend_type,
@@ -763,9 +772,6 @@ generate_fragment(struct llvmpipe_screen *screen,
 
    LLVMBuildRetVoid(builder);
 
-   LLVMDisposeBuilder(builder);
-
-
    /* Verify the LLVM IR.  If invalid, dump and abort */
 #ifdef DEBUG
    if(LLVMVerifyFunction(function, LLVMPrintMessageAction)) {
@@ -776,7 +782,7 @@ generate_fragment(struct llvmpipe_screen *screen,
 #endif
 
    /* Apply optimizations to LLVM IR */
-   LLVMRunFunctionPassManager(screen->pass, function);
+   LLVMRunFunctionPassManager(gallivm->passmgr, function);
 
    if ((gallivm_debug & GALLIVM_DEBUG_IR) || (LP_DEBUG & DEBUG_FS)) {
       /* Print the LLVM IR to stderr */
@@ -786,14 +792,14 @@ generate_fragment(struct llvmpipe_screen *screen,
 
    /* Dump byte code to a file */
    if (0) {
-      LLVMWriteBitcodeToFile(lp_build_module, "llvmpipe.bc");
+      LLVMWriteBitcodeToFile(gallivm->module, "llvmpipe.bc");
    }
 
    /*
     * Translate the LLVM IR into machine code.
     */
    {
-      void *f = LLVMGetPointerToGlobal(screen->engine, function);
+      void *f = LLVMGetPointerToGlobal(gallivm->engine, function);
 
       variant->jit_function[partial_mask] = (lp_jit_frag_func)pointer_to_func(f);
 
@@ -897,8 +903,13 @@ lp_debug_fs_variant(const struct lp_fragment_shader_variant *variant)
    debug_printf("\n");
 }
 
+
+/**
+ * Generate a new fragment shader variant from the shader code and
+ * other state indicated by the key.
+ */
 static struct lp_fragment_shader_variant *
-generate_variant(struct llvmpipe_screen *screen,
+generate_variant(struct llvmpipe_context *lp,
                  struct lp_fragment_shader *shader,
                  const struct lp_fragment_shader_variant_key *key)
 {
@@ -944,11 +955,11 @@ generate_variant(struct llvmpipe_screen *screen,
       lp_debug_fs_variant(variant);
    }
 
-   generate_fragment(screen, shader, variant, RAST_EDGE_TEST);
+   generate_fragment(lp, shader, variant, RAST_EDGE_TEST);
 
    if (variant->opaque) {
       /* Specialized shader, which doesn't need to read the color buffer. */
-      generate_fragment(screen, shader, variant, RAST_WHOLE);
+      generate_fragment(lp, shader, variant, RAST_WHOLE);
    } else {
       variant->jit_function[RAST_WHOLE] = variant->jit_function[RAST_EDGE_TEST];
    }
@@ -1033,7 +1044,8 @@ llvmpipe_create_fs_state(struct pipe_context *pipe,
 
    if (LP_DEBUG & DEBUG_TGSI) {
       unsigned attrib;
-      debug_printf("llvmpipe: Create fragment shader #%u %p:\n", shader->no, (void *) shader);
+      debug_printf("llvmpipe: Create fragment shader #%u %p:\n",
+                   shader->no, (void *) shader);
       tgsi_dump(templ->tokens, 0);
       debug_printf("usage masks:\n");
       for (attrib = 0; attrib < shader->info.base.num_inputs; ++attrib) {
@@ -1070,33 +1082,49 @@ llvmpipe_bind_fs_state(struct pipe_context *pipe, void *fs)
    llvmpipe->dirty |= LP_NEW_FS;
 }
 
-static void
-remove_shader_variant(struct llvmpipe_context *lp,
-                      struct lp_fragment_shader_variant *variant)
+
+/**
+ * Remove shader variant from two lists: the shader's variant list
+ * and the context's variant list.
+ */
+void
+llvmpipe_remove_shader_variant(struct llvmpipe_context *lp,
+                               struct lp_fragment_shader_variant *variant)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
    unsigned i;
 
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
-      debug_printf("llvmpipe: del fs #%u var #%u v created #%u v cached #%u v total cached #%u\n",
-                    variant->shader->no, variant->no, variant->shader->variants_created,
-                    variant->shader->variants_cached, lp->nr_fs_variants);
+      debug_printf("llvmpipe: del fs #%u var #%u v created #%u v cached"
+                   " #%u v total cached #%u\n",
+                   variant->shader->no,
+                   variant->no,
+                   variant->shader->variants_created,
+                   variant->shader->variants_cached,
+                   lp->nr_fs_variants);
    }
+
+   /* free all the variant's JIT'd functions */
    for (i = 0; i < Elements(variant->function); i++) {
       if (variant->function[i]) {
          if (variant->jit_function[i])
-            LLVMFreeMachineCodeForFunction(screen->engine,
+            LLVMFreeMachineCodeForFunction(lp->gallivm->engine,
                                            variant->function[i]);
          LLVMDeleteFunction(variant->function[i]);
       }
    }
+
+   /* remove from shader's list */
    remove_from_list(&variant->list_item_local);
    variant->shader->variants_cached--;
+
+   /* remove from context's list */
    remove_from_list(&variant->list_item_global);
    lp->nr_fs_variants--;
+
    FREE(variant);
 }
 
+
 static void
 llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
 {
@@ -1105,23 +1133,23 @@ llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
    struct lp_fs_variant_list_item *li;
 
    assert(fs != llvmpipe->fs);
-   (void) llvmpipe;
 
    /*
     * XXX: we need to flush the context until we have some sort of reference
     * counting in fragment shaders as they may still be binned
     * Flushing alone might not sufficient we need to wait on it too.
     */
-
    llvmpipe_finish(pipe, __FUNCTION__);
 
+   /* Delete all the variants */
    li = first_elem(&shader->variants);
    while(!at_end(&shader->variants, li)) {
       struct lp_fs_variant_list_item *next = next_elem(li);
-      remove_shader_variant(llvmpipe, li->base);
+      llvmpipe_remove_shader_variant(llvmpipe, li->base);
       li = next;
    }
 
+   /* Delete draw module's data */
    draw_delete_fragment_shader(llvmpipe->draw, shader->draw_data);
 
    assert(shader->variants_cached == 0);
@@ -1277,14 +1305,15 @@ make_variant_key(struct llvmpipe_context *lp,
    }
 }
 
+
+
 /**
- * Update fragment state.  This is called just prior to drawing
+ * Update fragment shader state.  This is called just prior to drawing
  * something when some fragment-related state has changed.
  */
 void 
 llvmpipe_update_fs(struct llvmpipe_context *lp)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
    struct lp_fragment_shader *shader = lp->fs;
    struct lp_fragment_shader_variant_key key;
    struct lp_fragment_shader_variant *variant = NULL;
@@ -1292,6 +1321,7 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
 
    make_variant_key(lp, shader, &key);
 
+   /* Search the variants for one which matches the key */
    li = first_elem(&shader->variants);
    while(!at_end(&shader->variants, li)) {
       if(memcmp(&li->base->key, &key, shader->variant_key_size) == 0) {
@@ -1302,36 +1332,49 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
    }
 
    if (variant) {
+      /* Move this variant to the head of the list to implement LRU
+       * deletion of shader's when we have too many.
+       */
       move_to_head(&lp->fs_variants_list, &variant->list_item_global);
    }
    else {
-      int64_t t0, t1;
-      int64_t dt;
+      /* variant not found, create it now */
+      int64_t t0, t1, dt;
       unsigned i;
+
+      /* First, check if we've exceeded the max number of shader variants.
+       * If so, free 25% of them (the least recently used ones).
+       */
       if (lp->nr_fs_variants >= LP_MAX_SHADER_VARIANTS) {
          struct pipe_context *pipe = &lp->pipe;
 
          /*
-          * XXX: we need to flush the context until we have some sort of reference
-          * counting in fragment shaders as they may still be binned
+          * XXX: we need to flush the context until we have some sort of
+          * reference counting in fragment shaders as they may still be binned
           * Flushing alone might not be sufficient we need to wait on it too.
           */
          llvmpipe_finish(pipe, __FUNCTION__);
 
          for (i = 0; i < LP_MAX_SHADER_VARIANTS / 4; i++) {
-            struct lp_fs_variant_list_item *item = last_elem(&lp->fs_variants_list);
-            remove_shader_variant(lp, item->base);
+            struct lp_fs_variant_list_item *item;
+            item = last_elem(&lp->fs_variants_list);
+            llvmpipe_remove_shader_variant(lp, item->base);
          }
       }
-      t0 = os_time_get();
-
-      variant = generate_variant(screen, shader, &key);
 
+      /*
+       * Generate the new variant.
+       */
+      t0 = os_time_get();
+      variant = generate_variant(lp, shader, &key);
       t1 = os_time_get();
       dt = t1 - t0;
       LP_COUNT_ADD(llvm_compile_time, dt);
       LP_COUNT_ADD(nr_llvm_compiles, 2);  /* emit vs. omit in/out test */
 
+      llvmpipe_variant_count++;
+
+      /* Put the new variant into the list */
       if (variant) {
          insert_at_head(&shader->variants, &variant->list_item_local);
          insert_at_head(&lp->fs_variants_list, &variant->list_item_global);
@@ -1340,6 +1383,7 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
       }
    }
 
+   /* Bind this variant */
    lp_setup_set_fs_variant(lp->setup, variant);
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.h b/src/gallium/drivers/llvmpipe/lp_state_fs.h
index 7d58c4936c..98410c6935 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.h
@@ -69,12 +69,15 @@ struct lp_fragment_shader_variant_key
    struct lp_sampler_static_state sampler[PIPE_MAX_SAMPLERS];
 };
 
