gallium/radeon: enable glsl-to-llvmglsl-to-llvm-05nov

11 files changed, 102 insertions, 7 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index 3c2a923181..fb065a15ab 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -1658,6 +1658,8 @@ void *ureg_create_shader( struct ureg_program *ureg,
    else
       memset(&state.stream_output, 0, sizeof(state.stream_output));
 
+   state.llvm_module = NULL;
+
    if (ureg->processor == TGSI_PROCESSOR_VERTEX)
       return pipe->create_vs_state( pipe, &state );
    else
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 219bd54304..140ca5fe37 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -279,6 +279,7 @@ struct r600_dsa_state {
 struct r600_pipe_shader;
 
 struct r600_pipe_shader_selector {
+   struct LLVMOpaqueModule * modinfo;
 	struct r600_pipe_shader *current;
 
 	struct tgsi_token       *tokens;
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 72cb58582f..6e68b5e5e8 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -26,6 +26,7 @@
 #include "r600_opcodes.h"
 #include "r600_shader.h"
 #include "r600d.h"
+#include "R600ReserveReg.h"
 
 #include "pipe/p_shader_tokens.h"
 #include "tgsi/tgsi_info.h"
@@ -141,6 +142,7 @@ int r600_pipe_shader_create(struct pipe_context *ctx,
 			}
 		}
 	}
+	shader->modinfo = sel->modinfo;
 	r = r600_shader_from_tgsi(rctx->screen, shader, key);
 	if (r) {
 		R600_ERR("translation from TGSI failed !\n");
@@ -344,6 +346,8 @@ static unsigned r600_alu_from_byte_stream(struct r600_shader_ctx *ctx,
 		alu.last = 1;
 	}
 #endif
+	
+	ctx->bc->ar_loaded = 1;
 
 	if (alu.execute_mask) {
 		alu.pred_sel = 0;
@@ -1348,6 +1352,31 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
 		}
 	}
 
+	int col1 = -1, bcol1 = -1, col2 = -1, bcol2 = -1;
+	unsigned is_eg = rscreen->chip_class >= EVERGREEN;
+	for (i = 0; i < ctx.shader->ninput; i++) {
+		if (ctx.shader->input[i].name == TGSI_SEMANTIC_COLOR) {
+			if (col1 == -1)
+				col1 = is_eg?ctx.shader->input[i].lds_pos:ctx.shader->input[i].gpr;
+			else
+				col2 = is_eg?ctx.shader->input[i].lds_pos:ctx.shader->input[i].gpr;
+		}
+		if (ctx.shader->input[i].name == TGSI_SEMANTIC_BCOLOR) {
+			if (bcol1 == -1)
+				bcol1 = is_eg?ctx.shader->input[i].lds_pos:ctx.shader->input[i].gpr;
+			else
+				bcol2 = is_eg?ctx.shader->input[i].lds_pos:ctx.shader->input[i].gpr;
+		}
+	}
+	
+	unsigned to_gpr_map[16], input_count = ctx.shader->ninput;
+	unsigned id = 0;
+	for (unsigned i = 0; i < input_count; i++) {
+		if (ctx.shader->input[i].name == TGSI_SEMANTIC_FACE)
+			continue;
+		to_gpr_map[id++] = ctx.shader->input[i].gpr;
+	}
+
 /* LLVM backend setup */
 #ifdef R600_USE_LLVM
 	if (use_llvm) {
@@ -1364,14 +1393,24 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
 		radeon_llvm_ctx.color_buffer_count = MAX2(key.nr_cbufs , 1);
 		radeon_llvm_ctx.chip_class = ctx.bc->chip_class;
 		radeon_llvm_ctx.fs_color_all = shader->fs_write_all && (rscreen->chip_class >= EVERGREEN);
-		mod = r600_tgsi_llvm(&radeon_llvm_ctx, tokens);
+		if (pipeshader->modinfo)
+		{
+			unsigned *input_map = (rscreen->chip_class >= EVERGREEN)? NULL : to_gpr_map;
+			mod = renumber_input_ref(pipeshader->modinfo, 4 * radeon_llvm_ctx.reserved_reg_count, shader->input[ctx.fragcoord_input].gpr, ctx.face_gpr,
+				shader->two_side, col1, bcol1, col2, bcol2, radeon_llvm_ctx.color_buffer_count, radeon_llvm_ctx.fs_color_all, input_map);
+		}
+		else
+		   mod = r600_tgsi_llvm(&radeon_llvm_ctx, tokens);
 		if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE)) {
 			dump = 1;
 		}
 		if (r600_llvm_compile(mod, &inst_bytes, &inst_byte_count,
 							rscreen->family, dump)) {
 			FREE(inst_bytes);
-			radeon_llvm_dispose(&radeon_llvm_ctx);
+			if (pipeshader->modinfo)
+				LLVMDisposeModule(mod);
+			else
+				radeon_llvm_dispose(&radeon_llvm_ctx);
 			use_llvm = 0;
 			fprintf(stderr, "R600 LLVM backend failed to compile "
 				"shader.  Falling back to TGSI\n");
@@ -1379,7 +1418,10 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
 			ctx.file_offset[TGSI_FILE_OUTPUT] =
 					ctx.file_offset[TGSI_FILE_INPUT];
 		}
-		radeon_llvm_dispose(&radeon_llvm_ctx);
+		if (pipeshader->modinfo)
+			LLVMDisposeModule(mod);
+		else
+			radeon_llvm_dispose(&radeon_llvm_ctx);
 	}
 #endif
 /* End of LLVM backend setup */
diff --git a/src/gallium/drivers/r600/r600_shader.h b/src/gallium/drivers/r600/r600_shader.h
index b58a58ab4d..de93564976 100644
--- a/src/gallium/drivers/r600/r600_shader.h
+++ b/src/gallium/drivers/r600/r600_shader.h
@@ -70,6 +70,7 @@ struct r600_shader_key {
 };
 
