[Mesa-dev] [PATCH 05/10] glsl: add several EmitNo* options, and MaxUnrollIterations
Luca Barbieri
luca at luca-barbieri.com
Sun Sep 5 18:30:47 PDT 2010
This increases the chance that GLSL programs will actually work.
Note that continues and returns are not yet lowered, so linking
will just fail if not supported.
---
src/glsl/glsl_parser_extras.cpp | 4 ++--
src/glsl/ir_optimization.h | 2 +-
src/glsl/linker.cpp | 2 +-
src/glsl/loop_analysis.h | 2 +-
src/glsl/loop_unroll.cpp | 10 ++++++----
src/glsl/main.cpp | 2 +-
src/mesa/drivers/dri/i965/brw_fs.cpp | 2 +-
src/mesa/main/mtypes.h | 7 ++++++-
src/mesa/main/shaderapi.c | 6 +++++-
src/mesa/program/ir_to_mesa.cpp | 26 ++++++++++++++++++++++----
src/mesa/state_tracker/st_extensions.c | 5 +----
11 files changed, 47 insertions(+), 21 deletions(-)
diff --git a/src/glsl/glsl_parser_extras.cpp b/src/glsl/glsl_parser_extras.cpp
index 390d1f0..51838b0 100644
--- a/src/glsl/glsl_parser_extras.cpp
+++ b/src/glsl/glsl_parser_extras.cpp
@@ -709,7 +709,7 @@ ast_struct_specifier::ast_struct_specifier(char *identifier,
}
bool
-do_common_optimization(exec_list *ir, bool linked)
+do_common_optimization(exec_list *ir, bool linked, unsigned max_unroll_iterations)
{
GLboolean progress = GL_FALSE;
@@ -742,7 +742,7 @@ do_common_optimization(exec_list *ir, bool linked)
loop_state *ls = analyze_loop_variables(ir);
progress = set_loop_controls(ir, ls) || progress;
- progress = unroll_loops(ir, ls) || progress;
+ progress = unroll_loops(ir, ls, max_unroll_iterations) || progress;
delete ls;
return progress;
diff --git a/src/glsl/ir_optimization.h b/src/glsl/ir_optimization.h
index 33f4bc7..df25673 100644
--- a/src/glsl/ir_optimization.h
+++ b/src/glsl/ir_optimization.h
@@ -28,7 +28,7 @@
* Prototypes for optimization passes to be called by the compiler and drivers.
*/
-bool do_common_optimization(exec_list *ir, bool linked);
+bool do_common_optimization(exec_list *ir, bool linked, unsigned max_unroll_iterations);
bool do_algebraic(exec_list *instructions);
bool do_constant_folding(exec_list *instructions);
diff --git a/src/glsl/linker.cpp b/src/glsl/linker.cpp
index e0823c3..b5a9439 100644
--- a/src/glsl/linker.cpp
+++ b/src/glsl/linker.cpp
@@ -1470,7 +1470,7 @@ link_shaders(GLcontext *ctx, struct gl_shader_program *prog)
* some of that unused.
*/
for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
- while (do_common_optimization(prog->_LinkedShaders[i]->ir, true))
+ while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32))
;
}
diff --git a/src/glsl/loop_analysis.h b/src/glsl/loop_analysis.h
index 893dd46..7b0511f 100644
--- a/src/glsl/loop_analysis.h
+++ b/src/glsl/loop_analysis.h
@@ -57,7 +57,7 @@ set_loop_controls(exec_list *instructions, loop_state *ls);
extern bool
-unroll_loops(exec_list *instructions, loop_state *ls);
+unroll_loops(exec_list *instructions, loop_state *ls, unsigned max_iterations);
/**
diff --git a/src/glsl/loop_unroll.cpp b/src/glsl/loop_unroll.cpp
index e204251..80f9217 100644
--- a/src/glsl/loop_unroll.cpp
+++ b/src/glsl/loop_unroll.cpp
@@ -27,10 +27,11 @@
class loop_unroll_visitor : public ir_hierarchical_visitor {
public:
- loop_unroll_visitor(loop_state *state)
+ loop_unroll_visitor(loop_state *state, unsigned max_iterations)
{
this->state = state;
this->progress = false;
+ this->max_iterations = max_iterations;
}
virtual ir_visitor_status visit_leave(ir_loop *ir);
@@ -38,6 +39,7 @@ public:
loop_state *state;
bool progress;
+ unsigned max_iterations;
};
@@ -62,7 +64,7 @@ loop_unroll_visitor::visit_leave(ir_loop *ir)
/* Don't try to unroll loops that have zillions of iterations either.