+
+/** doubly-linked list item */
 struct lp_fs_variant_list_item
 {
    struct lp_fragment_shader_variant *base;
    struct lp_fs_variant_list_item *next, *prev;
 };
 
+
 struct lp_fragment_shader_variant
 {
    struct lp_fragment_shader_variant_key key;
@@ -118,5 +121,9 @@ struct lp_fragment_shader
 void
 lp_debug_fs_variant(const struct lp_fragment_shader_variant *variant);
 
+void
+llvmpipe_remove_shader_variant(struct llvmpipe_context *lp,
+                               struct lp_fragment_shader_variant *variant);
+
 
 #endif /* LP_STATE_FS_H_ */
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.c b/src/gallium/drivers/llvmpipe/lp_state_setup.c
index 47f3d95320..ec3fdcadf4 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.c
@@ -31,6 +31,7 @@
 #include "util/u_simple_list.h"
 #include "os/os_time.h"
 #include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_intr.h"
@@ -83,24 +84,29 @@ struct lp_setup_args
    LLVMValueRef v2a;
 };
 
-static LLVMTypeRef type4f(void)
+
+
+static LLVMTypeRef
+type4f(struct gallivm_state *gallivm)
 {
-   return LLVMVectorType(LLVMFloatType(), 4);
+   return LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);
 }
 
 
 /* Equivalent of _mm_setr_ps(a,b,c,d)
  */
-static LLVMValueRef vec4f(LLVMBuilderRef bld,
-			  LLVMValueRef a, LLVMValueRef b, LLVMValueRef c, LLVMValueRef d,
-			  const char *name)
+static LLVMValueRef
+vec4f(struct gallivm_state *gallivm,
+      LLVMValueRef a, LLVMValueRef b, LLVMValueRef c, LLVMValueRef d,
+      const char *name)
 {
-   LLVMValueRef i0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef i1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef i2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-   LLVMValueRef i3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   LLVMBuilderRef bld = gallivm->builder;
+   LLVMValueRef i0 = lp_build_const_int32(gallivm, 0);
+   LLVMValueRef i1 = lp_build_const_int32(gallivm, 1);
+   LLVMValueRef i2 = lp_build_const_int32(gallivm, 2);
+   LLVMValueRef i3 = lp_build_const_int32(gallivm, 3);
 
-   LLVMValueRef res = LLVMGetUndef(type4f());
+   LLVMValueRef res = LLVMGetUndef(type4f(gallivm));
 
    res = LLVMBuildInsertElement(bld, res, a, i0, "");
    res = LLVMBuildInsertElement(bld, res, b, i1, "");
@@ -112,15 +118,17 @@ static LLVMValueRef vec4f(LLVMBuilderRef bld,
 
 /* Equivalent of _mm_set1_ps(a)
  */
-static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
-				      LLVMValueRef a,
-				      const char *name)
+static LLVMValueRef
+vec4f_from_scalar(struct gallivm_state *gallivm,
+                  LLVMValueRef a,
+                  const char *name)
 {
-   LLVMValueRef res = LLVMGetUndef(type4f());
+   LLVMBuilderRef bld = gallivm->builder;
+   LLVMValueRef res = LLVMGetUndef(type4f(gallivm));
    int i;
 
    for(i = 0; i < 4; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = lp_build_const_int32(gallivm, i);
       res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
    }
 
@@ -128,14 +136,15 @@ static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
 }
 
 static void
-store_coef(LLVMBuilderRef builder,
+store_coef(struct gallivm_state *gallivm,
 	   struct lp_setup_args *args,
 	   unsigned slot,
 	   LLVMValueRef a0,
 	   LLVMValueRef dadx,
 	   LLVMValueRef dady)
 {
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), slot, 0);
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMValueRef idx = lp_build_const_int32(gallivm, slot);
    
    LLVMBuildStore(builder,
 		  a0, 
@@ -153,14 +162,14 @@ store_coef(LLVMBuilderRef builder,
 
 
 static void 
-emit_constant_coef4( LLVMBuilderRef builder,
+emit_constant_coef4(struct gallivm_state *gallivm,
 		     struct lp_setup_args *args,
 		     unsigned slot,
 		     LLVMValueRef vert)
 {
-   LLVMValueRef zero      = LLVMConstReal(LLVMFloatType(), 0.0);
-   LLVMValueRef zerovec   = vec4f_from_scalar(builder, zero, "zero");
-   store_coef(builder, args, slot, vert, zerovec, zerovec);
+   LLVMValueRef zero      = lp_build_const_float(gallivm, 0.0);
+   LLVMValueRef zerovec   = vec4f_from_scalar(gallivm, zero, "zero");
+   store_coef(gallivm, args, slot, vert, zerovec, zerovec);
 }
 
 
@@ -170,30 +179,33 @@ emit_constant_coef4( LLVMBuilderRef builder,
  * \param frontface  is the triangle front facing?
  */
 static void 
-emit_facing_coef( LLVMBuilderRef builder,
+emit_facing_coef(struct gallivm_state *gallivm,
 		  struct lp_setup_args *args,
 		  unsigned slot )
 {
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
    LLVMValueRef a0_0 = args->facing;
-   LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, LLVMFloatType(), "");
-   LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
-   LLVMValueRef a0      = vec4f(builder, a0_0f, zero, zero, zero, "facing");
-   LLVMValueRef zerovec = vec4f_from_scalar(builder, zero, "zero");
+   LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, float_type, "");
+   LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+   LLVMValueRef a0 = vec4f(gallivm, a0_0f, zero, zero, zero, "facing");
+   LLVMValueRef zerovec = vec4f_from_scalar(gallivm, zero, "zero");
 
-   store_coef(builder, args, slot, a0, zerovec, zerovec);
+   store_coef(gallivm, args, slot, a0, zerovec, zerovec);
 }
 
 
 static LLVMValueRef
-vert_attrib(LLVMBuilderRef b,
+vert_attrib(struct gallivm_state *gallivm,
 	    LLVMValueRef vert,
 	    int attr,
 	    int elem,
 	    const char *name)
 {
+   LLVMBuilderRef b = gallivm->builder;
    LLVMValueRef idx[2];
-   idx[0] = LLVMConstInt(LLVMInt32Type(), attr, 0);
-   idx[1] = LLVMConstInt(LLVMInt32Type(), elem, 0);
+   idx[0] = lp_build_const_int32(gallivm, attr);
+   idx[1] = lp_build_const_int32(gallivm, elem);
    return LLVMBuildLoad(b, LLVMBuildGEP(b, vert, idx, 2, ""), name);
 }
 
@@ -214,16 +226,17 @@ vert_clamp(LLVMBuilderRef b,
 }
 
 static void
-lp_twoside(LLVMBuilderRef b, 
+lp_twoside(struct gallivm_state *gallivm,
            struct lp_setup_args *args,
            const struct lp_setup_variant_key *key,
            int bcolor_slot)
 {
+   LLVMBuilderRef b = gallivm->builder;
    LLVMValueRef a0_back, a1_back, a2_back;
-   LLVMValueRef idx2 = LLVMConstInt(LLVMInt32Type(), bcolor_slot, 0);
+   LLVMValueRef idx2 = lp_build_const_int32(gallivm, bcolor_slot);
 
    LLVMValueRef facing = args->facing;
-   LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing, LLVMConstInt(LLVMInt32Type(), 0, 0), ""); /** need i1 for if condition */
+   LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing, lp_build_const_int32(gallivm, 0), ""); /** need i1 for if condition */
    
    a0_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx2, 1, ""), "v0a_back");
    a1_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx2, 1, ""), "v1a_back");
@@ -241,29 +254,31 @@ lp_twoside(LLVMBuilderRef b,
 }
 
 static void
-lp_do_offset_tri(LLVMBuilderRef b, 
+lp_do_offset_tri(struct gallivm_state *gallivm,
                  struct lp_setup_args *args,
                  const struct lp_setup_variant_key *key)
 {
+   LLVMBuilderRef b = gallivm->builder;
    struct lp_build_context bld;
    LLVMValueRef zoffset, mult;
    LLVMValueRef z0_new, z1_new, z2_new;
    LLVMValueRef dzdx0, dzdx, dzdy0, dzdy;
    LLVMValueRef max, max_value;
    
-   LLVMValueRef one  = LLVMConstReal(LLVMFloatType(), 1.0);
-   LLVMValueRef zero  = LLVMConstReal(LLVMFloatType(), 0.0);
+   LLVMValueRef one  = lp_build_const_float(gallivm, 1.0);
+   LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+   LLVMValueRef two  = lp_build_const_int32(gallivm, 2);
 
