radeonsi: store shader disassemblies in memory for future users

[mesa.git] / src / gallium / drivers / radeon / radeon_llvm_emit.c

diff --git a/src/gallium/drivers/radeon/radeon_llvm_emit.c b/src/gallium/drivers/radeon/radeon_llvm_emit.c
index 1b17dd4613f25eddf1d966c01f09c1733d7ba8dd..04f62d13ca78d2bb43bd1493690fa0a225baee9b 100644 (file)
--- a/src/gallium/drivers/radeon/radeon_llvm_emit.c
+++ b/src/gallium/drivers/radeon/radeon_llvm_emit.c
@@ -26,6 +26,7 @@
 #include "radeon_llvm_emit.h"
 #include "radeon_elf_util.h"
 #include "util/u_memory.h"
+#include "pipe/p_shader_tokens.h"
 
 #include <llvm-c/Target.h>
 #include <llvm-c/TargetMachine.h>
@@ -39,6 +40,16 @@
 #define FS_STRING_LEN 30
 #define TRIPLE_STRING_LEN 7
 
+/**
+ * Shader types for the LLVM backend.
+ */
+enum radeon_llvm_shader_type {
+       RADEON_LLVM_SHADER_PS = 0,
+       RADEON_LLVM_SHADER_VS = 1,
+       RADEON_LLVM_SHADER_GS = 2,
+       RADEON_LLVM_SHADER_CS = 3,
+};
+
 /**
  * Set the shader type we want to compile
  *
@@ -46,35 +57,70 @@
  */
 void radeon_llvm_shader_type(LLVMValueRef F, unsigned type)
 {
-  char Str[2];
-  sprintf(Str, "%1d", type);
+       char Str[2];
+       enum radeon_llvm_shader_type llvm_type;
+
+       switch (type) {
+       case TGSI_PROCESSOR_VERTEX:
+       case TGSI_PROCESSOR_TESS_CTRL:
+       case TGSI_PROCESSOR_TESS_EVAL:
+               llvm_type = RADEON_LLVM_SHADER_VS;
+               break;
+       case TGSI_PROCESSOR_GEOMETRY:
+               llvm_type = RADEON_LLVM_SHADER_GS;
+               break;
+       case TGSI_PROCESSOR_FRAGMENT:
+               llvm_type = RADEON_LLVM_SHADER_PS;
+               break;
+       case TGSI_PROCESSOR_COMPUTE:
+               llvm_type = RADEON_LLVM_SHADER_CS;
+               break;
+       default:
+               assert(0);
+       }
 
-  LLVMAddTargetDependentFunctionAttr(F, "ShaderType", Str);
+       sprintf(Str, "%1d", llvm_type);
+
+       LLVMAddTargetDependentFunctionAttr(F, "ShaderType", Str);
+
+       if (type != TGSI_PROCESSOR_COMPUTE) {
+               LLVMAddTargetDependentFunctionAttr(F, "unsafe-fp-math", "true");
+       }
 }
 
-static void init_r600_target() {
+static void init_r600_target()
+{
        static unsigned initialized = 0;
        if (!initialized) {
+#if HAVE_LLVM < 0x0307
                LLVMInitializeR600TargetInfo();
                LLVMInitializeR600Target();
                LLVMInitializeR600TargetMC();
                LLVMInitializeR600AsmPrinter();
+#else
+               LLVMInitializeAMDGPUTargetInfo();
+               LLVMInitializeAMDGPUTarget();
+               LLVMInitializeAMDGPUTargetMC();
+               LLVMInitializeAMDGPUAsmPrinter();
+
+#endif
                initialized = 1;
        }
 }
 
-static LLVMTargetRef get_r600_target() {
+LLVMTargetRef radeon_llvm_get_r600_target(const char *triple)
+{
        LLVMTargetRef target = NULL;
+       char *err_message = NULL;
 
-       for (target = LLVMGetFirstTarget(); target;
-                                       target = LLVMGetNextTarget(target)) {
-               if (!strncmp(LLVMGetTargetName(target), "r600", 4)) {
-                       break;
-               }
-       }
+       init_r600_target();
 
