1 files changed, 58 insertions, 3 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index d288460a1b..13535dd638 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -36,23 +36,57 @@
 #include <llvm-c/Transforms/Scalar.h>
 
 #include "util/u_memory.h"
+#include "util/u_cpu_detect.h"
 #include "lp_screen.h"
 #include "lp_bld_intr.h"
+#include "lp_bld_misc.h"
 #include "lp_jit.h"
 
 
 static void
 lp_jit_init_globals(struct llvmpipe_screen *screen)
 {
-   /* struct lp_jit_context */
+   LLVMTypeRef texture_type;
+
+   /* struct lp_jit_texture */
    {
       LLVMTypeRef elem_types[4];
+
+      elem_types[LP_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_STRIDE] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_DATA]   = LLVMPointerType(LLVMInt8Type(), 0);
+
+      texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, width,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_WIDTH);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, height,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_HEIGHT);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, stride,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_STRIDE);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, data,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_DATA);
+      LP_CHECK_STRUCT_SIZE(struct lp_jit_texture,
+                           screen->target, texture_type);
+
+      LLVMAddTypeName(screen->module, "texture", texture_type);
+   }
+
+   /* struct lp_jit_context */
+   {
+      LLVMTypeRef elem_types[5];
       LLVMTypeRef context_type;
 
       elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* constants */
       elem_types[1] = LLVMPointerType(LLVMInt8Type(), 0);  /* samplers */
       elem_types[2] = LLVMFloatType();                     /* alpha_ref_value */
       elem_types[3] = LLVMPointerType(LLVMInt8Type(), 0);  /* blend_color */
+      elem_types[4] = LLVMArrayType(texture_type, PIPE_MAX_SAMPLERS); /* textures */
 
       context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
 
@@ -64,6 +98,9 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
                              screen->target, context_type, 2);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, blend_color,
                              screen->target, context_type, 3);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, textures,
+                             screen->target, context_type,
+                             LP_JIT_CONTEXT_TEXTURES_INDEX);
       LP_CHECK_STRUCT_SIZE(struct lp_jit_context,
                            screen->target, context_type);
 
@@ -112,12 +149,23 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
 {
    char *error = NULL;
 
+   util_cpu_detect();
+
+#if 0
+   /* For simulating less capable machines */
+   util_cpu_caps.has_sse3 = 0;
+   util_cpu_caps.has_sse4_1 = 0;
+#endif
+
+   LLVMLinkInJIT();
+   LLVMInitializeNativeTarget();
+
    screen->module = LLVMModuleCreateWithName("llvmpipe");
 
    screen->provider = LLVMCreateModuleProviderForExistingModule(screen->module);
 
    if (LLVMCreateJITCompiler(&screen->engine, screen->provider, 1, &error)) {
-      fprintf(stderr, "%s\n", error);
+      _debug_printf("%s\n", error);
       LLVMDisposeMessage(error);
       abort();
    }
@@ -128,8 +176,15 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
    LLVMAddTargetData(screen->target, screen->pass);
    /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
     * but there are more on SVN. */
+   /* TODO: Add more passes */
    LLVMAddConstantPropagationPass(screen->pass);
-   LLVMAddInstructionCombiningPass(screen->pass);
+   if(util_cpu_caps.has_sse4_1) {
+      /* FIXME: There is a bug in this pass, whereby the combination of fptosi
+       * and sitofp (necessary for trunc/floor/ceil/round implementation)
+       * somehow becomes invalid code.
+       */
+      LLVMAddInstructionCombiningPass(screen->pass);
+   }
    LLVMAddPromoteMemoryToRegisterPass(screen->pass);
    LLVMAddGVNPass(screen->pass);
    LLVMAddCFGSimplificationPass(screen->pass);