    /* edge vectors: e = v0 - v2, f = v1 - v2 */
-   LLVMValueRef v0_x = vert_attrib(b, args->v0, 0, 0, "v0_x");
-   LLVMValueRef v1_x = vert_attrib(b, args->v1, 0, 0, "v1_x");
-   LLVMValueRef v2_x = vert_attrib(b, args->v2, 0, 0, "v2_x");
-   LLVMValueRef v0_y = vert_attrib(b, args->v0, 0, 1, "v0_y");
-   LLVMValueRef v1_y = vert_attrib(b, args->v1, 0, 1, "v1_y");
-   LLVMValueRef v2_y = vert_attrib(b, args->v2, 0, 1, "v2_y");
-   LLVMValueRef v0_z = vert_attrib(b, args->v0, 0, 2, "v0_z");
-   LLVMValueRef v1_z = vert_attrib(b, args->v1, 0, 2, "v1_z");
-   LLVMValueRef v2_z = vert_attrib(b, args->v2, 0, 2, "v2_z");
+   LLVMValueRef v0_x = vert_attrib(gallivm, args->v0, 0, 0, "v0_x");
+   LLVMValueRef v1_x = vert_attrib(gallivm, args->v1, 0, 0, "v1_x");
+   LLVMValueRef v2_x = vert_attrib(gallivm, args->v2, 0, 0, "v2_x");
+   LLVMValueRef v0_y = vert_attrib(gallivm, args->v0, 0, 1, "v0_y");
+   LLVMValueRef v1_y = vert_attrib(gallivm, args->v1, 0, 1, "v1_y");
+   LLVMValueRef v2_y = vert_attrib(gallivm, args->v2, 0, 1, "v2_y");
+   LLVMValueRef v0_z = vert_attrib(gallivm, args->v0, 0, 2, "v0_z");
+   LLVMValueRef v1_z = vert_attrib(gallivm, args->v1, 0, 2, "v1_z");
+   LLVMValueRef v2_z = vert_attrib(gallivm, args->v2, 0, 2, "v2_z");
  
    /* edge vectors: e = v0 - v2, f = v1 - v2 */
    LLVMValueRef dx02 = LLVMBuildFSub(b, v0_x, v2_x, "dx02");
@@ -288,7 +303,7 @@ lp_do_offset_tri(LLVMBuilderRef b,
    LLVMValueRef res2  = LLVMBuildFSub(b, dz02_dx12, dx02_dz12, "res2");
 
    /* dzdx = fabsf(res1 * inv_det), dydx = fabsf(res2 * inv_det)*/
-   lp_build_context_init(&bld, b, lp_type_float(32));
+   lp_build_context_init(&bld, gallivm, lp_type_float(32));
    dzdx0 = LLVMBuildFMul(b, res1, inv_det, "dzdx");
    dzdx  = lp_build_abs(&bld, dzdx0);
    dzdy0 = LLVMBuildFMul(b, res2, inv_det, "dzdy");
@@ -298,8 +313,8 @@ lp_do_offset_tri(LLVMBuilderRef b,
    max = LLVMBuildFCmp(b, LLVMRealUGT, dzdx, dzdy, "");
    max_value = LLVMBuildSelect(b, max, dzdx, dzdy, "max"); 
 
-   mult = LLVMBuildFMul(b, max_value, LLVMConstReal(LLVMFloatType(), key->scale), "");
-   zoffset = LLVMBuildFAdd(b, LLVMConstReal(LLVMFloatType(), key->units), mult, "zoffset");
+   mult = LLVMBuildFMul(b, max_value, lp_build_const_float(gallivm, key->scale), "");
+   zoffset = LLVMBuildFAdd(b, lp_build_const_float(gallivm, key->units), mult, "zoffset");
 
    /* clamp and do offset */
    z0_new = vert_clamp(b, LLVMBuildFAdd(b, v0_z, zoffset, ""), zero, one);
@@ -308,18 +323,19 @@ lp_do_offset_tri(LLVMBuilderRef b,
 
    /* insert into args->a0.z, a1.z, a2.z:
     */   
-   args->v0a = LLVMBuildInsertElement(b, args->v0a, z0_new, LLVMConstInt(LLVMInt32Type(), 2, 0), "");
-   args->v1a = LLVMBuildInsertElement(b, args->v1a, z1_new, LLVMConstInt(LLVMInt32Type(), 2, 0), "");
-   args->v2a = LLVMBuildInsertElement(b, args->v2a, z2_new, LLVMConstInt(LLVMInt32Type(), 2, 0), "");
+   args->v0a = LLVMBuildInsertElement(b, args->v0a, z0_new, two, "");
+   args->v1a = LLVMBuildInsertElement(b, args->v1a, z1_new, two, "");
+   args->v2a = LLVMBuildInsertElement(b, args->v2a, z2_new, two, "");
 }
 
 static void
-load_attribute(LLVMBuilderRef b, 
+load_attribute(struct gallivm_state *gallivm,
                struct lp_setup_args *args,
                const struct lp_setup_variant_key *key,
                unsigned vert_attr)
 {
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
+   LLVMBuilderRef b = gallivm->builder;
+   LLVMValueRef idx = lp_build_const_int32(gallivm, vert_attr);
 
    /* Load the vertex data
     */
@@ -331,25 +347,26 @@ load_attribute(LLVMBuilderRef b,
    /* Potentially modify it according to twoside, offset, etc:
     */
    if (vert_attr == 0 && (key->scale != 0.0f || key->units != 0.0f)) {
-      lp_do_offset_tri(b, args, key);
+      lp_do_offset_tri(gallivm, args, key);
    }
 
    if (key->twoside) {
       if (vert_attr == key->color_slot && key->bcolor_slot != ~0)
-         lp_twoside(b, args, key, key->bcolor_slot);
+         lp_twoside(gallivm, args, key, key->bcolor_slot);
       else if (vert_attr == key->spec_slot && key->bspec_slot != ~0)
-         lp_twoside(b, args, key, key->bspec_slot);
+         lp_twoside(gallivm, args, key, key->bspec_slot);
    }
 }
 
 static void 
-emit_coef4( LLVMBuilderRef b,
+emit_coef4( struct gallivm_state *gallivm,
 	    struct lp_setup_args *args,
 	    unsigned slot,
 	    LLVMValueRef a0,
 	    LLVMValueRef a1,
 	    LLVMValueRef a2)
 {
+   LLVMBuilderRef b = gallivm->builder;
    LLVMValueRef dy20_ooa = args->dy20_ooa;
    LLVMValueRef dy01_ooa = args->dy01_ooa;
    LLVMValueRef dx20_ooa = args->dx20_ooa;
@@ -381,17 +398,17 @@ emit_coef4( LLVMBuilderRef b,
    LLVMValueRef attr_v0       = LLVMBuildFAdd(b, dadx_x0, dady_y0, "attr_v0"); 
    LLVMValueRef attr_0        = LLVMBuildFSub(b, a0, attr_v0, "attr_0"); 
 
-   store_coef(b, args, slot, attr_0, dadx, dady);
+   store_coef(gallivm, args, slot, attr_0, dadx, dady);
 }
 
 
 static void 
-emit_linear_coef( LLVMBuilderRef b,
+emit_linear_coef( struct gallivm_state *gallivm,
 		  struct lp_setup_args *args,
 		  unsigned slot)
 {
    /* nothing to do anymore */
-   emit_coef4(b,
+   emit_coef4(gallivm,
               args, slot, 
               args->v0a,
               args->v1a,
@@ -408,30 +425,32 @@ emit_linear_coef( LLVMBuilderRef b,
  * divide the interpolated value by the interpolated W at that fragment.
  */
 static void 
-emit_perspective_coef( LLVMBuilderRef b,
+emit_perspective_coef( struct gallivm_state *gallivm,
 		       struct lp_setup_args *args,
 		       unsigned slot)
 {
+   LLVMBuilderRef b = gallivm->builder;
+
    /* premultiply by 1/w  (v[0][3] is always 1/w):
     */
-   LLVMValueRef v0_oow = vec4f_from_scalar(b, vert_attrib(b, args->v0, 0, 3, ""), "v0_oow");
-   LLVMValueRef v1_oow = vec4f_from_scalar(b, vert_attrib(b, args->v1, 0, 3, ""), "v1_oow");
-   LLVMValueRef v2_oow = vec4f_from_scalar(b, vert_attrib(b, args->v2, 0, 3, ""), "v2_oow");
+   LLVMValueRef v0_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v0, 0, 3, ""), "v0_oow");
+   LLVMValueRef v1_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v1, 0, 3, ""), "v1_oow");
+   LLVMValueRef v2_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v2, 0, 3, ""), "v2_oow");
 
    LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, args->v0a, v0_oow, "v0_oow_v0a");
    LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, args->v1a, v1_oow, "v1_oow_v1a");
    LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, args->v2a, v2_oow, "v2_oow_v2a");
 
-   emit_coef4(b, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
+   emit_coef4(gallivm, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
 }
 
 
 static void
-emit_position_coef( LLVMBuilderRef builder,
+emit_position_coef( struct gallivm_state *gallivm,
 		    struct lp_setup_args *args,
 		    int slot )
 {
-   emit_linear_coef(builder, args, slot);
+   emit_linear_coef(gallivm, args, slot);
 }
 
 
@@ -441,7 +460,7 @@ emit_position_coef( LLVMBuilderRef builder,
  * Compute the inputs-> dadx, dady, a0 values.
  */
 static void 
-emit_tri_coef( LLVMBuilderRef builder,
+emit_tri_coef( struct gallivm_state *gallivm,
 	       const struct lp_setup_variant_key *key,
 	       struct lp_setup_args *args )
 {
@@ -449,8 +468,8 @@ emit_tri_coef( LLVMBuilderRef builder,
 
    /* The internal position input is in slot zero:
     */
-   load_attribute(builder, args, key, 0);
-   emit_position_coef(builder, args, 0);
+   load_attribute(gallivm, args, key, 0);
+   emit_position_coef(gallivm, args, 0);
 
