aco: add tests for lower_subdword

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28836>

src/amd/compiler/tests/helpers.cpp

@@ -221,6 +221,22 @@ finish_setup_reduce_temp_test()
    aco_print_program(program.get(), output);
 }
 
+void
+finish_lower_subdword_test()
+{
+   finish_program(program.get());
+   if (!aco::validate_ir(program.get())) {
+      fail_test("Validation before lower_subdword failed");
+      return;
+   }
+   aco::lower_subdword(program.get());
+   if (!aco::validate_ir(program.get())) {
+      fail_test("Validation after lower_subdword failed");
+      return;
+   }
+   aco_print_program(program.get(), output);
+}
+
 void
 finish_ra_test(ra_test_policy policy, bool lower)
 {

src/amd/compiler/tests/helpers.h

@@ -70,6 +70,7 @@ void finish_program(aco::Program* program, bool endpgm = true);
 void finish_validator_test();
 void finish_opt_test();
 void finish_setup_reduce_temp_test();
+void finish_lower_subdword_test();
 void finish_ra_test(aco::ra_test_policy, bool lower = false);
 void finish_optimizer_postRA_test();
 void finish_to_hw_instr_test();

src/amd/compiler/tests/meson.build

@@ -14,6 +14,7 @@ aco_tests_files = files(
   'test_insert_nops.cpp',
   'test_insert_waitcnt.cpp',
   'test_isel.cpp',
+  'test_lower_subdword.cpp',
   'test_optimizer.cpp',
   'test_reduce_assign.cpp',
   'test_regalloc.cpp',