-       if (!target) {
-               fprintf(stderr, "Can't find target r600\n");
+       if (LLVMGetTargetFromTriple(triple, &target, &err_message)) {
+               fprintf(stderr, "Cannot find target for triple %s ", triple);
+               if (err_message) {
+                       fprintf(stderr, "%s\n", err_message);
+               }
+               LLVMDisposeMessage(err_message);
                return NULL;
        }
        return target;
@@ -82,7 +128,8 @@ static LLVMTargetRef get_r600_target() {
 
 #if HAVE_LLVM >= 0x0305
 
-static void radeonDiagnosticHandler(LLVMDiagnosticInfoRef di, void *context) {
+static void radeonDiagnosticHandler(LLVMDiagnosticInfoRef di, void *context)
+{
        if (LLVMGetDiagInfoSeverity(di) == LLVMDSError) {
                unsigned int *diagnosticflag = (unsigned int *)context;
                char *diaginfo_message = LLVMGetDiagInfoDescription(di);
@@ -101,13 +148,13 @@ static void radeonDiagnosticHandler(LLVMDiagnosticInfoRef di, void *context) {
  * @returns 0 for success, 1 for failure
  */
 unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binary,
-                                         const char * gpu_family, unsigned dump) {
+                         const char *gpu_family, unsigned dump, LLVMTargetMachineRef tm)
+{
 
-       LLVMTargetRef target;
-       LLVMTargetMachineRef tm;
        char cpu[CPU_STRING_LEN];
        char fs[FS_STRING_LEN];
        char *err;
+       bool dispose_tm = false;
        LLVMContextRef llvm_ctx;
        unsigned rval = 0;
        LLVMMemoryBufferRef out_buffer;
@@ -116,22 +163,25 @@ unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binar
        char triple[TRIPLE_STRING_LEN];
        LLVMBool mem_err;
 
-       /* initialise */
-       init_r600_target();
-
-       target = get_r600_target();
-       if (!target) {
-               return 1;
+       if (!tm) {
+               strncpy(triple, "r600--", TRIPLE_STRING_LEN);
+               LLVMTargetRef target = radeon_llvm_get_r600_target(triple);
+               if (!target) {
+                       return 1;
+               }
+               strncpy(cpu, gpu_family, CPU_STRING_LEN);
+               memset(fs, 0, sizeof(fs));
+               if (dump) {
+                       strncpy(fs, "+DumpCode", FS_STRING_LEN);
+               }
+               tm = LLVMCreateTargetMachine(target, triple, cpu, fs,
+                                 LLVMCodeGenLevelDefault, LLVMRelocDefault,
+                                                 LLVMCodeModelDefault);
+               dispose_tm = true;
        }
-
-       strncpy(cpu, gpu_family, CPU_STRING_LEN);
-       memset(fs, 0, sizeof(fs));
        if (dump) {
                LLVMDumpModule(M);
-               strncpy(fs, "+DumpCode", FS_STRING_LEN);
        }
-       strncpy(triple, "r600--", TRIPLE_STRING_LEN);
-
        /* Setup Diagnostic Handler*/
        llvm_ctx = LLVMGetModuleContext(M);
 
@@ -141,9 +191,6 @@ unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binar
        rval = 0;
 
        /* Compile IR*/
-       tm = LLVMCreateTargetMachine(target, triple, cpu, fs,
-                                 LLVMCodeGenLevelDefault, LLVMRelocDefault,
-                                                 LLVMCodeModelDefault);
        mem_err = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile, &err,
                                                                 &out_buffer);
 
@@ -163,10 +210,13 @@ unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binar
        buffer_size = LLVMGetBufferSize(out_buffer);
        buffer_data = LLVMGetBufferStart(out_buffer);
 
-       radeon_elf_read(buffer_data, buffer_size, binary, dump);
+       radeon_elf_read(buffer_data, buffer_size, binary);
 
        /* Clean up */
        LLVMDisposeMemoryBuffer(out_buffer);
-       LLVMDisposeTargetMachine(tm);
+
+       if (dispose_tm) {
+               LLVMDisposeTargetMachine(tm);
+       }
        return rval;
 }