aboutsummaryrefslogtreecommitdiffstats
path: root/src/compiler
diff options
context:
space:
mode:
Diffstat (limited to 'src/compiler')
-rw-r--r--src/compiler/nir/meson.build12
-rw-r--r--src/compiler/nir/nir.h3
-rw-r--r--src/compiler/nir/nir_opt_comparison_pre.c2
-rw-r--r--src/compiler/nir/tests/comparison_pre_tests.cpp318
4 files changed, 334 insertions, 1 deletions
diff --git a/src/compiler/nir/meson.build b/src/compiler/nir/meson.build
index 01ddcdf0ea0..169c31aacd5 100644
--- a/src/compiler/nir/meson.build
+++ b/src/compiler/nir/meson.build
@@ -306,4 +306,16 @@ if with_tests
link_with : libmesa_util,
)
)
+
+ test(
+ 'comparison_pre',
+ executable(
+ 'comparison_pre',
+ files('tests/comparison_pre_tests.cpp'),
+ c_args : [c_vis_args, c_msvc_compat_args, no_override_init_args],
+ include_directories : [inc_common],
+ dependencies : [dep_thread, idep_gtest, idep_nir],
+ link_with : libmesa_util,
+ )
+ )
endif
diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 55a39a493e9..92dad1c4bb3 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -3625,6 +3625,9 @@ bool nir_lower_phis_to_regs_block(nir_block *block);
bool nir_lower_ssa_defs_to_regs_block(nir_block *block);
bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl *impl);
+/* This is here for unit tests. */
+bool nir_opt_comparison_pre_impl(nir_function_impl *impl);
+
bool nir_opt_comparison_pre(nir_shader *shader);
bool nir_opt_algebraic(nir_shader *shader);
diff --git a/src/compiler/nir/nir_opt_comparison_pre.c b/src/compiler/nir/nir_opt_comparison_pre.c
index 221379b3a23..33295e5eea6 100644
--- a/src/compiler/nir/nir_opt_comparison_pre.c
+++ b/src/compiler/nir/nir_opt_comparison_pre.c
@@ -346,7 +346,7 @@ comparison_pre_block(nir_block *block, struct block_queue *bq, nir_builder *bld)
return progress;
}
-static bool
+bool
nir_opt_comparison_pre_impl(nir_function_impl *impl)
{
struct block_queue bq;
diff --git a/src/compiler/nir/tests/comparison_pre_tests.cpp b/src/compiler/nir/tests/comparison_pre_tests.cpp
new file mode 100644
index 00000000000..f31879be6c4
--- /dev/null
+++ b/src/compiler/nir/tests/comparison_pre_tests.cpp
@@ -0,0 +1,318 @@
+/*
+ * Copyright © 2019 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 <gtest/gtest.h>
+#include "nir.h"
+#include "nir_builder.h"
+
+class comparison_pre_test : public ::testing::Test {
+protected:
+ comparison_pre_test()
+ {
+ static const nir_shader_compiler_options options = { };
+ nir_builder_init_simple_shader(&bld, NULL, MESA_SHADER_VERTEX, &options);
+
+ v1 = nir_imm_vec4(&bld, -2.0, -1.0, 1.0, 2.0);
+ v2 = nir_imm_vec4(&bld, 2.0, 1.0, -1.0, -2.0);
+ v3 = nir_imm_vec4(&bld, 3.0, 4.0, 5.0, 6.0);
+ }
+
+ ~comparison_pre_test()
+ {
+ ralloc_free(bld.shader);
+ }
+
+ struct nir_builder bld;
+
+ nir_ssa_def *v1;
+ nir_ssa_def *v2;
+ nir_ssa_def *v3;
+
+ const uint8_t xxxx[4] = { 0, 0, 0, 0 };
+ const uint8_t wwww[4] = { 3, 3, 3, 3 };
+};
+
+TEST_F(comparison_pre_test, a_lt_b_vs_neg_a_plus_b)
+{
+ /* Before:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 1 ssa_6 = flt ssa_5, ssa_3
+ *
+ * if ssa_6 {
+ * vec1 32 ssa_7 = fneg ssa_5
+ * vec1 32 ssa_8 = fadd ssa_7, ssa_3
+ * } else {
+ * }
+ *
+ * After:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 32 ssa_9 = fneg ssa_5
+ * vec1 32 ssa_10 = fadd ssa_3, ssa_9
+ * vec1 32 ssa_11 = load_const (0.0)
+ * vec1 1 ssa_12 = flt ssa_11, ssa_10
+ * vec1 32 ssa_13 = mov ssa_10
+ * vec1 1 ssa_14 = mov ssa_12
+ *
+ * if ssa_14 {
+ * vec1 32 ssa_7 = fneg ssa_5
+ * } else {
+ * }
+ */
+ nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
+ nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
+
+ nir_ssa_def *flt = nir_flt(&bld, a, one);
+
+ nir_if *nif = nir_push_if(&bld, flt);
+
+ nir_fadd(&bld, nir_fneg(&bld, a), one);
+
+ nir_pop_if(&bld, nif);
+
+ EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
+}
+
+TEST_F(comparison_pre_test, a_lt_b_vs_a_minus_b)
+{
+ /* Before:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 1 ssa_6 = flt ssa_3, ssa_5
+ *
+ * if ssa_6 {
+ * vec1 32 ssa_7 = fneg ssa_5
+ * vec1 32 ssa_8 = fadd ssa_3, ssa_7
+ * } else {
+ * }
+ *
+ * After:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 32 ssa_9 = fneg ssa_5
