[Mesa-dev] [PATCH] mesa: Add partial constant propagation pass for Mesa IR
Ian Romanick
idr at freedesktop.org
Mon Aug 15 16:27:49 PDT 2011
From: Ian Romanick <ian.d.romanick at intel.com>
This cleans up some code generated by the IR-to-Mesa pass for i915.
In particular, some shaders involving arrays of constant matrices
result in really bad code.
v2: Silence several warnings from merging the gl_constant_value work.
Fix DP[23] folding. Add support for a bunch more opcodes that appear
in piglit runs on i915.
---
src/mesa/SConscript | 1 +
src/mesa/program/prog_opt_constant_fold.c | 451 +++++++++++++++++++++++++++++
src/mesa/program/prog_optimize.c | 2 +
src/mesa/program/prog_optimize.h | 3 +
src/mesa/sources.mak | 1 +
5 files changed, 458 insertions(+), 0 deletions(-)
create mode 100644 src/mesa/program/prog_opt_constant_fold.c
diff --git a/src/mesa/SConscript b/src/mesa/SConscript
index 05aa0e8..b0c3334 100644
--- a/src/mesa/SConscript
+++ b/src/mesa/SConscript
@@ -293,6 +293,7 @@ program_sources = [
'program/prog_instruction.c',
'program/prog_noise.c',
'program/prog_optimize.c',
+ 'program/prog_opt_constant_fold.c',
'program/prog_parameter.c',
'program/prog_parameter_layout.c',
'program/prog_print.c',
diff --git a/src/mesa/program/prog_opt_constant_fold.c b/src/mesa/program/prog_opt_constant_fold.c
new file mode 100644
index 0000000..e2418b5
--- /dev/null
+++ b/src/mesa/program/prog_opt_constant_fold.c
@@ -0,0 +1,451 @@
+/*
+ * Copyright © 2010 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/macros.h"
+#include "program.h"
+#include "prog_instruction.h"
+#include "prog_optimize.h"
+#include "prog_parameter.h"
+#include <stdbool.h>
+
+static bool
+src_regs_are_constant(const struct prog_instruction *inst, unsigned num_srcs)
+{
+ unsigned i;
+
+ for (i = 0; i < num_srcs; i++) {
+ if (inst->SrcReg[i].File != PROGRAM_CONSTANT)
+ return false;
+ }
+
+ return true;
+}
+
+static struct prog_src_register
+src_reg_for_float(struct gl_program *prog, float val)
+{
+ struct prog_src_register src;
+ unsigned swiz;
+
+ memset(&src, 0, sizeof(src));
+
+ src.File = PROGRAM_CONSTANT;
+ src.Index = _mesa_add_unnamed_constant(prog->Parameters,
+ (gl_constant_value *) &val, 1, &swiz);
+ src.Swizzle = swiz;
+ return src;
+}
+
+static struct prog_src_register
+src_reg_for_vec4(struct gl_program *prog, const float *val)
+{
+ struct prog_src_register src;
+ unsigned swiz;
+
+ memset(&src, 0, sizeof(src));
+
+ src.File = PROGRAM_CONSTANT;
+ src.Index = _mesa_add_unnamed_constant(prog->Parameters,
+ (gl_constant_value *) val, 4, &swiz);
+ src.Swizzle = swiz;
+ return src;
+}
+
+static bool
+src_regs_are_same(const struct prog_src_register *a,
+ const struct prog_src_register *b)
+{
+ return (a->File == b->File)
+ && (a->Index == b->Index)
+ && (a->Swizzle == b->Swizzle)
+ && (a->Abs == b->Abs)
+ && (a->Negate == b->Negate)
+ && (a->RelAddr == 0)
+ && (b->RelAddr == 0);
+}
+
+static void
+get_value(struct gl_program *prog, struct prog_src_register *r, float *data)
+{
+ const gl_constant_value *const value =
+ prog->Parameters->ParameterValues[r->Index];
+
+ data[0] = value[GET_SWZ(r->Swizzle, 0)].f;
+ data[1] = value[GET_SWZ(r->Swizzle, 1)].f;
+ data[2] = value[GET_SWZ(r->Swizzle, 2)].f;
+ data[3] = value[GET_SWZ(r->Swizzle, 3)].f;
+
+ if (r->Abs) {
+ data[0] = fabsf(data[0]);
+ data[1] = fabsf(data[1]);
+ data[2] = fabsf(data[2]);
+ data[3] = fabsf(data[3]);
+ }
+
+ if (r->Negate & 0x01) {
+ data[0] = -data[0];
+ }
+
+ if (r->Negate & 0x02) {
+ data[1] = -data[1];
+ }
+
+ if (r->Negate & 0x04) {
+ data[2] = -data[2];
+ }
+
+ if (r->Negate & 0x08) {
+ data[3] = -data[3];
+ }
+}
+
+/**
+ * Try to replace instructions that produce a constant result with simple moves
+ *
+ * The hope is that a following copy propagation pass will eliminate the
+ * unnecessary move instructions.