    /* setup interpolation for all the remaining attributes:
     */
@@ -459,24 +478,24 @@ emit_tri_coef( LLVMBuilderRef builder,
       if (key->inputs[slot].interp == LP_INTERP_CONSTANT ||
           key->inputs[slot].interp == LP_INTERP_LINEAR ||
           key->inputs[slot].interp == LP_INTERP_PERSPECTIVE)
-         load_attribute(builder, args, key, key->inputs[slot].src_index);
+         load_attribute(gallivm, args, key, key->inputs[slot].src_index);
 
       switch (key->inputs[slot].interp) {
       case LP_INTERP_CONSTANT:
 	 if (key->flatshade_first) {
-	    emit_constant_coef4(builder, args, slot+1, args->v0a);
+	    emit_constant_coef4(gallivm, args, slot+1, args->v0a);
 	 }
 	 else {
-	    emit_constant_coef4(builder, args, slot+1, args->v2a);
+	    emit_constant_coef4(gallivm, args, slot+1, args->v2a);
 	 }
 	 break;
 
       case LP_INTERP_LINEAR:
-	 emit_linear_coef(builder, args, slot+1);
+	 emit_linear_coef(gallivm, args, slot+1);
          break;
 
       case LP_INTERP_PERSPECTIVE:
-	 emit_perspective_coef(builder, args, slot+1);
+	 emit_perspective_coef(gallivm, args, slot+1);
          break;
 
       case LP_INTERP_POSITION:
@@ -487,7 +506,7 @@ emit_tri_coef( LLVMBuilderRef builder,
          break;
 
       case LP_INTERP_FACING:
-         emit_facing_coef(builder, args, slot+1);
+         emit_facing_coef(gallivm, args, slot+1);
          break;
 
       default:
@@ -500,7 +519,7 @@ emit_tri_coef( LLVMBuilderRef builder,
 /* XXX: This is generic code, share with fs/vs codegen:
  */
 static lp_jit_setup_triangle
-finalize_function(struct llvmpipe_screen *screen,
+finalize_function(struct gallivm_state *gallivm,
 		  LLVMBuilderRef builder,
 		  LLVMValueRef function)
 {
@@ -516,7 +535,7 @@ finalize_function(struct llvmpipe_screen *screen,
 #endif
 
    /* Apply optimizations to LLVM IR */
-   LLVMRunFunctionPassManager(screen->pass, function);
+   LLVMRunFunctionPassManager(gallivm->passmgr, function);
 
    if (gallivm_debug & GALLIVM_DEBUG_IR)
    {
@@ -528,7 +547,7 @@ finalize_function(struct llvmpipe_screen *screen,
    /*
     * Translate the LLVM IR into machine code.
     */
-   f = LLVMGetPointerToGlobal(screen->engine, function);
+   f = LLVMGetPointerToGlobal(gallivm->engine, function);
 
    if (gallivm_debug & GALLIVM_DEBUG_ASM)
    {
@@ -568,21 +587,23 @@ set_noalias(LLVMBuilderRef builder,
 }
 
 static void
-init_args(LLVMBuilderRef b, 
+init_args(struct gallivm_state *gallivm,
 	  struct lp_setup_args *args,
 	  const struct lp_setup_variant *variant)
 {
-   LLVMValueRef v0_x = vert_attrib(b, args->v0, 0, 0, "v0_x");
-   LLVMValueRef v0_y = vert_attrib(b, args->v0, 0, 1, "v0_y");
+   LLVMBuilderRef b = gallivm->builder;
 
-   LLVMValueRef v1_x = vert_attrib(b, args->v1, 0, 0, "v1_x");
-   LLVMValueRef v1_y = vert_attrib(b, args->v1, 0, 1, "v1_y");
+   LLVMValueRef v0_x = vert_attrib(gallivm, args->v0, 0, 0, "v0_x");
+   LLVMValueRef v0_y = vert_attrib(gallivm, args->v0, 0, 1, "v0_y");
 
-   LLVMValueRef v2_x = vert_attrib(b, args->v2, 0, 0, "v2_x");
-   LLVMValueRef v2_y = vert_attrib(b, args->v2, 0, 1, "v2_y");
+   LLVMValueRef v1_x = vert_attrib(gallivm, args->v1, 0, 0, "v1_x");
+   LLVMValueRef v1_y = vert_attrib(gallivm, args->v1, 0, 1, "v1_y");
 
-   LLVMValueRef pixel_center = LLVMConstReal(LLVMFloatType(),
-					     variant->key.pixel_center_half ? 0.5 : 0);
+   LLVMValueRef v2_x = vert_attrib(gallivm, args->v2, 0, 0, "v2_x");
+   LLVMValueRef v2_y = vert_attrib(gallivm, args->v2, 0, 1, "v2_y");
+
+   LLVMValueRef pixel_center = lp_build_const_float(gallivm,
+                                   variant->key.pixel_center_half ? 0.5 : 0);
 
    LLVMValueRef x0_center = LLVMBuildFSub(b, v0_x, pixel_center, "x0_center" );
    LLVMValueRef y0_center = LLVMBuildFSub(b, v0_y, pixel_center, "y0_center" );
@@ -592,7 +613,7 @@ init_args(LLVMBuilderRef b,
    LLVMValueRef dx20 = LLVMBuildFSub(b, v2_x, v0_x, "dx20");
    LLVMValueRef dy20 = LLVMBuildFSub(b, v2_y, v0_y, "dy20");
 
-   LLVMValueRef one  = LLVMConstReal(LLVMFloatType(), 1.0);
+   LLVMValueRef one  = lp_build_const_float(gallivm, 1.0);
    LLVMValueRef e    = LLVMBuildFMul(b, dx01, dy20, "e");
    LLVMValueRef f    = LLVMBuildFMul(b, dx20, dy01, "f");
    LLVMValueRef ooa  = LLVMBuildFDiv(b, one, LLVMBuildFSub(b, e, f, ""), "ooa");
@@ -602,14 +623,14 @@ init_args(LLVMBuilderRef b,
    LLVMValueRef dx20_ooa = LLVMBuildFMul(b, dx20, ooa, "dx20_ooa");
    LLVMValueRef dx01_ooa = LLVMBuildFMul(b, dx01, ooa, "dx01_ooa");
 
-   args->dy20_ooa  = vec4f_from_scalar(b, dy20_ooa, "dy20_ooa_4f");
-   args->dy01_ooa  = vec4f_from_scalar(b, dy01_ooa, "dy01_ooa_4f");
+   args->dy20_ooa  = vec4f_from_scalar(gallivm, dy20_ooa, "dy20_ooa_4f");
+   args->dy01_ooa  = vec4f_from_scalar(gallivm, dy01_ooa, "dy01_ooa_4f");
 
-   args->dx20_ooa  = vec4f_from_scalar(b, dx20_ooa, "dx20_ooa_4f");
-   args->dx01_ooa  = vec4f_from_scalar(b, dx01_ooa, "dx01_ooa_4f");
+   args->dx20_ooa  = vec4f_from_scalar(gallivm, dx20_ooa, "dx20_ooa_4f");
+   args->dx01_ooa  = vec4f_from_scalar(gallivm, dx01_ooa, "dx01_ooa_4f");
 
-   args->x0_center = vec4f_from_scalar(b, x0_center, "x0_center_4f");
-   args->y0_center = vec4f_from_scalar(b, y0_center, "y0_center_4f");
+   args->x0_center = vec4f_from_scalar(gallivm, x0_center, "x0_center_4f");
+   args->y0_center = vec4f_from_scalar(gallivm, y0_center, "y0_center_4f");
 }
 
 /**
@@ -617,7 +638,7 @@ init_args(LLVMBuilderRef b,
  *
  */
 static struct lp_setup_variant *
-generate_setup_variant(struct llvmpipe_screen *screen,
+generate_setup_variant(struct gallivm_state *gallivm,
 		       struct lp_setup_variant_key *key,
                        struct llvmpipe_context *lp)
 {
@@ -628,7 +649,7 @@ generate_setup_variant(struct llvmpipe_screen *screen,
    LLVMTypeRef func_type;
    LLVMTypeRef arg_types[7];
    LLVMBasicBlockRef block;
-   LLVMBuilderRef builder;
+   LLVMBuilderRef builder = gallivm->builder;
    int64_t t0, t1;
 
    if (0)
@@ -653,19 +674,20 @@ generate_setup_variant(struct llvmpipe_screen *screen,
     * the vertices.
     */
 
-   vec4f_type = LLVMVectorType(LLVMFloatType(), 4);
+   vec4f_type = LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);
 
    arg_types[0] = LLVMPointerType(vec4f_type, 0);        /* v0 */
    arg_types[1] = LLVMPointerType(vec4f_type, 0);        /* v1 */
    arg_types[2] = LLVMPointerType(vec4f_type, 0);        /* v2 */
-   arg_types[3] = LLVMInt32Type();			/* facing */
+   arg_types[3] = LLVMInt32TypeInContext(gallivm->context); /* facing */
    arg_types[4] = LLVMPointerType(vec4f_type, 0);	/* a0, aligned */
    arg_types[5] = LLVMPointerType(vec4f_type, 0);	/* dadx, aligned */
    arg_types[6] = LLVMPointerType(vec4f_type, 0);	/* dady, aligned */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
+                                arg_types, Elements(arg_types), 0);
 
-   variant->function = LLVMAddFunction(screen->module, func_name, func_type);
+   variant->function = LLVMAddFunction(gallivm->module, func_name, func_type);
    if (!variant->function)
       goto fail;
 
@@ -690,19 +712,18 @@ generate_setup_variant(struct llvmpipe_screen *screen,
    /*
     * Function body
     */
-   block = LLVMAppendBasicBlock(variant->function, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(gallivm->context,
+                                         variant->function, "entry");
    LLVMPositionBuilderAtEnd(builder, block);
 
    set_noalias(builder, variant->function, arg_types, Elements(arg_types));
-   init_args(builder, &args, variant);
-   emit_tri_coef(builder, &variant->key, &args);
+   init_args(gallivm, &args, variant);
+   emit_tri_coef(gallivm, &variant->key, &args);
 
    lp_emit_emms(builder);
    LLVMBuildRetVoid(builder);
-   LLVMDisposeBuilder(builder);
 
