gallivm: Add an alternative to LLVMDumpValue that works with Windows GUI apps.

[mesa.git] / src / gallium / auxiliary / draw / draw_llvm.c

diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index 27383221b9bcb7684d3aed32f0df86994dd54485..cfafa183c81e16c25a67b0dad68b94ceaa051fc3 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -161,9 +161,16 @@ create_vertex_header(struct draw_llvm *llvm, int data_elems)
 struct draw_llvm *
 draw_llvm_create(struct draw_context *draw)
 {
-   struct draw_llvm *llvm = CALLOC_STRUCT( draw_llvm );
+   struct draw_llvm *llvm;
 
+#ifdef PIPE_ARCH_X86
    util_cpu_detect();
+   /* require SSE2 due to LLVM PR6960. */
+   if (!util_cpu_caps.has_sse2)
+       return NULL;
+#endif
+
+   llvm = CALLOC_STRUCT( draw_llvm );
 
    llvm->draw = draw;
    llvm->engine = draw->engine;
@@ -182,6 +189,8 @@ draw_llvm_create(struct draw_context *draw)
    /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
     * but there are more on SVN. */
    /* TODO: Add more passes */
+   LLVMAddCFGSimplificationPass(llvm->pass);
+   LLVMAddPromoteMemoryToRegisterPass(llvm->pass);
    LLVMAddConstantPropagationPass(llvm->pass);
    if(util_cpu_caps.has_sse4_1) {
       /* FIXME: There is a bug in this pass, whereby the combination of fptosi
@@ -190,15 +199,13 @@ draw_llvm_create(struct draw_context *draw)
        */
       LLVMAddInstructionCombiningPass(llvm->pass);
    }
-   LLVMAddPromoteMemoryToRegisterPass(llvm->pass);
    LLVMAddGVNPass(llvm->pass);
-   LLVMAddCFGSimplificationPass(llvm->pass);
 
    init_globals(llvm);
 
 
 #if 0
-   LLVMDumpModule(llvm->module);
+   LLVMDumpModule(lp_build_module);
 #endif
 
    return llvm;
@@ -207,6 +214,8 @@ draw_llvm_create(struct draw_context *draw)
 void
 draw_llvm_destroy(struct draw_llvm *llvm)
 {
+   LLVMDisposePassManager(llvm->pass);
+
    FREE(llvm);
 }
 
@@ -682,7 +691,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
     */
 #ifdef DEBUG
    if(LLVMVerifyFunction(variant->function, LLVMPrintMessageAction)) {
-      LLVMDumpValue(variant->function);
+      lp_debug_dump_value(variant->function);
       assert(0);
    }
 #endif
@@ -690,7 +699,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    LLVMRunFunctionPassManager(llvm->pass, variant->function);
 
    if (0) {
-      LLVMDumpValue(variant->function);
+      lp_debug_dump_value(variant->function);
       debug_printf("\n");
    }
    variant->jit_func = (draw_jit_vert_func)LLVMGetPointerToGlobal(llvm->draw->engine, variant->function);
@@ -713,10 +722,12 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    struct draw_context *draw = llvm->draw;
    unsigned i, j;
    struct lp_build_context bld;
+   struct lp_build_context bld_int;
    struct lp_build_loop_state lp_loop;
    struct lp_type vs_type = lp_type_float_vec(32);
    const int max_vertices = 4;
    LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
+   LLVMValueRef fetch_max;
 
    arg_types[0] = llvm->context_ptr_type;               /* context */
    arg_types[1] = llvm->vertex_header_ptr_type;         /* vertex_header */
@@ -759,9 +770,14 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    LLVMPositionBuilderAtEnd(builder, block);
 
    lp_build_context_init(&bld, builder, vs_type);
+   lp_build_context_init(&bld_int, builder, lp_type_int(32));
 
    step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0);
 
+   fetch_max = LLVMBuildSub(builder, fetch_count,
+                            LLVMConstInt(LLVMInt32Type(), 1, 0),
+                            "fetch_max");
+
    lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), &lp_loop);
    {
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
@@ -780,8 +796,15 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
             builder,
             lp_loop.counter,
             LLVMConstInt(LLVMInt32Type(), i, 0), "");
-         LLVMValueRef fetch_ptr = LLVMBuildGEP(builder, fetch_elts,
-                                               &true_index, 1, "");
+         LLVMValueRef fetch_ptr;
+
+         /* make sure we're not out of bounds which can happen
+          * if fetch_count % 4 != 0, because on the last iteration
+          * a few of the 4 vertex fetches will be out of bounds */
+         true_index = lp_build_min(&bld_int, true_index, fetch_max);
+
+         fetch_ptr = LLVMBuildGEP(builder, fetch_elts,
+                                  &true_index, 1, "");
          true_index = LLVMBuildLoad(builder, fetch_ptr, "fetch_elt");
          for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
             struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
@@ -819,7 +842,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
     */
 #ifdef DEBUG
    if(LLVMVerifyFunction(variant->function_elts, LLVMPrintMessageAction)) {
-      LLVMDumpValue(variant->function_elts);
+      lp_debug_dump_value(variant->function_elts);
       assert(0);
    }
 #endif
@@ -827,7 +850,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    LLVMRunFunctionPassManager(llvm->pass, variant->function_elts);
 
    if (0) {
-      LLVMDumpValue(variant->function_elts);
+      lp_debug_dump_value(variant->function_elts);
       debug_printf("\n");
    }
    variant->jit_func_elts = (draw_jit_vert_func_elts)LLVMGetPointerToGlobal(