+ */
+GLboolean
+_mesa_constant_fold(struct gl_program *prog)
+{
+ bool progress = false;
+ unsigned i;
+
+ for (i = 0; i < prog->NumInstructions; i++) {
+ struct prog_instruction *const inst = &prog->Instructions[i];
+
+ switch (inst->Opcode) {
+ case OPCODE_ADD:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = a[0] + b[0];
+ result[1] = a[1] + b[1];
+ result[2] = a[2] + b[2];
+ result[3] = a[3] + b[3];
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_CMP:
+ /* FINISHME: We could also optimize CMP instructions where the first
+ * FINISHME: source is a constant that is either all < 0.0 or all
+ * FINISHME: >= 0.0.
+ */
+ if (src_regs_are_constant(inst, 3)) {
+ float a[4];
+ float b[4];
+ float c[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+ get_value(prog, &inst->SrcReg[2], c);
+
+ result[0] = a[0] < 0.0f ? b[0] : c[0];
+ result[1] = a[1] < 0.0f ? b[1] : c[1];
+ result[2] = a[2] < 0.0f ? b[2] : c[2];
+ result[3] = a[3] < 0.0f ? b[3] : c[3];
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+ inst->SrcReg[2].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[2].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_DP2:
+ case OPCODE_DP3:
+ case OPCODE_DP4:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result;
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ /* It seems like a loop could be used here, but we cleverly put
+ * DP2A between DP2 and DP3. Subtracting DP2 (or similar) from
+ * the opcode results in various failures of the loop control.
+ */
+ result = (a[0] * b[0]) + (a[1] * b[1]);
+
+ if (inst->Opcode >= OPCODE_DP3)
+ result += a[2] * b[2];
+
+ if (inst->Opcode == OPCODE_DP4)
+ result += a[3] * b[3];
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_MUL:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = a[0] * b[0];
+ result[1] = a[1] * b[1];
+ result[2] = a[2] * b[2];
+ result[3] = a[3] * b[3];
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_SEQ:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = (a[0] == b[0]) ? 1.0f : 0.0f;
+ result[1] = (a[1] == b[1]) ? 1.0f : 0.0f;
+ result[2] = (a[2] == b[2]) ? 1.0f : 0.0f;
+ result[3] = (a[3] == b[3]) ? 1.0f : 0.0f;
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ } else if (src_regs_are_same(&inst->SrcReg[0], &inst->SrcReg[1])) {
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, 1.0f);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_SGE:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = (a[0] >= b[0]) ? 1.0f : 0.0f;
+ result[1] = (a[1] >= b[1]) ? 1.0f : 0.0f;
+ result[2] = (a[2] >= b[2]) ? 1.0f : 0.0f;
+ result[3] = (a[3] >= b[3]) ? 1.0f : 0.0f;
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ } else if (src_regs_are_same(&inst->SrcReg[0], &inst->SrcReg[1])) {
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, 1.0f);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_SGT:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = (a[0] > b[0]) ? 1.0f : 0.0f;
+ result[1] = (a[1] > b[1]) ? 1.0f : 0.0f;
+ result[2] = (a[2] > b[2]) ? 1.0f : 0.0f;
+ result[3] = (a[3] > b[3]) ? 1.0f : 0.0f;
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ } else if (src_regs_are_same(&inst->SrcReg[0], &inst->SrcReg[1])) {