-   variant->jit_function = finalize_function(screen, builder,
+   variant->jit_function = finalize_function(gallivm, builder,
 					     variant->function);
    if (!variant->jit_function)
       goto fail;
@@ -722,7 +743,7 @@ fail:
    if (variant) {
       if (variant->function) {
 	 if (variant->jit_function)
-	    LLVMFreeMachineCodeForFunction(screen->engine,
+	    LLVMFreeMachineCodeForFunction(gallivm->engine,
 					   variant->function);
 	 LLVMDeleteFunction(variant->function);
       }
@@ -773,8 +794,6 @@ static void
 remove_setup_variant(struct llvmpipe_context *lp,
 		     struct lp_setup_variant *variant)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
-
    if (gallivm_debug & GALLIVM_DEBUG_IR) {
       debug_printf("llvmpipe: del setup_variant #%u total %u\n",
 		   variant->no, lp->nr_setup_variants);
@@ -782,7 +801,7 @@ remove_setup_variant(struct llvmpipe_context *lp,
 
    if (variant->function) {
       if (variant->jit_function)
-	 LLVMFreeMachineCodeForFunction(screen->engine,
+	 LLVMFreeMachineCodeForFunction(lp->gallivm->engine,
 					variant->function);
       LLVMDeleteFunction(variant->function);
    }
@@ -825,8 +844,6 @@ cull_setup_variants(struct llvmpipe_context *lp)
 void 
 llvmpipe_update_setup(struct llvmpipe_context *lp)
 {
-   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
-
    struct lp_setup_variant_key *key = &lp->setup_variant.key;
    struct lp_setup_variant *variant = NULL;
    struct lp_setup_variant_list_item *li;
@@ -849,9 +866,11 @@ llvmpipe_update_setup(struct llvmpipe_context *lp)
 	 cull_setup_variants(lp);
       }
 
-      variant = generate_setup_variant(screen, key, lp);
+      variant = generate_setup_variant(lp->gallivm, key, lp);
       insert_at_head(&lp->setup_variants_list, &variant->list_item_global);
       lp->nr_setup_variants++;
+
+      llvmpipe_variant_count++;
    }
 
    lp_setup_set_setup_variant(lp->setup,
diff --git a/src/gallium/drivers/llvmpipe/lp_test.h b/src/gallium/drivers/llvmpipe/lp_test.h
index 90422e4258..c64f3e149f 100644
--- a/src/gallium/drivers/llvmpipe/lp_test.h
+++ b/src/gallium/drivers/llvmpipe/lp_test.h
@@ -64,13 +64,14 @@ write_tsv_header(FILE *fp);
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n);
+test_some(struct gallivm_state *gallivm,unsigned verbose, FILE *fp,
+          unsigned long n);
 
 boolean
-test_single(unsigned verbose, FILE *fp);
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp);
 
 boolean
-test_all(unsigned verbose, FILE *fp);
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp);
 
 
 #if defined(PIPE_CC_MSVC)
diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c
index 8b6b5e1298..b3ca134131 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c
@@ -163,11 +163,13 @@ dump_blend_type(FILE *fp,
 
 
 static LLVMValueRef
-add_blend_test(LLVMModuleRef module,
+add_blend_test(struct gallivm_state *gallivm,
                const struct pipe_blend_state *blend,
                enum vector_mode mode,
                struct lp_type type)
 {
+   LLVMModuleRef module = gallivm->module;
+   LLVMContextRef context = gallivm->context;
    LLVMTypeRef vec_type;
    LLVMTypeRef args[4];
    LLVMValueRef func;
@@ -179,18 +181,18 @@ add_blend_test(LLVMModuleRef module,
    LLVMBuilderRef builder;
    const unsigned rt = 0;
 
-   vec_type = lp_build_vec_type(type);
+   vec_type = lp_build_vec_type(gallivm, type);
 
    args[3] = args[2] = args[1] = args[0] = LLVMPointerType(vec_type, 0);
-   func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 4, 0));
+   func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidTypeInContext(context), args, 4, 0));
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
    src_ptr = LLVMGetParam(func, 0);
    dst_ptr = LLVMGetParam(func, 1);
    const_ptr = LLVMGetParam(func, 2);
    res_ptr = LLVMGetParam(func, 3);
 
-   block = LLVMAppendBasicBlock(func, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(context, func, "entry");
+   builder = gallivm->builder;
    LLVMPositionBuilderAtEnd(builder, block);
 
    if (mode == AoS) {
@@ -203,7 +205,7 @@ add_blend_test(LLVMModuleRef module,
       dst = LLVMBuildLoad(builder, dst_ptr, "dst");
       con = LLVMBuildLoad(builder, const_ptr, "const");
 
-      res = lp_build_blend_aos(builder, blend, type, rt, src, dst, con, 3);
+      res = lp_build_blend_aos(gallivm, blend, type, rt, src, dst, con, 3);
 
       lp_build_name(res, "res");
 
@@ -218,7 +220,7 @@ add_blend_test(LLVMModuleRef module,
       unsigned i;
 
       for(i = 0; i < 4; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+         LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
          src[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, src_ptr, &index, 1, ""), "");
          dst[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), "");
          con[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), "");
@@ -227,10 +229,10 @@ add_blend_test(LLVMModuleRef module,
          lp_build_name(dst[i], "dst.%c", "rgba"[i]);
       }
 
-      lp_build_blend_soa(builder, blend, type, rt, src, dst, con, res);
+      lp_build_blend_soa(gallivm, blend, type, rt, src, dst, con, res);
 
       for(i = 0; i < 4; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+         LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
          lp_build_name(res[i], "res.%c", "rgba"[i]);
          LLVMBuildStore(builder, res[i], LLVMBuildGEP(builder, res_ptr, &index, 1, ""));
       }
@@ -238,7 +240,6 @@ add_blend_test(LLVMModuleRef module,
 
    LLVMBuildRetVoid(builder);;
 
-   LLVMDisposeBuilder(builder);
    return func;
 }
 
@@ -465,16 +466,16 @@ compute_blend_ref(const struct pipe_blend_state *blend,
 
 PIPE_ALIGN_STACK
 static boolean
-test_one(unsigned verbose,
+test_one(struct gallivm_state *gallivm,
+         unsigned verbose,
          FILE *fp,
          const struct pipe_blend_state *blend,
          enum vector_mode mode,
          struct lp_type type)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef func = NULL;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    char *error = NULL;
    blend_test_ptr_t blend_test_ptr;
    boolean success;
@@ -487,9 +488,7 @@ test_one(unsigned verbose,
    if(verbose >= 1)
       dump_blend_type(stdout, blend, mode, type);
 
-   module = LLVMModuleCreateWithName("test");
-
-   func = add_blend_test(module, blend, mode, type);
+   func = add_blend_test(gallivm, blend, mode, type);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       LLVMDumpModule(module);
@@ -497,24 +496,6 @@ test_one(unsigned verbose,
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
-   if(verbose >= 2)
-      LLVMDumpModule(module);
-
    code = LLVMGetPointerToGlobal(engine, func);
    blend_test_ptr = voidptr_to_blend_test_ptr_t(code);
 
@@ -715,9 +696,6 @@ test_one(unsigned verbose,
 
    LLVMFreeMachineCodeForFunction(engine, func);
 
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
@@ -773,7 +751,7 @@ const unsigned num_types = sizeof(blend_types)/sizeof(blend_types[0]);
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    const unsigned *rgb_func;
    const unsigned *rgb_src_factor;
@@ -809,7 +787,7 @@ test_all(unsigned verbose, FILE *fp)
                            blend.rt[0].alpha_dst_factor  = *alpha_dst_factor;
                            blend.rt[0].colormask         = PIPE_MASK_RGBA;
 
-                           if(!test_one(verbose, fp, &blend, mode, *type))
+                           if(!test_one(gallivm, verbose, fp, &blend, mode, *type))
                              success = FALSE;
 
                         }
@@ -826,7 +804,8 @@ test_all(unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
    const unsigned *rgb_func;
    const unsigned *rgb_src_factor;
@@ -868,7 +847,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
       blend.rt[0].alpha_dst_factor  = *alpha_dst_factor;
       blend.rt[0].colormask         = PIPE_MASK_RGBA;
 
-      if(!test_one(verbose, fp, &blend, mode, *type))
+      if(!test_one(gallivm, verbose, fp, &blend, mode, *type))
         success = FALSE;
    }
 
@@ -877,7 +856,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c
index 3ba42bf11a..f4a2f360c7 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c
@@ -97,64 +97,65 @@ dump_conv_types(FILE *fp,
 
 
 static LLVMValueRef
-add_conv_test(LLVMModuleRef module,
+add_conv_test(struct gallivm_state *gallivm,
               struct lp_type src_type, unsigned num_srcs,
               struct lp_type dst_type, unsigned num_dsts)
 {
+   LLVMModuleRef module = gallivm->module;
+   LLVMContextRef context = gallivm->context;
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMTypeRef args[2];
    LLVMValueRef func;
    LLVMValueRef src_ptr;
    LLVMValueRef dst_ptr;
    LLVMBasicBlockRef block;
-   LLVMBuilderRef builder;
    LLVMValueRef src[LP_MAX_VECTOR_LENGTH];
    LLVMValueRef dst[LP_MAX_VECTOR_LENGTH];
    unsigned i;
 
