radeonsi: store shader disassemblies in memory for future users

This will be used by the new ddebug pipe. I'm including it now to avoid
conflicts with other patches.

src/gallium/drivers/r600/evergreen_compute.c

@@ -226,7 +226,7 @@ void *evergreen_create_compute_state(
 	}
 #else
 	memset(&shader->binary, 0, sizeof(shader->binary));
-	radeon_elf_read(code, header->num_bytes, &shader->binary, true);
+	radeon_elf_read(code, header->num_bytes, &shader->binary);
 	r600_create_shader(&shader->bc, &shader->binary, &use_kill);
 
 	shader->code_bo = r600_compute_buffer_alloc_vram(ctx->screen,

src/gallium/drivers/radeon/r600_pipe_common.h

@@ -129,9 +129,8 @@ struct radeon_shader_binary {
 	struct radeon_shader_reloc *relocs;
 	unsigned reloc_count;
 
-	/** Set to 1 if the disassembly for this binary has been dumped to
-	 *  stderr. */
-	int disassembled;
+	/** Disassembled shader in a string. */
+	char *disasm_string;
 };
 
 struct r600_resource {

src/gallium/drivers/radeon/radeon_elf_util.c

@@ -103,8 +103,7 @@ static void parse_relocs(Elf *elf, Elf_Data *relocs, Elf_Data *symbols,
 }
 
 void radeon_elf_read(const char *elf_data, unsigned elf_size,
-					struct radeon_shader_binary *binary,
-					unsigned debug)
+		     struct radeon_shader_binary *binary)
 {
 	char *elf_buffer;
 	Elf *elf;
@@ -124,7 +123,6 @@ void radeon_elf_read(const char *elf_data, unsigned elf_size,
 	elf = elf_memory(elf_buffer, elf_size);
 
 	elf_getshdrstrndx(elf, &section_str_index);
-	binary->disassembled = 0;
 
 	while ((section = elf_nextscn(elf, section))) {
 		const char *name;
@@ -145,12 +143,11 @@ void radeon_elf_read(const char *elf_data, unsigned elf_size,
 			binary->config_size = section_data->d_size;
 			binary->config = MALLOC(binary->config_size * sizeof(unsigned char));
 			memcpy(binary->config, section_data->d_buf, binary->config_size);
-		} else if (debug && !strcmp(name, ".AMDGPU.disasm")) {
-			binary->disassembled = 1;
+		} else if (!strcmp(name, ".AMDGPU.disasm")) {
+			/* Always read disassembly if it's available. */
 			section_data = elf_getdata(section, section_data);
-			fprintf(stderr, "\nShader Disassembly:\n\n");
-			fprintf(stderr, "%.*s\n", (int)section_data->d_size,
-						  (char *)section_data->d_buf);
+			binary->disasm_string = strndup(section_data->d_buf,
+							section_data->d_size);
 		} else if (!strncmp(name, ".rodata", 7)) {
 			section_data = elf_getdata(section, section_data);
 			binary->rodata_size = section_data->d_size;

src/gallium/drivers/radeon/radeon_elf_util.h

@@ -37,7 +37,7 @@ struct radeon_shader_reloc;
  * radeon_shader_binary object.
  */
 void radeon_elf_read(const char *elf_data, unsigned elf_size,
-		struct radeon_shader_binary *binary, unsigned debug);
+		     struct radeon_shader_binary *binary);
 
 /**
  * @returns A pointer to the start of the configuration information for

src/gallium/drivers/radeon/radeon_llvm_emit.c

@@ -210,7 +210,7 @@ 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);

src/gallium/drivers/radeonsi/si_compute.c

@@ -137,7 +137,7 @@ static void *si_create_compute_state(
 	}
 #else
 
-	radeon_elf_read(code, header->num_bytes, &program->shader.binary, true);
+	radeon_elf_read(code, header->num_bytes, &program->shader.binary);
 
 	/* init_scratch_buffer patches the shader code with the scratch address,
 	 * so we need to call it before si_shader_binary_read() which uploads

src/gallium/drivers/radeonsi/si_shader.c

@@ -3811,7 +3811,10 @@ int si_shader_binary_read(struct si_screen *sscreen, struct si_shader *shader)
 	si_shader_binary_upload(sscreen, shader);
 
 	if (dump) {
-		if (!binary->disassembled) {
+		if (binary->disasm_string) {
+			fprintf(stderr, "\nShader Disassembly:\n\n");
+			fprintf(stderr, "%s\n", binary->disasm_string);
+		} else {
 			fprintf(stderr, "SI CODE:\n");
 			for (i = 0; i < binary->code_size; i+=4 ) {
 				fprintf(stderr, "@0x%x: %02x%02x%02x%02x\n", i, binary->code[i + 3],
@@ -3849,7 +3852,8 @@ int si_compile_llvm(struct si_screen *sscreen, struct si_shader *shader,
 	if (shader->scratch_bytes_per_wave == 0) {
 		FREE(shader->binary.code);
 		FREE(shader->binary.relocs);
-		memset(&shader->binary, 0, sizeof(shader->binary));
+		memset(&shader->binary, 0,
+		       offsetof(struct radeon_shader_binary, disasm_string));
 	}
 	return r;
 }
@@ -4176,4 +4180,5 @@ void si_shader_destroy(struct pipe_context *ctx, struct si_shader *shader)
 
 	FREE(shader->binary.code);
 	FREE(shader->binary.relocs);
+	FREE(shader->binary.disasm_string);
 }