+ * vec1 32 ssa_10 = fadd ssa_3, ssa_9
+ * vec1 32 ssa_11 = load_const (0.0)
+ * vec1 1 ssa_12 = flt ssa_10, ssa_11
+ * vec1 32 ssa_13 = mov ssa_10
+ * vec1 1 ssa_14 = mov ssa_12
+ *
+ * if ssa_14 {
+ * vec1 32 ssa_7 = fneg ssa_5
+ * } else {
+ * }
+ */
+ nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
+ nir_ssa_def *b = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
+
+ nir_ssa_def *flt = nir_flt(&bld, one, b);
+
+ nir_if *nif = nir_push_if(&bld, flt);
+
+ nir_fadd(&bld, one, nir_fneg(&bld, b));
+
+ nir_pop_if(&bld, nif);
+
+ EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
+}
+
+TEST_F(comparison_pre_test, neg_a_lt_b_vs_a_plus_b)
+{
+ /* Before:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 32 ssa_6 = fneg ssa_5
+ * vec1 1 ssa_7 = flt ssa_6, ssa_3
+ *
+ * if ssa_7 {
+ * vec1 32 ssa_8 = fadd ssa_5, ssa_3
+ * } else {
+ * }
+ *
+ * After:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 32 ssa_9 = fneg ssa_5
+ * vec1 32 ssa_9 = fneg ssa_6
+ * vec1 32 ssa_10 = fadd ssa_3, ssa_9
+ * vec1 32 ssa_11 = load_const ( 0.0)
+ * vec1 1 ssa_12 = flt ssa_11, ssa_10
+ * vec1 32 ssa_13 = mov ssa_10
+ * vec1 1 ssa_14 = mov ssa_12
+ *
+ * if ssa_14 {
+ * } else {
+ * }
+ */
+
+ nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
+ nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
+
+ nir_ssa_def *flt = nir_flt(&bld, nir_fneg(&bld, a), one);
+
+ nir_if *nif = nir_push_if(&bld, flt);
+
+ nir_fadd(&bld, a, one);
+
+ nir_pop_if(&bld, nif);
+
+ EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
+}
+
+TEST_F(comparison_pre_test, a_lt_neg_b_vs_a_plus_b)
+{
+ /* Before:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 32 ssa_6 = fneg ssa_5
+ * vec1 1 ssa_7 = flt ssa_3, ssa_6
+ *
+ * if ssa_7 {
+ * vec1 32 ssa_8 = fadd ssa_3, ssa_5
+ * } else {
+ * }
+ *
+ * After:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec1 32 ssa_3 = load_const ( 1.0)
+ * vec4 32 ssa_4 = fadd ssa_0, ssa_2
+ * vec1 32 ssa_5 = mov ssa_4.x
+ * vec1 32 ssa_9 = fneg ssa_5
+ * vec1 32 ssa_9 = fneg ssa_6
+ * vec1 32 ssa_10 = fadd ssa_3, ssa_9
+ * vec1 32 ssa_11 = load_const ( 0.0)
+ * vec1 1 ssa_12 = flt ssa_10, ssa_11
+ * vec1 32 ssa_13 = mov ssa_10
+ * vec1 1 ssa_14 = mov ssa_12
+ *
+ * if ssa_14 {
+ * } else {
+ * }
+ */
+ nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
+ nir_ssa_def *b = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
+
+ nir_ssa_def *flt = nir_flt(&bld, one, nir_fneg(&bld, b));
+
+ nir_if *nif = nir_push_if(&bld, flt);
+
+ nir_fadd(&bld, one, b);
+
+ nir_pop_if(&bld, nif);
+
+ EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
+}
+
+TEST_F(comparison_pre_test, non_scalar_add_result)
+{
+ /* The optimization pass should not do anything because the result of the
+ * fadd is not a scalar.
+ *
+ * Before:
+ *
+ * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
+ * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
+ * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
+ * vec4 32 ssa_3 = fadd ssa_0, ssa_2
+ * vec1 1 ssa_4 = flt ssa_0.x, ssa_3.x
+ *
+ * if ssa_4 {
+ * vec2 32 ssa_5 = fadd ssa_1.xx, ssa_3.xx
+ * } else {
+ * }
+ *
+ * After:
+ *
+ * No change.
+ */
+ nir_ssa_def *a = nir_fadd(&bld, v1, v3);
+
+ nir_alu_instr *flt = nir_alu_instr_create(bld.shader, nir_op_flt);
+
+ flt->src[0].src = nir_src_for_ssa(v1);
+ flt->src[1].src = nir_src_for_ssa(a);
+
+ memcpy(&flt->src[0].swizzle, xxxx, sizeof(xxxx));
+ memcpy(&flt->src[1].swizzle, xxxx, sizeof(xxxx));
+
+ nir_builder_alu_instr_finish_and_insert(&bld, flt);
+
+ flt->dest.dest.ssa.num_components = 1;
+ flt->dest.write_mask = 1;
+
+ nir_if *nif = nir_push_if(&bld, &flt->dest.dest.ssa);
+
+ nir_alu_instr *fadd = nir_alu_instr_create(bld.shader, nir_op_fadd);
+
+ fadd->src[0].src = nir_src_for_ssa(v2);
+ fadd->src[1].src = nir_src_for_ssa(a);
+
+ memcpy(&fadd->src[0].swizzle, xxxx, sizeof(xxxx));
+ memcpy(&fadd->src[1].swizzle, xxxx, sizeof(xxxx));
+
+ nir_builder_alu_instr_finish_and_insert(&bld, fadd);
+
+ fadd->dest.dest.ssa.num_components = 2;
+ fadd->dest.write_mask = 3;
+
+ nir_pop_if(&bld, nif);
+
+ EXPECT_FALSE(nir_opt_comparison_pre_impl(bld.impl));
+}
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-29 01:08:25 +0000