-   args[0] = LLVMPointerType(lp_build_vec_type(src_type), 0);
-   args[1] = LLVMPointerType(lp_build_vec_type(dst_type), 0);
+   args[0] = LLVMPointerType(lp_build_vec_type(gallivm, src_type), 0);
+   args[1] = LLVMPointerType(lp_build_vec_type(gallivm, dst_type), 0);
 
-   func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 2, 0));
+   func = LLVMAddFunction(module, "test",
+                          LLVMFunctionType(LLVMVoidTypeInContext(context),
+                                           args, 2, 0));
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
    src_ptr = LLVMGetParam(func, 0);
    dst_ptr = LLVMGetParam(func, 1);
 
-   block = LLVMAppendBasicBlock(func, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMPositionBuilderAtEnd(builder, block);
 
    for(i = 0; i < num_srcs; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
       LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, "");
       src[i] = LLVMBuildLoad(builder, ptr, "");
    }
 
-   lp_build_conv(builder, src_type, dst_type, src, num_srcs, dst, num_dsts);
+   lp_build_conv(gallivm, src_type, dst_type, src, num_srcs, dst, num_dsts);
 
    for(i = 0; i < num_dsts; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
       LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, "");
       LLVMBuildStore(builder, dst[i], ptr);
    }
 
    LLVMBuildRetVoid(builder);;
 
-   LLVMDisposeBuilder(builder);
    return func;
 }
 
 
 PIPE_ALIGN_STACK
 static boolean
-test_one(unsigned verbose,
+test_one(struct gallivm_state *gallivm, unsigned verbose,
          FILE *fp,
          struct lp_type src_type,
          struct lp_type dst_type)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    LLVMValueRef func = NULL;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
    char *error = NULL;
    conv_test_ptr_t conv_test_ptr;
    boolean success;
@@ -193,9 +194,7 @@ test_one(unsigned verbose,
 
    eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type));
 
-   module = LLVMModuleCreateWithName("test");
-
-   func = add_conv_test(module, src_type, num_srcs, dst_type, num_dsts);
+   func = add_conv_test(gallivm, src_type, num_srcs, dst_type, num_dsts);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       LLVMDumpModule(module);
@@ -203,21 +202,6 @@ test_one(unsigned verbose,
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    if(verbose >= 2)
       LLVMDumpModule(module);
 
@@ -342,9 +326,6 @@ test_one(unsigned verbose,
 
    LLVMFreeMachineCodeForFunction(engine, func);
 
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
@@ -390,7 +371,7 @@ const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]);
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    const struct lp_type *src_type;
    const struct lp_type *dst_type;
@@ -405,7 +386,7 @@ test_all(unsigned verbose, FILE *fp)
          if(src_type->norm != dst_type->norm)
             continue;
 
-         if(!test_one(verbose, fp, *src_type, *dst_type))
+         if(!test_one(gallivm, verbose, fp, *src_type, *dst_type))
            success = FALSE;
 
       }
@@ -416,7 +397,8 @@ test_all(unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
    const struct lp_type *src_type;
    const struct lp_type *dst_type;
@@ -430,7 +412,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
          dst_type = &conv_types[rand() % num_types];
       } while (src_type == dst_type || src_type->norm != dst_type->norm);
 
-      if(!test_one(verbose, fp, *src_type, *dst_type))
+      if(!test_one(gallivm, verbose, fp, *src_type, *dst_type))
         success = FALSE;
    }
 
@@ -439,7 +421,7 @@ test_some(unsigned verbose, FILE *fp, unsigned long n)
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    /*    float, fixed,  sign,  norm, width, len */
    struct lp_type f32x4_type =
@@ -449,7 +431,7 @@ test_single(unsigned verbose, FILE *fp)
 
    boolean success;
 
-   success = test_one(verbose, fp, f32x4_type, ub8x4_type);
+   success = test_one(gallivm, verbose, fp, f32x4_type, ub8x4_type);
 
    return success;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 2855d7cea4..4152ca6cf6 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -30,13 +30,6 @@
 #include <stdio.h>
 #include <float.h>
 
-#include "gallivm/lp_bld.h"
-#include "gallivm/lp_bld_debug.h"
-#include "gallivm/lp_bld_init.h"
-#include <llvm-c/Analysis.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "util/u_memory.h"
 #include "util/u_pointer.h"
 #include "util/u_string.h"
@@ -44,7 +37,11 @@
 #include "util/u_format_tests.h"
 #include "util/u_format_s3tc.h"
 
+#include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_format.h"
+#include "gallivm/lp_bld_init.h"
+
 #include "lp_test.h"
 
 
@@ -78,56 +75,57 @@ typedef void
 
 
 static LLVMValueRef
-add_fetch_rgba_test(unsigned verbose,
+add_fetch_rgba_test(struct gallivm_state *gallivm, unsigned verbose,
                     const struct util_format_description *desc,
                     struct lp_type type)
 {
    char name[256];
+   LLVMContextRef context = gallivm->context;
+   LLVMModuleRef module = gallivm->module;
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMPassManagerRef passmgr = gallivm->passmgr;
    LLVMTypeRef args[4];
    LLVMValueRef func;
    LLVMValueRef packed_ptr;
-   LLVMValueRef offset = LLVMConstNull(LLVMInt32Type());
+   LLVMValueRef offset = LLVMConstNull(LLVMInt32TypeInContext(context));
    LLVMValueRef rgba_ptr;
    LLVMValueRef i;
    LLVMValueRef j;
    LLVMBasicBlockRef block;
-   LLVMBuilderRef builder;
    LLVMValueRef rgba;
 
    util_snprintf(name, sizeof name, "fetch_%s_%s", desc->short_name,
                  type.floating ? "float" : "unorm8");
 
-   args[0] = LLVMPointerType(lp_build_vec_type(type), 0);
-   args[1] = LLVMPointerType(LLVMInt8Type(), 0);
-   args[3] = args[2] = LLVMInt32Type();
+   args[0] = LLVMPointerType(lp_build_vec_type(gallivm, type), 0);
+   args[1] = LLVMPointerType(LLVMInt8TypeInContext(context), 0);
+   args[3] = args[2] = LLVMInt32TypeInContext(context);
 
-   func = LLVMAddFunction(lp_build_module, name,
-                          LLVMFunctionType(LLVMVoidType(), args, Elements(args), 0));
+   func = LLVMAddFunction(module, name,
+                          LLVMFunctionType(LLVMVoidTypeInContext(context),
+                                           args, Elements(args), 0));
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
    rgba_ptr = LLVMGetParam(func, 0);
    packed_ptr = LLVMGetParam(func, 1);
    i = LLVMGetParam(func, 2);
    j = LLVMGetParam(func, 3);
 
-   block = LLVMAppendBasicBlock(func, "entry");
-   builder = LLVMCreateBuilder();
+   block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMPositionBuilderAtEnd(builder, block);
 
-   rgba = lp_build_fetch_rgba_aos(builder, desc, type,
+   rgba = lp_build_fetch_rgba_aos(gallivm, desc, type,
                                   packed_ptr, offset, i, j);
 
    LLVMBuildStore(builder, rgba, rgba_ptr);
 
    LLVMBuildRetVoid(builder);
 
-   LLVMDisposeBuilder(builder);
-
    if (LLVMVerifyFunction(func, LLVMPrintMessageAction)) {
       LLVMDumpValue(func);
       abort();
    }
 
-   LLVMRunFunctionPassManager(lp_build_pass, func);
+   LLVMRunFunctionPassManager(passmgr, func);
 
    if (verbose >= 1) {
       LLVMDumpValue(func);
@@ -139,10 +137,11 @@ add_fetch_rgba_test(unsigned verbose,
 
 PIPE_ALIGN_STACK
 static boolean
-test_format_float(unsigned verbose, FILE *fp,
+test_format_float(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
                   const struct util_format_description *desc)
 {
    LLVMValueRef fetch = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    fetch_ptr_t fetch_ptr;
    PIPE_ALIGN_VAR(16) float unpacked[4];
    boolean first = TRUE;
@@ -150,9 +149,9 @@ test_format_float(unsigned verbose, FILE *fp,
    unsigned i, j, k, l;
    void *f;
 
-   fetch = add_fetch_rgba_test(verbose, desc, lp_float32_vec4_type());
+   fetch = add_fetch_rgba_test(gallivm, verbose, desc, lp_float32_vec4_type());
 
-   f = LLVMGetPointerToGlobal(lp_build_engine, fetch);
+   f = LLVMGetPointerToGlobal(engine, fetch);
    fetch_ptr = (fetch_ptr_t) pointer_to_func(f);
 
    if (verbose >= 2) {
@@ -208,7 +207,7 @@ test_format_float(unsigned verbose, FILE *fp,
       }
    }
 
-   LLVMFreeMachineCodeForFunction(lp_build_engine, fetch);
+   LLVMFreeMachineCodeForFunction(engine, fetch);
    LLVMDeleteFunction(fetch);
 
    if(fp)
@@ -220,7 +219,8 @@ test_format_float(unsigned verbose, FILE *fp,
 
 PIPE_ALIGN_STACK
 static boolean
-test_format_unorm8(unsigned verbose, FILE *fp,
+test_format_unorm8(struct gallivm_state *gallivm,
+                   unsigned verbose, FILE *fp,
                    const struct util_format_description *desc)
 {
    LLVMValueRef fetch = NULL;
@@ -231,9 +231,9 @@ test_format_unorm8(unsigned verbose, FILE *fp,
    unsigned i, j, k, l;
    void *f;
 
-   fetch = add_fetch_rgba_test(verbose, desc, lp_unorm8_vec4_type());
+   fetch = add_fetch_rgba_test(gallivm, verbose, desc, lp_unorm8_vec4_type());
 
