radeonsi: print LLVM IRs to ddebug logs

Getting LLVM IRs of hanging shaders have never been easier.

Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>

6 files changed, 26 insertions, 1 deletions

diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c
index 5e981d617c4..74674e70e72 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.c
+++ b/src/gallium/drivers/radeon/r600_pipe_common.c
@@ -67,6 +67,7 @@ void radeon_shader_binary_clean(struct radeon_shader_binary *b)
 	FREE(b->global_symbol_offsets);
 	FREE(b->relocs);
 	FREE(b->disasm_string);
+	FREE(b->llvm_ir_string);
 }
 
 /*
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h
index c145dc3a069..1ad69f85bb4 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -155,6 +155,7 @@ struct radeon_shader_binary {
 
 	/** Disassembled shader in a string. */
 	char *disasm_string;
+	char *llvm_ir_string;
 };
 
 void radeon_shader_binary_init(struct radeon_shader_binary *b);
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index f15e58904ec..ad2a86a812b 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -139,6 +139,9 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen,
 	if (sscreen->b.debug_flags & DBG_CHECK_VM)
 		flags |= PIPE_CONTEXT_DEBUG;
 
+	if (flags & PIPE_CONTEXT_DEBUG)
+		sscreen->record_llvm_ir = true; /* racy but not critical */
+
 	sctx->b.b.screen = screen; /* this must be set first */
 	sctx->b.b.priv = priv;
 	sctx->b.b.destroy = si_destroy_context;
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 9d15cbf2225..fc7e73e01a7 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -87,6 +87,7 @@ struct si_screen {
 
 	/* Whether shaders are monolithic (1-part) or separate (3-part). */
 	bool				use_monolithic_shaders;
+	bool				record_llvm_ir;
 
 	pipe_mutex			shader_parts_mutex;
 	struct si_shader_part		*vs_prologs;
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 4dcc8c835cd..77d1a8b96e8 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -6226,6 +6226,12 @@ void si_shader_dump(struct si_screen *sscreen, struct si_shader *shader,
 		    struct pipe_debug_callback *debug, unsigned processor,
 		    FILE *file)
 {
+	if (file != stderr && shader->binary.llvm_ir_string) {
+		fprintf(file, "\n%s - main shader part - LLVM IR:\n\n",
+			si_get_shader_name(shader, processor));
+		fprintf(file, "%s\n", shader->binary.llvm_ir_string);
+	}
+
 	if (file != stderr ||
 	    (r600_can_dump_shader(&sscreen->b, processor) &&
 	     !(sscreen->b.debug_flags & DBG_NO_ASM))) {
@@ -6271,6 +6277,12 @@ int si_compile_llvm(struct si_screen *sscreen,
 		}
 	}
 
+	if (sscreen->record_llvm_ir) {
+		char *ir = LLVMPrintModuleToString(mod);
+		binary->llvm_ir_string = strdup(ir);
+		LLVMDisposeMessage(ir);
+	}
+
 	if (!si_replace_shader(count, binary)) {
 		r = radeon_llvm_compile(mod, binary, tm, debug);
 		if (r)
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index f78d0821f92..766ef2c6549 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -96,6 +96,8 @@ static uint32_t *read_chunk(uint32_t *ptr, void **data, unsigned *size)
 {
 	*size = *ptr++;
 	assert(*data == NULL);
+	if (!*size)
+		return ptr;
 	*data = malloc(*size);
 	return read_data(ptr, *data, *size);
 }
@@ -110,6 +112,8 @@ static void *si_get_shader_binary(struct si_shader *shader)
 	unsigned relocs_size = shader->binary.reloc_count *
 			       sizeof(shader->binary.relocs[0]);
 	unsigned disasm_size = strlen(shader->binary.disasm_string) + 1;
+	unsigned llvm_ir_size = shader->binary.llvm_ir_string ?
+				strlen(shader->binary.llvm_ir_string) + 1 : 0;
 	unsigned size =
 		4 + /* total size */
 		4 + /* CRC32 of the data below */
@@ -118,7 +122,8 @@ static void *si_get_shader_binary(struct si_shader *shader)
 		4 + align(shader->binary.code_size, 4) +
 		4 + align(shader->binary.rodata_size, 4) +
 		4 + align(relocs_size, 4) +
-		4 + align(disasm_size, 4);
+		4 + align(disasm_size, 4) +
+		4 + align(llvm_ir_size, 4);
 	void *buffer = CALLOC(1, size);
 	uint32_t *ptr = (uint32_t*)buffer;
 
@@ -134,6 +139,7 @@ static void *si_get_shader_binary(struct si_shader *shader)
 	ptr = write_chunk(ptr, shader->binary.rodata, shader->binary.rodata_size);
 	ptr = write_chunk(ptr, shader->binary.relocs, relocs_size);
 	ptr = write_chunk(ptr, shader->binary.disasm_string, disasm_size);
+	ptr = write_chunk(ptr, shader->binary.llvm_ir_string, llvm_ir_size);
 	assert((char *)ptr - (char *)buffer == size);
 
 	/* Compute CRC32. */
@@ -165,6 +171,7 @@ static bool si_load_shader_binary(struct si_shader *shader, void *binary)
 	ptr = read_chunk(ptr, (void**)&shader->binary.relocs, &chunk_size);
 	shader->binary.reloc_count = chunk_size / sizeof(shader->binary.relocs[0]);
 	ptr = read_chunk(ptr, (void**)&shader->binary.disasm_string, &chunk_size);
+	ptr = read_chunk(ptr, (void**)&shader->binary.llvm_ir_string, &chunk_size);
 
 	return true;
 }