 struct r600_pipe_shader {
+	struct LLVMOpaqueModule * modinfo;
 	struct r600_pipe_shader_selector *selector;
 	struct r600_pipe_shader	*next_variant;
 	struct r600_shader	shader;
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 926cb1ae88..2cb1b5c627 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -33,6 +33,7 @@
 #include "util/u_upload_mgr.h"
 #include "tgsi/tgsi_parse.h"
 #include <byteswap.h>
+#include <llvm-c/Core.h>
 
 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
 
@@ -746,6 +747,7 @@ static int r600_shader_select(struct pipe_context *ctx,
 			key = r600_shader_selector_key(ctx, sel);
 		}
 
+		shader->modinfo = sel->modinfo;
 		shader->key = key;
 		sel->num_shaders++;
 	}
@@ -774,6 +776,7 @@ static void *r600_create_shader_state(struct pipe_context *ctx,
 	sel->type = pipe_shader_type;
 	sel->tokens = tgsi_dup_tokens(state->tokens);
 	sel->so = state->stream_output;
+	sel->modinfo = state->llvm_module;
 
 	r = r600_shader_select(ctx, sel, NULL);
 	if (r)
@@ -868,6 +871,9 @@ static void r600_delete_ps_state(struct pipe_context *ctx, void *state)
 		rctx->ps_shader = NULL;
 	}
 
+	if (sel->modinfo)
+		LLVMDisposeModule(sel->modinfo);
+
 	r600_delete_shader_selector(ctx, sel);
 }
 
@@ -880,6 +886,9 @@ static void r600_delete_vs_state(struct pipe_context *ctx, void *state)
 		rctx->vs_shader = NULL;
 	}
 
+	if (sel->modinfo)
+		LLVMDisposeModule(sel->modinfo);
+
 	r600_delete_shader_selector(ctx, sel);
 }
 
diff --git a/src/gallium/drivers/radeon/radeon_llvm_emit.cpp b/src/gallium/drivers/radeon/radeon_llvm_emit.cpp
index 21661a0500..045b090712 100644
--- a/src/gallium/drivers/radeon/radeon_llvm_emit.cpp
+++ b/src/gallium/drivers/radeon/radeon_llvm_emit.cpp
@@ -128,6 +128,7 @@ radeon_llvm_compile(LLVMModuleRef M, unsigned char ** bytes,
 #else
    PM.add(new DataLayout(*AMDGPUTargetMachine.getDataLayout()));
 #endif
+   PM.add(createScalarReplAggregatesPass(4096));
    PM.add(createPromoteMemoryToRegisterPass());
    AMDGPUTargetMachine.setAsmVerbosityDefault(true);
 
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index f84bfd79d6..d3d2d371cb 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -49,6 +49,7 @@
 extern "C" {
 #endif
 
+typedef  struct LLVMOpaqueModule * LLVMModuleRef;
 
 /**
  * Implementation limits
@@ -209,6 +210,7 @@ struct pipe_stream_output_info
 struct pipe_shader_state
 {
    const struct tgsi_token *tokens;
+   LLVMModuleRef llvm_module;
    struct pipe_stream_output_info stream_output;
 };
 
diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index c030a6b372..6481327409 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -45,6 +45,7 @@
 #include "main/mtypes.h"
 #include "main/shaderobj.h"
 #include "program/hash_table.h"
+#include "st_glsl_to_llvm.h"
 
 extern "C" {
 #include "main/shaderapi.h"
@@ -4997,6 +4998,32 @@ get_mesa_program(struct gl_context *ctx,
    return prog;
 }
 