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, 0.0f);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_SLE:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = (a[0] <= b[0]) ? 1.0f : 0.0f;
+ result[1] = (a[1] <= b[1]) ? 1.0f : 0.0f;
+ result[2] = (a[2] <= b[2]) ? 1.0f : 0.0f;
+ result[3] = (a[3] <= b[3]) ? 1.0f : 0.0f;
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ } else if (src_regs_are_same(&inst->SrcReg[0], &inst->SrcReg[1])) {
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, 1.0f);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_SLT:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = (a[0] < b[0]) ? 1.0f : 0.0f;
+ result[1] = (a[1] < b[1]) ? 1.0f : 0.0f;
+ result[2] = (a[2] < b[2]) ? 1.0f : 0.0f;
+ result[3] = (a[3] < b[3]) ? 1.0f : 0.0f;
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ } else if (src_regs_are_same(&inst->SrcReg[0], &inst->SrcReg[1])) {
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, 0.0f);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ case OPCODE_SNE:
+ if (src_regs_are_constant(inst, 2)) {
+ float a[4];
+ float b[4];
+ float result[4];
+
+ get_value(prog, &inst->SrcReg[0], a);
+ get_value(prog, &inst->SrcReg[1], b);
+
+ result[0] = (a[0] != b[0]) ? 1.0f : 0.0f;
+ result[1] = (a[1] != b[1]) ? 1.0f : 0.0f;
+ result[2] = (a[2] != b[2]) ? 1.0f : 0.0f;
+ result[3] = (a[3] != b[3]) ? 1.0f : 0.0f;
+
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_vec4(prog, result);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ } else if (src_regs_are_same(&inst->SrcReg[0], &inst->SrcReg[1])) {
+ inst->Opcode = OPCODE_MOV;
+ inst->SrcReg[0] = src_reg_for_float(prog, 0.0f);
+
+ inst->SrcReg[1].File = PROGRAM_UNDEFINED;
+ inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ progress = true;
+ }
+ break;
+
+ default:
+ break;
+ }
+ }
+
+ return progress;
+}
diff --git a/src/mesa/program/prog_optimize.c b/src/mesa/program/prog_optimize.c
index 3340ce0..25d9684 100644
--- a/src/mesa/program/prog_optimize.c
+++ b/src/mesa/program/prog_optimize.c
@@ -1358,6 +1358,8 @@ _mesa_optimize_program(struct gl_context *ctx, struct gl_program *program)
any_change = GL_TRUE;
if (_mesa_remove_dead_code_local(program))
any_change = GL_TRUE;
+
+ any_change = _mesa_constant_fold(program) || any_change;
_mesa_reallocate_registers(program);
} while (any_change);
}
diff --git a/src/mesa/program/prog_optimize.h b/src/mesa/program/prog_optimize.h
index 463f5fc..9854fb7 100644
--- a/src/mesa/program/prog_optimize.h
+++ b/src/mesa/program/prog_optimize.h
@@ -44,4 +44,7 @@ _mesa_find_temp_intervals(const struct prog_instruction *instructions,
extern void
_mesa_optimize_program(struct gl_context *ctx, struct gl_program *program);
+extern GLboolean
+_mesa_constant_fold(struct gl_program *prog);
+
#endif
diff --git a/src/mesa/sources.mak b/src/mesa/sources.mak
index ed008f8..5e77e0f 100644
--- a/src/mesa/sources.mak
+++ b/src/mesa/sources.mak
@@ -251,6 +251,7 @@ PROGRAM_SOURCES = \
program/prog_instruction.c \
program/prog_noise.c \
program/prog_optimize.c \
+ program/prog_opt_constant_fold.c \
program/prog_parameter.c \
program/prog_parameter_layout.c \
program/prog_print.c \
--
1.7.4.4
More information about the mesa-dev
mailing list