*/
- if (ls->max_iterations > 32)
+ if (ls->max_iterations > max_iterations)
return visit_continue;
if (ls->num_loop_jumps > 0)
@@ -90,9 +92,9 @@ loop_unroll_visitor::visit_leave(ir_loop *ir)
bool
-unroll_loops(exec_list *instructions, loop_state *ls)
+unroll_loops(exec_list *instructions, loop_state *ls, unsigned max_iterations)
{
- loop_unroll_visitor v(ls);
+ loop_unroll_visitor v(ls, max_iterations);
v.run(instructions);
diff --git a/src/glsl/main.cpp b/src/glsl/main.cpp
index 30f11a5..8e73eb8 100644
--- a/src/glsl/main.cpp
+++ b/src/glsl/main.cpp
@@ -178,7 +178,7 @@ compile_shader(struct gl_shader *shader)
loop_state *ls = analyze_loop_variables(shader->ir);
progress = set_loop_controls(shader->ir, ls) || progress;
- progress = unroll_loops(shader->ir, ls) || progress;
+ progress = unroll_loops(shader->ir, ls, 32) || progress;
delete ls;
} while (progress);
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index 34c5d52..025f7c2 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -141,7 +141,7 @@ brw_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
do {
progress = false;
- progress = do_common_optimization(shader->ir, true) || progress;
+ progress = do_common_optimization(shader->ir, true, 32) || progress;
} while (progress);
validate_ir_tree(shader->ir);
diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
index 657df68..ccf3700 100644
--- a/src/mesa/main/mtypes.h
+++ b/src/mesa/main/mtypes.h
@@ -2187,7 +2187,6 @@ struct gl_shader_compiler_options
{
/** Driver-selectable options: */
GLboolean EmitHighLevelInstructions; /**< IF/ELSE/ENDIF vs. BRA, etc. */
- GLboolean EmitContReturn; /**< Emit CONT/RET opcodes? */
GLboolean EmitCondCodes; /**< Use condition codes? */
GLboolean EmitComments; /**< Annotated instructions */
GLboolean EmitNVTempInitialization; /**< 0-fill NV temp registers */
@@ -2196,6 +2195,12 @@ struct gl_shader_compiler_options
* support control flow.
*/
GLboolean EmitNoIfs;
+ GLboolean EmitNoLoops;
+ GLboolean EmitNoFunctions;
+ GLboolean EmitNoCont; /**< Emit CONT opcode? */
+ GLboolean EmitNoMainReturn; /**< Emit CONT/RET opcodes? */
+
+ GLuint MaxUnrollIterations;
struct gl_sl_pragmas DefaultPragmas; /**< Default #pragma settings */
};
diff --git a/src/mesa/main/shaderapi.c b/src/mesa/main/shaderapi.c
index 2977a29..c32c09f 100644
--- a/src/mesa/main/shaderapi.c
+++ b/src/mesa/main/shaderapi.c
@@ -97,10 +97,14 @@ _mesa_init_shader_state(GLcontext *ctx)
struct gl_shader_compiler_options options;
GLuint i;
options.EmitHighLevelInstructions = GL_TRUE;
- options.EmitContReturn = GL_TRUE;
options.EmitCondCodes = GL_FALSE;
options.EmitComments = GL_FALSE;
options.EmitNoIfs = GL_FALSE;
+ options.EmitNoLoops = GL_FALSE;
+ options.EmitNoFunctions = GL_FALSE;
+ options.EmitNoCont = GL_FALSE;
+ options.EmitNoMainReturn = GL_FALSE;
+ options.MaxUnrollIterations = 32;
/* Default pragma settings */
options.DefaultPragmas.IgnoreOptimize = GL_FALSE;
diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index d73e338..d7de13f 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -2541,6 +2541,7 @@ get_mesa_program(GLcontext *ctx, struct gl_shader_program *shader_program,
GLenum target;
const char *target_string;
GLboolean progress;
+ GLboolean in_sub = GL_FALSE;