src/amd/compiler/tests/test_lower_subdword.cpp

@@ -0,0 +1,142 @@
+/*
+ * Copyright © 2024 Valve Corporation
+ *
+ * SPDX-License-Identifier: MIT
+ */
+#include "helpers.h"
+#include <stdarg.h>
+
+using namespace aco;
+
+BEGIN_TEST(lower_subdword.simple_extract)
+   //>> v1: %a, v2: %b = p_startpgm
+   if (!setup_cs("v1 v2", GFX6))
+      return;
+
+   //! v1: %aw0 = p_parallelcopy %a
+   bld.pseudo(aco_opcode::p_extract_vector, bld.def(v2b), inputs[0], Operand::c32(0));
+
+   //! v1: %aw1 = v_lshrrev_b32 16, %a
+   bld.pseudo(aco_opcode::p_extract_vector, bld.def(v2b), inputs[0], Operand::c32(1));
+
+   //! v1: %ab3 = v_lshrrev_b32 24, %a
+   bld.pseudo(aco_opcode::p_extract_vector, bld.def(v1b), inputs[0], Operand::c32(3));
+
+   //! v1: %_,  v1: %bd1 = p_split_vector %b
+   //! v1: %6 = v_lshrrev_b32 16, %bd1
+   bld.pseudo(aco_opcode::p_extract_vector, bld.def(v2b), inputs[1], Operand::c32(3));
+
+   //! v1: %bd0,  v1: %_ = p_split_vector %b
+   //! v1: %bb2 = v_lshrrev_b32 16, %bd0
+   bld.pseudo(aco_opcode::p_extract_vector, bld.def(v1b), inputs[1], Operand::c32(2));
+
+   //! v1: %bd0_2,  v1: %bd1_2 = p_split_vector %b
+   //! v1: %bv3bhi = v_alignbyte_b32 %bd1_2, %bd0_2, 3
+   bld.pseudo(aco_opcode::p_extract_vector, bld.def(v3b), inputs[1], Operand::c32(1));
+
+   finish_lower_subdword_test();
+END_TEST
+
+BEGIN_TEST(lower_subdword.simple_split)
+   //>> v1: %a, v2: %b = p_startpgm
+   if (!setup_cs("v1 v2", GFX6))
+      return;
+
+   //! v1: %o1_0 = p_parallelcopy %a
+   //! v1: %o1_1 = v_lshrrev_b32 8, %a
+   //! v1: %o1_2 = v_lshrrev_b32 24, %a
+   bld.pseudo(aco_opcode::p_split_vector, bld.def(v1b), bld.def(v2b), bld.def(v1b), inputs[0]);
+
+   //! v1: %tmp2_0,  v1: %_ = p_split_vector %b
+   //! v1: %o2_0 = p_parallelcopy %tmp2_0
+   //! v1: %tmp2_1,  v1: %_ = p_split_vector %b
+   //! v1: %o2_1 = v_lshrrev_b32 8, %tmp2_1
+   //! v1: %tmp2_2,  v1: %_ = p_split_vector %b
+   //! v1: %o2_2 = v_lshrrev_b32 16, %tmp2_2
+   //! v1: %tmp2_3,  v1: %_ = p_split_vector %b
+   //! v1: %o2_3 = v_lshrrev_b32 24, %tmp2_3
+   //! v1: %_,  v1: %tmp2_4 = p_split_vector %b
+   //! v1: %o2_4 = p_parallelcopy %tmp2_4
+   //! v1: %_,  v1: %tmp2_5 = p_split_vector %b
+   //! v1: %o2_5 = v_lshrrev_b32 8, %tmp2_5
+   //! v1: %_,  v1: %tmp2_6 = p_split_vector %b
+   //! v1: %o2_6 = v_lshrrev_b32 16, %tmp2_6
+   //! v1: %_,  v1: %tmp2_7 = p_split_vector %b
+   //! v1: %o2_7 = v_lshrrev_b32 24, %tmp2_7
+   bld.pseudo(aco_opcode::p_split_vector, bld.def(v1b), bld.def(v1b), bld.def(v1b), bld.def(v1b),
+              bld.def(v1b), bld.def(v1b), bld.def(v1b), bld.def(v1b), inputs[1]);
+
+   finish_lower_subdword_test();
+END_TEST
+
+BEGIN_TEST(lower_subdword.simple_create)
+   //>> v1: %b,  v1: %w,  v1: %tb,  s1: %sd,  v1: %vd = p_startpgm
+   if (!setup_cs("v1b v2b v3b s1 v1 ", GFX6))
+      return;
+
+   //! v1: %tmp1_0 = v_lshlrev_b32 24, %b
+   //! v1: %tmp1_1 = v_alignbyte_b32 %b, %tmp1_0, 1
+   //! v1: %tmp1_2 = v_alignbyte_b32 %b, %tmp1_1, 1
+   //! v1: %o1 = v_alignbyte_b32 %b, %tmp1_2, 1
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v1), inputs[0], inputs[0], inputs[0], inputs[0]);
+
+   //! v1: %tmp2_0 = v_lshlrev_b32 24, %b
+   //! v1: %tmp2_1 = v_alignbyte_b32 %w, %tmp2_0, 2
+   //! v1: %o2 = v_alignbyte_b32 %b, %tmp2_1, 1
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v1), inputs[0], inputs[1], inputs[0]);
+
+   //! v1: %tmp3_0 = v_lshlrev_b32 24, %b
+   //! v1: %tmp3_1 = v_alignbyte_b32 %tb, %tmp3_0, 3
+   //! v1: %tmp3_2 = v_lshlrev_b32 16, %w
+   //! v1: %tmp3_3 = v_alignbyte_b32 %b, %tmp3_2, 1
+   //! v1: %tmp3_4 = v_alignbyte_b32 %b, %tmp3_3, 1
+   //! v2: %o3 = p_create_vector %tmp3_1, %tmp3_4
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v2), inputs[0], inputs[2], inputs[1], inputs[0],
+              inputs[0]);
+
+   //! v1: %tmp4_0 = p_parallelcopy %sd
+   //! v1: %tmp4_1 = p_parallelcopy %vd
+   //! v1: %tmp4_2 = v_lshlrev_b32 16, %w
+   //! v1: %tmp4_3 = v_alignbyte_b32 %tb, %tmp4_2, 2
+   //! v1: %tmp4_4 = v_lshlrev_b32 8, %tb
+   //! v1: %tmp4_5 = v_alignbyte_b32 %tb, %tmp4_4, 3
+   //! v4: %o4 = p_create_vector %tmp4_0, %tmp4_1, %tmp4_3, %tmp4_5
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v4), inputs[3], inputs[4], inputs[1], inputs[2],
+              inputs[2]);
+
+   //! v1: %tmp5_0 = v_lshlrev_b32 24, %b
+   //! v1: %o5 = v_alignbyte_b32 %b, %tmp5_0, 3
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v2b), inputs[0], inputs[0]);
+
+   finish_lower_subdword_test();
+END_TEST
+
+BEGIN_TEST(lower_subdword.create_const_undef)
+   //>> v1: %b,  v1: %w = p_startpgm
+   if (!setup_cs("v1b v2b", GFX6))
+      return;
+
+   //! v1: %o1 = p_parallelcopy %b
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v1), inputs[0], Operand(v1b), Operand(v2b));
+
+   //! v1: %o2 = p_parallelcopy 0x40300
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v3b), Operand::c8(0), Operand::c8(3),
+              Operand::c8(4));
+
+   //! s1: %tmp1 = p_parallelcopy 0x64000000
+   //! v1: %o3 = v_alignbyte_b32 %w, %tmp1, 3
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v1), Operand::c8(100), inputs[1], Operand(v1b));
+
+   //! s1: %tmp2 = p_parallelcopy 0x64000000
+   //! v1: %tmp3 = v_alignbyte_b32 %b, %tmp2, 2
+   //! v1: %tmp4 = v_alignbyte_b32 %b, %tmp3, 1
+   //! v1: %tmp5 = p_parallelcopy %w
+   //! v2: %o4 = p_create_vector %tmp4, %tmp5
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v6b), Operand::c8(100), inputs[0], Operand(v1b),
+              inputs[0], inputs[1]);
+
+   //! v1: %o5 = v_alignbyte_b32 %b, 0, 3
+   bld.pseudo(aco_opcode::p_create_vector, bld.def(v2b), Operand(v1b), inputs[0]);
+
+   finish_lower_subdword_test();
+END_TEST