-   f = LLVMGetPointerToGlobal(lp_build_engine, fetch);
+   f = LLVMGetPointerToGlobal(gallivm->engine, fetch);
    fetch_ptr = (fetch_ptr_t) pointer_to_func(f);
 
    if (verbose >= 2) {
@@ -290,7 +290,7 @@ test_format_unorm8(unsigned verbose, FILE *fp,
    if (!success)
       LLVMDumpValue(fetch);
 
-   LLVMFreeMachineCodeForFunction(lp_build_engine, fetch);
+   LLVMFreeMachineCodeForFunction(gallivm->engine, fetch);
    LLVMDeleteFunction(fetch);
 
    if(fp)
@@ -303,16 +303,17 @@ test_format_unorm8(unsigned verbose, FILE *fp,
 
 
 static boolean
-test_one(unsigned verbose, FILE *fp,
+test_one(struct gallivm_state *gallivm,
+         unsigned verbose, FILE *fp,
          const struct util_format_description *format_desc)
 {
    boolean success = TRUE;
 
-   if (!test_format_float(verbose, fp, format_desc)) {
+   if (!test_format_float(gallivm, verbose, fp, format_desc)) {
      success = FALSE;
    }
 
-   if (!test_format_unorm8(verbose, fp, format_desc)) {
+   if (!test_format_unorm8(gallivm, verbose, fp, format_desc)) {
      success = FALSE;
    }
 
@@ -321,7 +322,7 @@ test_one(unsigned verbose, FILE *fp,
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    enum pipe_format format;
    boolean success = TRUE;
@@ -349,7 +350,7 @@ test_all(unsigned verbose, FILE *fp)
          continue;
       }
 
-      if (!test_one(verbose, fp, format_desc)) {
+      if (!test_one(gallivm, verbose, fp, format_desc)) {
            success = FALSE;
       }
    }
@@ -359,14 +360,15 @@ test_all(unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c
index 7a0d06ae2c..149ee6f125 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_main.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_main.c
@@ -380,6 +380,7 @@ int main(int argc, char **argv)
    unsigned i;
    boolean success;
    boolean single = FALSE;
+   struct gallivm_state *gallivm;
 
    for(i = 1; i < argc; ++i) {
       if(strcmp(argv[i], "-v") == 0)
@@ -394,21 +395,23 @@ int main(int argc, char **argv)
 
    lp_build_init();
 
+   gallivm = gallivm_create();
+
    util_cpu_detect();
 
    if(fp) {
       /* Warm up the caches */
-      test_some(0, NULL, 100);
+      test_some(gallivm, 0, NULL, 100);
 
       write_tsv_header(fp);
    }
       
    if (single)
-      success = test_single(verbose, fp);
+      success = test_single(gallivm, verbose, fp);
    else if (n)
-      success = test_some(verbose, fp, n);
+      success = test_some(gallivm, verbose, fp, n);
    else
-      success = test_all(verbose, fp);
+      success = test_all(gallivm, verbose, fp);
 
    if(fp)
       fclose(fp);
diff --git a/src/gallium/drivers/llvmpipe/lp_test_printf.c b/src/gallium/drivers/llvmpipe/lp_test_printf.c
index 4653f30e39..620cdb57c1 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_printf.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_printf.c
@@ -35,11 +35,6 @@
 #include "gallivm/lp_bld_assert.h"
 #include "gallivm/lp_bld_printf.h"
 
-#include <llvm-c/Analysis.h>
-#include <llvm-c/ExecutionEngine.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "lp_test.h"
 
 
@@ -63,48 +58,45 @@ typedef void (*test_printf_t)(int i);
 
 
 static LLVMValueRef
-add_printf_test(LLVMModuleRef module)
+add_printf_test(struct gallivm_state *gallivm)
 {
-   LLVMTypeRef args[1] = { LLVMIntType(32) };
-   LLVMValueRef func = LLVMAddFunction(module, "test_printf", LLVMFunctionType(LLVMVoidType(), args, 1, 0));
-   LLVMBuilderRef builder = LLVMCreateBuilder();
-   LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
+   LLVMModuleRef module = gallivm->module;
+   LLVMTypeRef args[1] = { LLVMIntTypeInContext(gallivm->context, 32) };
+   LLVMValueRef func = LLVMAddFunction(module, "test_printf", LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context), args, 1, 0));
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(gallivm->context, func, "entry");
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
    LLVMPositionBuilderAtEnd(builder, block);
-   lp_build_printf(builder, "hello, world\n");
-   lp_build_printf(builder, "print 5 6: %d %d\n", LLVMConstInt(LLVMInt32Type(), 5, 0),
-				LLVMConstInt(LLVMInt32Type(), 6, 0));
+   lp_build_printf(gallivm, "hello, world\n");
+   lp_build_printf(gallivm, "print 5 6: %d %d\n", LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 5, 0),
+				LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 6, 0));
 
    /* Also test lp_build_assert().  This should not fail. */
-   lp_build_assert(builder, LLVMConstInt(LLVMInt32Type(), 1, 0), "assert(1)");
+   lp_build_assert(gallivm, LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 1, 0), "assert(1)");
 
    LLVMBuildRetVoid(builder);
-   LLVMDisposeBuilder(builder);
+
    return func;
 }
 
 
 PIPE_ALIGN_STACK
 static boolean
-test_printf(unsigned verbose, FILE *fp, const struct printf_test_case *testcase)
+test_printf(struct gallivm_state *gallivm,
+            unsigned verbose, FILE *fp,
+            const struct printf_test_case *testcase)
 {
-   LLVMModuleRef module = NULL;
-   LLVMValueRef test = NULL;
-   LLVMExecutionEngineRef engine = NULL;
-   LLVMModuleProviderRef provider = NULL;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
+   LLVMModuleRef module = gallivm->module;
+   LLVMValueRef test;
    char *error = NULL;
-   test_printf_t test_printf;
-   float unpacked[4];
-   unsigned packed;
+   test_printf_t test_printf_func;
    boolean success = TRUE;
    void *code;
 
-   module = LLVMModuleCreateWithName("test");
-
-   test = add_printf_test(module);
+   test = add_printf_test(gallivm);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       LLVMDumpModule(module);
@@ -112,74 +104,40 @@ test_printf(unsigned verbose, FILE *fp, const struct printf_test_case *testcase)
    }
    LLVMDisposeMessage(error);
 
-   provider = LLVMCreateModuleProviderForExistingModule(module);
-#if 0
-   if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) {
-      fprintf(stderr, "%s\n", error);
-      LLVMDisposeMessage(error);
-      abort();
-   }
-#else
-   (void) provider;
-   engine = lp_build_engine;
-#endif
-
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    code = LLVMGetPointerToGlobal(engine, test);
-   test_printf = (test_printf_t)pointer_to_func(code);
-
-   memset(unpacked, 0, sizeof unpacked);
-   packed = 0;
-
+   test_printf_func = (test_printf_t) pointer_to_func(code);
 
    // LLVMDumpModule(module);
 
-   test_printf(0);
+   test_printf_func(0);
 
    LLVMFreeMachineCodeForFunction(engine, test);
 
-   LLVMDisposeExecutionEngine(engine);
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    boolean success = TRUE;
 
-   test_printf(verbose, fp, NULL);
+   test_printf(gallivm, verbose, fp, NULL);
 
    return success;
 }
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_round.c b/src/gallium/drivers/llvmpipe/lp_test_round.c
index 816518e508..4edee4af12 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_round.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_round.c
@@ -34,11 +34,6 @@
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_arit.h"
 
-#include <llvm-c/Analysis.h>
-#include <llvm-c/ExecutionEngine.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
-
 #include "lp_test.h"
 
 
@@ -64,18 +59,21 @@ typedef LLVMValueRef (*lp_func_t)(struct lp_build_context *, LLVMValueRef);
 
 
 static LLVMValueRef
-add_test(LLVMModuleRef module, const char *name, lp_func_t lp_func)
+add_test(struct gallivm_state *gallivm, const char *name, lp_func_t lp_func)
 {
-   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4);
+   LLVMModuleRef module = gallivm->module;
+   LLVMContextRef context = gallivm->context;
+   LLVMBuilderRef builder = gallivm->builder;
+
+   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(context), 4);
    LLVMTypeRef args[1] = { v4sf };
    LLVMValueRef func = LLVMAddFunction(module, name, LLVMFunctionType(v4sf, args, 1, 0));
    LLVMValueRef arg1 = LLVMGetParam(func, 0);
-   LLVMBuilderRef builder = LLVMCreateBuilder();
-   LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMValueRef ret;
    struct lp_build_context bld;
 
-   lp_build_context_init(&bld, builder, lp_float32_vec4_type());
+   lp_build_context_init(&bld, gallivm, lp_float32_vec4_type());
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
@@ -84,7 +82,7 @@ add_test(LLVMModuleRef module, const char *name, lp_func_t lp_func)
    ret = lp_func(&bld, arg1);
 
    LLVMBuildRet(builder, ret);
-   LLVMDisposeBuilder(builder);
+
    return func;
 }
 
@@ -117,12 +115,11 @@ compare(v4sf x, v4sf y)
 
 PIPE_ALIGN_STACK
 static boolean
-test_round(unsigned verbose, FILE *fp)
+test_round(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef test_round = NULL, test_trunc, test_floor, test_ceil;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
    char *error = NULL;
    test_round_t round_func, trunc_func, floor_func, ceil_func;
    float unpacked[4];
@@ -130,12 +127,10 @@ test_round(unsigned verbose, FILE *fp)
    boolean success = TRUE;
    int i;
 