struct gl_shader_compiler_options *options =
&ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)];
@@ -2647,16 +2648,32 @@ get_mesa_program(GLcontext *ctx, struct gl_shader_program *shader_program,
if (mesa_inst->SrcReg[src].RelAddr)
prog->IndirectRegisterFiles |= 1 << mesa_inst->SrcReg[src].File;
- if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) {
+ if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF)
fail_link(shader_program, "Couldn't flatten if statement\n");
- }
+
+ if (options->EmitNoLoops && mesa_inst->Opcode == OPCODE_BGNLOOP)
+ fail_link(shader_program, "Couldn't unroll loop\n");
+
+ if (options->EmitNoLoops && mesa_inst->Opcode == OPCODE_CONT)
+ fail_link(shader_program, "Couldn't convert continue to if/else\n");
+
+ if (options->EmitNoFunctions
+ && (mesa_inst->Opcode == OPCODE_BGNSUB || mesa_inst->Opcode == OPCODE_CAL))
+ fail_link(shader_program, "Couldn't inline function\n");
+
+ if (options->EmitNoMainReturn && !in_sub && mesa_inst->Opcode == OPCODE_RET)
+ fail_link(shader_program, "Couldn't convert return in main to if/else\n");
switch (mesa_inst->Opcode) {
case OPCODE_BGNSUB:
+ ASSERT(!in_sub);
+ in_sub = GL_TRUE;
inst->function->inst = i;
mesa_inst->Comment = strdup(inst->function->sig->function_name());
break;
case OPCODE_ENDSUB:
+ ASSERT(in_sub);
+ in_sub = GL_FALSE;
mesa_inst->Comment = strdup(inst->function->sig->function_name());
break;
case OPCODE_CAL:
@@ -2720,6 +2737,7 @@ _mesa_ir_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
bool progress;
+ unsigned max_unroll_iterations = 32;
exec_list *ir = prog->_LinkedShaders[i]->ir;
struct gl_shader_compiler_options *options =
&ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)];
@@ -2733,7 +2751,7 @@ _mesa_ir_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
do_div_to_mul_rcp(ir);
do_explog_to_explog2(ir);
- progress = do_common_optimization(ir, true) || progress;
+ progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
if (options->EmitNoIfs)
progress = do_if_to_cond_assign(ir) || progress;
@@ -2805,7 +2823,7 @@ _mesa_glsl_compile_shader(GLcontext *ctx, struct gl_shader *shader)
/* Do some optimization at compile time to reduce shader IR size
* and reduce later work if the same shader is linked multiple times
*/
- while (do_common_optimization(shader->ir, false))
+ while (do_common_optimization(shader->ir, false, 32))
;
validate_ir_tree(shader->ir);
diff --git a/src/mesa/state_tracker/st_extensions.c b/src/mesa/state_tracker/st_extensions.c
index 82be149..91bd5a9 100644
--- a/src/mesa/state_tracker/st_extensions.c
+++ b/src/mesa/state_tracker/st_extensions.c
@@ -135,11 +135,8 @@ void st_init_limits(struct st_context *st)
= CLAMP(screen->get_param(screen, PIPE_CAP_MAX_RENDER_TARGETS),
1, MAX_DRAW_BUFFERS);
- /* Is TGSI_OPCODE_CONT supported? */
- /* XXX separate query for early function return? */
for(i = 0; i < MESA_SHADER_TYPES; ++i)
- st->ctx->ShaderCompilerOptions[i].EmitContReturn =
- screen->get_param(screen, PIPE_CAP_TGSI_CONT_SUPPORTED);
+ st->ctx->ShaderCompilerOptions[i].EmitNoCont = !screen->get_param(screen, PIPE_CAP_TGSI_CONT_SUPPORTED);
/* Quads always follow GL provoking rules. */
c->QuadsFollowProvokingVertexConvention = GL_FALSE;
--
1.7.0.4
More information about the mesa-dev
mailing list