+extern "C"
+void * st_glsl_to_llvm_visit(glsl_to_tgsi_visitor *glsl_to_tgsi, unsigned shader_type, unsigned clamp_color)
+{
+  const exec_list *ir;
+  if (!glsl_to_tgsi->shader_program)
+    return NULL;
+  switch (shader_type) {
+  case GL_FRAGMENT_SHADER:
+    if (glsl_to_tgsi->shader_program->_LinkedShaders[1])
+      ir = glsl_to_tgsi->shader_program->_LinkedShaders[1]->ir;
+    else
+      return NULL;
+    break;
+  case GL_VERTEX_SHADER:
+    if (glsl_to_tgsi->shader_program->_LinkedShaders[0])
+      ir = glsl_to_tgsi->shader_program->_LinkedShaders[0]->ir;
+    else
+      return NULL;
+    break;
+  default:
+    abort();
+  }
+
+  return glsl_to_llvm_visit(ir, glsl_to_tgsi->shader_program, glsl_to_tgsi->prog, shader_type, clamp_color > 0);
+}
+
 extern "C" {
 
 struct gl_shader *
diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.h b/src/mesa/state_tracker/st_glsl_to_tgsi.h
index a3fe91f7e8..9880460758 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.h
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.h
@@ -34,6 +34,8 @@ struct gl_shader;
 struct gl_shader_program;
 struct glsl_to_tgsi_visitor;
 
+void * st_glsl_to_llvm_visit(struct glsl_to_tgsi_visitor *, unsigned, unsigned);
+
 enum pipe_error st_translate_program(
    struct gl_context *ctx,
    uint procType,
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index a9111b523a..4efc1aa418 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -52,6 +52,7 @@
 #include "st_context.h"
 #include "st_program.h"
 #include "st_mesa_to_tgsi.h"
+#include "st_glsl_to_llvm.h"
 #include "cso_cache/cso_context.h"
 
 
@@ -332,7 +333,7 @@ st_translate_vertex_program(struct st_context *st,
       debug_printf("\n");
    }
 
-   if (stvp->glsl_to_tgsi)
+   if (stvp->glsl_to_tgsi) {
       error = st_translate_program(st->ctx,
                                    TGSI_PROCESSOR_VERTEX,
                                    ureg,
@@ -352,7 +353,9 @@ st_translate_vertex_program(struct st_context *st,
                                    stvp->output_semantic_index,
                                    key->passthrough_edgeflags,
                                    key->clamp_color);
-   else
+      if (_mesa_getenv("GLSL_TO_LLVM"))
+         vpv->tgsi.llvm_module = st_glsl_to_llvm_visit(stvp->glsl_to_tgsi, GL_VERTEX_SHADER, key->clamp_color);
+   } else
       error = st_translate_mesa_program(st->ctx,
                                         TGSI_PROCESSOR_VERTEX,
                                         ureg,
@@ -702,7 +705,7 @@ st_translate_fragment_program(struct st_context *st,
       }
    }
 
-   if (stfp->glsl_to_tgsi)
+   if (stfp->glsl_to_tgsi) {
       st_translate_program(st->ctx,
                            TGSI_PROCESSOR_FRAGMENT,
                            ureg,
@@ -721,7 +724,9 @@ st_translate_fragment_program(struct st_context *st,
                            fs_output_semantic_name,
                            fs_output_semantic_index, FALSE,
                            key->clamp_color );
-   else
+      if (_mesa_getenv("GLSL_TO_LLVM"))
+         variant->tgsi.llvm_module = st_glsl_to_llvm_visit(stfp->glsl_to_tgsi, GL_FRAGMENT_SHADER, key->clamp_color);
+   } else
       st_translate_mesa_program(st->ctx,
                                 TGSI_PROCESSOR_FRAGMENT,
                                 ureg,
diff --git a/src/mesa/state_tracker/st_program.h b/src/mesa/state_tracker/st_program.h
index 23a262cccc..62f73ef514 100644
--- a/src/mesa/state_tracker/st_program.h
+++ b/src/mesa/state_tracker/st_program.h
@@ -39,6 +39,7 @@
 #include "pipe/p_state.h"
 #include "st_context.h"
 #include "st_glsl_to_tgsi.h"
+#include <llvm-c/Core.h>
 
 
 /** Fragment program variant key */
@@ -90,6 +91,7 @@ struct st_fragment_program
 {
    struct gl_fragment_program Base;
    struct glsl_to_tgsi_visitor* glsl_to_tgsi;
+   LLVMModuleRef llvm_module;
 
    struct st_fp_variant *variants;
 };
@@ -145,6 +147,7 @@ struct st_vertex_program
 {
    struct gl_vertex_program Base;  /**< The Mesa vertex program */
    struct glsl_to_tgsi_visitor* glsl_to_tgsi;
+   LLVMModuleRef llvm_module;
 
    /** maps a Mesa VERT_ATTRIB_x to a packed TGSI input index */
    GLuint input_to_index[VERT_ATTRIB_MAX];