aco/tests: add setup_reduce_temp.divergent_if_phi

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20621>

src/amd/compiler/aco_builder_h.py

@@ -545,7 +545,7 @@ formats = [("pseudo", [Format.PSEUDO], 'Pseudo_instruction', list(itertools.prod
            ("exp", [Format.EXP], 'Export_instruction', [(0, 4), (0, 5)]),
            ("branch", [Format.PSEUDO_BRANCH], 'Pseudo_branch_instruction', itertools.product([1], [0, 1])),
            ("barrier", [Format.PSEUDO_BARRIER], 'Pseudo_barrier_instruction', [(0, 0)]),
-           ("reduction", [Format.PSEUDO_REDUCTION], 'Pseudo_reduction_instruction', [(3, 2)]),
+           ("reduction", [Format.PSEUDO_REDUCTION], 'Pseudo_reduction_instruction', [(3, 3)]),
            ("vop1", [Format.VOP1], 'VOP1_instruction', [(0, 0), (1, 1), (2, 2)]),
            ("vop1_sdwa", [Format.VOP1, Format.SDWA], 'SDWA_instruction', [(1, 1)]),
            ("vop2", [Format.VOP2], 'VOP2_instruction', itertools.product([1, 2], [2, 3])),

src/amd/compiler/tests/helpers.cpp

@@ -175,6 +175,21 @@ void finish_opt_test()
    aco_print_program(program.get(), output);
 }
 
+void finish_setup_reduce_temp_test()
+{
+   finish_program(program.get());
+   if (!aco::validate_ir(program.get())) {
+      fail_test("Validation before setup_reduce_temp failed");
+      return;
+   }
+   aco::setup_reduce_temp(program.get());
+   if (!aco::validate_ir(program.get())) {
+      fail_test("Validation after setup_reduce_temp failed");
+      return;
+   }
+   aco_print_program(program.get(), output);
+}
+
 void finish_ra_test(ra_test_policy policy, bool lower)
 {
    finish_program(program.get());
@@ -382,7 +397,7 @@ void emit_divergent_if_else(Program* prog, aco::Builder& b, Operand cond, std::f
 
    if_block->kind |= block_kind_branch;
    invert->kind |= block_kind_invert;
-   endif_block->kind |= block_kind_merge;
+   endif_block->kind |= block_kind_merge | (if_block->kind & block_kind_top_level);
 
    /* Set up logical CF */
    then_logical->logical_preds.push_back(if_block->index);

src/amd/compiler/tests/helpers.h

@@ -80,6 +80,7 @@ bool setup_cs(const char *input_spec, enum amd_gfx_level gfx_level,
 void finish_program(aco::Program *program);
 void finish_validator_test();
 void finish_opt_test();
+void finish_setup_reduce_temp_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

@@ -29,6 +29,7 @@ aco_tests_files = files(
   'test_insert_waitcnt.cpp',
   'test_isel.cpp',
   'test_optimizer.cpp',
+  'test_reduce_assign.cpp',
   'test_regalloc.cpp',
   'test_optimizer_postRA.cpp',
   'test_sdwa.cpp',

src/amd/compiler/tests/test_optimizer_postRA.cpp

@@ -514,7 +514,7 @@ BEGIN_TEST(optimizer_postRA.dpp_across_cf)
 
    /* --- merge block --- */
    //! BB6
-   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, merge, */
+   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, top-level, merge, */
    //! s2: %0:exec = p_parallelcopy %saved_exec:s[84-85]
 
    //! v1: %res10:v[12] = v_add_f32 %a:v[0], %b:v[1] row_mirror bound_ctrl:1
@@ -593,7 +593,7 @@ BEGIN_TEST(optimizer_postRA.dpp_across_cf_overwritten)
 
    /* --- merge block --- */
    //! BB6
-   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, merge, */
+   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, top-level, merge, */
    //! s2: %0:exec = p_parallelcopy %saved_exec:s[84-85]
 
    //! v1: %result:v[12] = v_add_f32 %dpp_mov_tmp:v[12], %b:v[1]
@@ -665,7 +665,7 @@ BEGIN_TEST(optimizer_postRA.scc_nocmp_across_cf)
 
    /* --- merge block --- */
    //! BB6
-   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, merge, */
+   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, top-level, merge, */
    //! s2: %0:exec = p_parallelcopy %saved_exec:s[84-85]
 
    //! s2: %tmp_salu:s[8-9], s1: %br_scc:scc = s_and_b64 %a:s[2-3], 0x40018
@@ -748,7 +748,7 @@ BEGIN_TEST(optimizer_postRA.scc_nocmp_across_cf_partially_overwritten)
 
    /* --- merge block --- */
    //! BB6
-   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, merge, */
+   //! /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, top-level, merge, */
    //! s2: %0:exec = p_parallelcopy %saved_exec:s[84-85]
 
    //! s1: %br_scc:scc = s_cmp_lg_u32 %tmp_salu:s[8-9], 0

src/amd/compiler/tests/test_reduce_assign.cpp

@@ -0,0 +1,59 @@
+/*
+ * Copyright © 2022 Valve 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 "helpers.h"
+
+using namespace aco;
+
+BEGIN_TEST(setup_reduce_temp.divergent_if_phi)
+   /*
+    * This must have an end_linear_vgpr after the phi (to ensure it's live during the phi copies):
+    * v0 = start_linear_vgpr
+    * divergent_if (...) {
+    *
+    * } else {
+    *    use_linear_vgpr(v0)
+    * }
+    * ... = phi ...
+   */
+   //TODO: fix the RA validator to spot this
+   //>> s2: %_, v1: %a = p_startpgm
+   if (!setup_cs("s2 v1", GFX9))
+      return;
+
+   //>> lv1: %lv = p_start_linear_vgpr
+   emit_divergent_if_else(program.get(), bld, Operand(inputs[0]), [&]() -> void {
+      //>> s1: %_, s2: %_, s1: %_:scc = p_reduce %a, %lv, lv1: undef op:umin32 cluster_size:64
+      Instruction* reduce = bld.reduction(aco_opcode::p_reduce, bld.def(s1),
+                                          bld.def(bld.lm), bld.def(s1, scc), inputs[1],
+                                          Operand(v1.as_linear()), Operand(v1.as_linear()), umin32);
+      reduce->reduction().cluster_size = bld.lm.bytes() * 8;
+   }, [&]() -> void {
+      /* nothing */
+   });
+   bld.pseudo(aco_opcode::p_phi, bld.def(v1), Operand::c32(1), Operand::zero());
+   //>> /* logical preds: BB1, BB4, / linear preds: BB4, BB5, / kind: uniform, top-level, merge, */
+   //! p_end_linear_vgpr %lv
+
+   finish_setup_reduce_temp_test();
+END_TEST