-   module = LLVMModuleCreateWithName("test");
-
-   test_round = add_test(module, "round", lp_build_round);
-   test_trunc = add_test(module, "trunc", lp_build_trunc);
-   test_floor = add_test(module, "floor", lp_build_floor);
-   test_ceil = add_test(module, "ceil", lp_build_ceil);
+   test_round = add_test(gallivm, "round", lp_build_round);
+   test_trunc = add_test(gallivm, "trunc", lp_build_trunc);
+   test_floor = add_test(gallivm, "floor", lp_build_floor);
+   test_ceil = add_test(gallivm, "ceil", lp_build_ceil);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       printf("LLVMVerifyModule: %s\n", error);
@@ -144,21 +139,6 @@ test_round(unsigned verbose, FILE *fp)
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    round_func = (test_round_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_round));
    trunc_func = (test_round_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_trunc));
    floor_func = (test_round_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_floor));
@@ -229,17 +209,13 @@ test_round(unsigned verbose, FILE *fp)
    LLVMFreeMachineCodeForFunction(engine, test_floor);
    LLVMFreeMachineCodeForFunction(engine, test_ceil);
 
-   LLVMDisposeExecutionEngine(engine);
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
 #else /* !PIPE_ARCH_SSE */
 
 static boolean
-test_round(unsigned verbose, FILE *fp)
+test_round(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    return TRUE;
 }
@@ -248,20 +224,21 @@ test_round(unsigned verbose, FILE *fp)
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
-   return test_round(verbose, fp);
+   return test_round(gallivm, verbose, fp);
 }
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_sincos.c b/src/gallium/drivers/llvmpipe/lp_test_sincos.c
index 79939b1a39..77f6cb4b79 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_sincos.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_sincos.c
@@ -29,15 +29,11 @@
 #include <stdlib.h>
 #include <stdio.h>
 
+#include "util/u_pointer.h"
+
 #include "gallivm/lp_bld.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_arit.h"
-#include "util/u_pointer.h"
-
-#include <llvm-c/Analysis.h>
-#include <llvm-c/ExecutionEngine.h>
-#include <llvm-c/Target.h>
-#include <llvm-c/Transforms/Scalar.h>
 
 #include "lp_test.h"
 
@@ -61,25 +57,25 @@ write_tsv_header(FILE *fp)
 typedef __m128 (*test_sincos_t)(__m128);
 
 static LLVMValueRef
-add_sincos_test(LLVMModuleRef module, boolean sin)
+add_sincos_test(struct gallivm_state *gallivm, LLVMModuleRef module,
+                LLVMContextRef context, boolean sin)
 {
-   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatType(), 4);
+   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(context), 4);
    LLVMTypeRef args[1] = { v4sf };
    LLVMValueRef func = LLVMAddFunction(module, "sincos", LLVMFunctionType(v4sf, args, 1, 0));
    LLVMValueRef arg1 = LLVMGetParam(func, 0);
-   LLVMBuilderRef builder = LLVMCreateBuilder();
-   LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(context, func, "entry");
    LLVMValueRef ret;
    struct lp_build_context bld;
 
-   lp_build_context_init(&bld, builder, lp_float32_vec4_type());
+   lp_build_context_init(&bld, gallivm, lp_float32_vec4_type());
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
    LLVMPositionBuilderAtEnd(builder, block);
    ret = sin ? lp_build_sin(&bld, arg1) : lp_build_cos(&bld, arg1);
    LLVMBuildRet(builder, ret);
-   LLVMDisposeBuilder(builder);
    return func;
 }
 
@@ -95,22 +91,20 @@ printv(char* string, v4sf value)
 
 PIPE_ALIGN_STACK
 static boolean
-test_sincos(unsigned verbose, FILE *fp)
+test_sincos(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
-   LLVMModuleRef module = NULL;
+   LLVMModuleRef module = gallivm->module;
    LLVMValueRef test_sin = NULL, test_cos = NULL;
-   LLVMExecutionEngineRef engine = lp_build_engine;
-   LLVMPassManagerRef pass = NULL;
+   LLVMExecutionEngineRef engine = gallivm->engine;
+   LLVMContextRef context = gallivm->context;
    char *error = NULL;
    test_sincos_t sin_func;
    test_sincos_t cos_func;
    float unpacked[4];
    boolean success = TRUE;
 
-   module = LLVMModuleCreateWithName("test");
-
-   test_sin = add_sincos_test(module, TRUE);
-   test_cos = add_sincos_test(module, FALSE);
+   test_sin = add_sincos_test(gallivm, module, context, TRUE);
+   test_cos = add_sincos_test(gallivm, module, context,FALSE);
 
    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
       printf("LLVMVerifyModule: %s\n", error);
@@ -119,21 +113,6 @@ test_sincos(unsigned verbose, FILE *fp)
    }
    LLVMDisposeMessage(error);
 
-#if 0
-   pass = LLVMCreatePassManager();
-   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
-   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
-    * but there are more on SVN. */
-   LLVMAddConstantPropagationPass(pass);
-   LLVMAddInstructionCombiningPass(pass);
-   LLVMAddPromoteMemoryToRegisterPass(pass);
-   LLVMAddGVNPass(pass);
-   LLVMAddCFGSimplificationPass(pass);
-   LLVMRunPassManager(pass, module);
-#else
-   (void)pass;
-#endif
-
    sin_func = (test_sincos_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_sin));
    cos_func = (test_sincos_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_cos));
 
@@ -152,9 +131,6 @@ test_sincos(unsigned verbose, FILE *fp)
    LLVMFreeMachineCodeForFunction(engine, test_sin);
    LLVMFreeMachineCodeForFunction(engine, test_cos);
 
-   if(pass)
-      LLVMDisposePassManager(pass);
-
    return success;
 }
 
@@ -170,24 +146,25 @@ test_sincos(unsigned verbose, FILE *fp)
 
 
 boolean
-test_all(unsigned verbose, FILE *fp)
+test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    boolean success = TRUE;
 
-   test_sincos(verbose, fp);
+   test_sincos(gallivm, verbose, fp);
 
    return success;
 }
 
 
 boolean
-test_some(unsigned verbose, FILE *fp, unsigned long n)
+test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+          unsigned long n)
 {
-   return test_all(verbose, fp);
+   return test_all(gallivm, verbose, fp);
 }
 
 boolean
-test_single(unsigned verbose, FILE *fp)
+test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 {
    printf("no test_single()");
    return TRUE;
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.c b/src/gallium/drivers/llvmpipe/lp_tex_sample.c
index f417fc8a9e..ed4282937f 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample.c
@@ -43,6 +43,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_shader_tokens.h"
 #include "gallivm/lp_bld_debug.h"
+#include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_sample.h"
 #include "gallivm/lp_bld_tgsi.h"
@@ -89,7 +90,7 @@ struct lp_llvm_sampler_soa
  */
 static LLVMValueRef
 lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
-                       LLVMBuilderRef builder,
+                       struct gallivm_state *gallivm,
                        unsigned unit,
                        unsigned member_index,
                        const char *member_name,
@@ -97,6 +98,7 @@ lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
 {
    struct llvmpipe_sampler_dynamic_state *state =
       (struct llvmpipe_sampler_dynamic_state *)base;
+   LLVMBuilderRef builder = gallivm->builder;
    LLVMValueRef indices[4];
    LLVMValueRef ptr;
    LLVMValueRef res;
@@ -104,13 +106,13 @@ lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
    assert(unit < PIPE_MAX_SAMPLERS);
 
    /* context[0] */
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[0] = lp_build_const_int32(gallivm, 0);
    /* context[0].textures */
-   indices[1] = LLVMConstInt(LLVMInt32Type(), LP_JIT_CTX_TEXTURES, 0);
+   indices[1] = lp_build_const_int32(gallivm, LP_JIT_CTX_TEXTURES);
    /* context[0].textures[unit] */
-   indices[2] = LLVMConstInt(LLVMInt32Type(), unit, 0);
+   indices[2] = lp_build_const_int32(gallivm, unit);
    /* context[0].textures[unit].member */
-   indices[3] = LLVMConstInt(LLVMInt32Type(), member_index, 0);
+   indices[3] = lp_build_const_int32(gallivm, member_index);
 
    ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
 
@@ -137,10 +139,10 @@ lp_llvm_texture_member(const struct lp_sampler_dynamic_state *base,
 #define LP_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load)  \
    static LLVMValueRef \
    lp_llvm_texture_##_name( const struct lp_sampler_dynamic_state *base, \
-                            LLVMBuilderRef builder, \
+                            struct gallivm_state *gallivm, \
                             unsigned unit) \
    { \
-      return lp_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \
+      return lp_llvm_texture_member(base, gallivm, unit, _index, #_name, _emit_load ); \
    }
 
 
@@ -170,7 +172,7 @@ lp_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
  */
 static void
 lp_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base,
-                                     LLVMBuilderRef builder,
+                                     struct gallivm_state *gallivm,
                                      struct lp_type type,
                                      unsigned unit,
                                      unsigned num_coords,
@@ -186,11 +188,11 @@ lp_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base,
    assert(unit < PIPE_MAX_SAMPLERS);
    
    if (LP_PERF & PERF_NO_TEX) {
-      lp_build_sample_nop(type, texel);
+      lp_build_sample_nop(gallivm, type, texel);
       return;
    }
 
-   lp_build_sample_soa(builder,
+   lp_build_sample_soa(gallivm,
                        &sampler->dynamic_state.static_state[unit],
                        &sampler->dynamic_state.base,
                        type,