nir: Use nir_src_bit_size instead of alu1->dest.dest.ssa.bit_size

This is important because, for example nir_op_fne has
dest.dest.ssa.bit_size == 1, but the source operands can be 16-, 32-, or
64-bits.  Fixing this helps partial redundancy elimination for compares
in a few more shaders.

v2: Add unit tests for nir_opt_comparison_pre that are fixed by this
commit.

All Intel platforms had similar results.
total instructions in shared programs: 17179408 -> 17179081 (<.01%)
instructions in affected programs: 43958 -> 43631 (-0.74%)
helped: 118
HURT: 2
helped stats (abs) min: 1 max: 5 x̄: 2.87 x̃: 2
helped stats (rel) min: 0.06% max: 4.12% x̄: 1.19% x̃: 0.81%
HURT stats (abs)   min: 6 max: 6 x̄: 6.00 x̃: 6
HURT stats (rel)   min: 5.83% max: 6.06% x̄: 5.94% x̃: 5.94%
95% mean confidence interval for instructions value: -3.08 -2.37
95% mean confidence interval for instructions %-change: -1.30% -0.85%
Instructions are helped.

total cycles in shared programs: 360959066 -> 360942386 (<.01%)
cycles in affected programs: 774274 -> 757594 (-2.15%)
helped: 111
HURT: 4
helped stats (abs) min: 1 max: 1591 x̄: 169.49 x̃: 36
helped stats (rel) min: <.01% max: 24.43% x̄: 8.86% x̃: 2.24%
HURT stats (abs)   min: 1 max: 2068 x̄: 533.25 x̃: 32
HURT stats (rel)   min: 0.02% max: 5.10% x̄: 3.06% x̃: 3.56%
95% mean confidence interval for cycles value: -200.61 -89.47
95% mean confidence interval for cycles %-change: -10.32% -6.58%
Cycles are helped.

Reviewed-by: Jason Ekstrand <[email protected]> [v1]
Suggested-by: Jason Ekstrand <[email protected]>
Reviewed-by: Matt Turner <[email protected]>
Fixes: be1cc3552bc ("nir: Add nir_const_value_negative_equal")

1 files changed, 213 insertions, 0 deletions

diff --git a/src/compiler/nir/tests/comparison_pre_tests.cpp b/src/compiler/nir/tests/comparison_pre_tests.cpp
index f31879be6c4..fe1cc23fb3b 100644
--- a/src/compiler/nir/tests/comparison_pre_tests.cpp
+++ b/src/compiler/nir/tests/comparison_pre_tests.cpp
@@ -260,6 +260,219 @@ TEST_F(comparison_pre_test, a_lt_neg_b_vs_a_plus_b)
    EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
 }
 
+TEST_F(comparison_pre_test, imm_lt_b_vs_neg_imm_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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 1 ssa_7 = flt ssa_3, ssa_6
+    *
+    * if ssa_7 {
+    *    vec1 32 ssa_8 = fadd ssa_4, ssa_6
+    * } 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 32 ssa_9 = fneg ssa_3
+    * vec1 32 ssa_10 = fadd ssa_6, 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 *neg_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, one, a);
+
+   nir_if *nif = nir_push_if(&bld, flt);
+
+   nir_fadd(&bld, neg_one, a);
+
+   nir_pop_if(&bld, nif);
+
+   EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
+}
+
+TEST_F(comparison_pre_test, a_lt_imm_vs_a_minus_imm)
+{
+   /* 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 1 ssa_7 = flt ssa_6, ssa_3
+    *
+    * if ssa_6 {
+    *    vec1 32 ssa_8 = fadd ssa_6, ssa_4
+    * } 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 32 ssa_9 = fneg ssa_3
+    * vec1 32 ssa_10 = fadd ssa_6, 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 *neg_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, a, neg_one);
+
+   nir_pop_if(&bld, nif);
+
+   EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
+}
+
+TEST_F(comparison_pre_test, neg_imm_lt_a_vs_a_plus_imm)
+{
+   /* 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 1 ssa_7 = flt ssa_4, ssa_6
+    *
+    * if ssa_7 {
+    *    vec1 32 ssa_8 = fadd ssa_6, 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 32 ssa_9 = fneg ssa_4
+    * vec1 32 ssa_10 = fadd ssa_6, 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 *neg_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, neg_one, a);
+
+   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_imm_vs_a_plus_imm)
+{
+   /* 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 1 ssa_7 = flt ssa_6, ssa_4
+    *
+    * if ssa_7 {
+    *    vec1 32 ssa_8 = fadd ssa_6, 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)
+    * vec1 32 ssa_4 = load_const (-1.0)
+    * vec4 32 ssa_5 = fadd ssa_0, ssa_2
+    * vec1 32 ssa_6 = mov ssa_5.x
+    * vec1 32 ssa_9 = fneg ssa_4
+    * vec1 32 ssa_10 = fadd ssa_6, 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 *neg_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, neg_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, non_scalar_add_result)
 {
    /* The optimization pass should not do anything because the result of the