From 61fa007e4842f8e0e7c777d4fdb23b3c7118d48f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Daniel=20Sch=C3=BCrmann?= <daniel@schuermann.dev>
Date: Fri, 24 Jan 2025 10:05:34 +0100
Subject: [PATCH] aco/isel: fix empty exec tracking for uniform branches

Totals from 5 (0.01% of 79395) affected shaders: (Navi31)

Instrs: 54730 -> 54715 (-0.03%)
CodeSize: 276928 -> 276852 (-0.03%)
Latency: 215212 -> 214874 (-0.16%)
InvThroughput: 40154 -> 40150 (-0.01%)
Copies: 6824 -> 6821 (-0.04%); split: -0.06%, +0.01%
Branches: 1625 -> 1615 (-0.62%)
SALU: 5682 -> 5678 (-0.07%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33206>
---
 .../compiler/aco_instruction_selection.cpp    | 12 ++--
 src/amd/compiler/tests/test_isel.cpp          | 55 +++++++++++++++++++
 2 files changed, 60 insertions(+), 7 deletions(-)

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 285c3f0c5e6..120675adea3 100644
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -10186,12 +10186,11 @@ begin_uniform_if_then(isel_context* ctx, if_context* ic, Temp cond)
    ic->BB_if_idx = ctx->block->index;
    ic->BB_endif = Block();
    ic->BB_endif.kind |= ctx->block->kind & block_kind_top_level;
-
-   ctx->cf_info.has_branch = false;
-   ctx->cf_info.parent_loop.has_divergent_branch = false;
+   assert(!ctx->cf_info.has_branch && !ctx->cf_info.parent_loop.has_divergent_branch);
 
    ic->had_divergent_discard_old = ctx->cf_info.had_divergent_discard;
    ic->has_divergent_continue_old = ctx->cf_info.parent_loop.has_divergent_continue;
+   ic->exec_old = ctx->cf_info.exec;
 
    /** emit then block */
    if (ic->cond.id())
@@ -10228,6 +10227,8 @@ begin_uniform_if_else(isel_context* ctx, if_context* ic, bool logical_else)
    ic->has_divergent_continue_then = ctx->cf_info.parent_loop.has_divergent_continue;
    ctx->cf_info.parent_loop.has_divergent_continue = ic->has_divergent_continue_old;
 
+   std::swap(ctx->cf_info.exec, ic->exec_old);
+
    /** emit else block */
    Block* BB_else = ctx->program->create_and_insert_block();
    if (logical_else) {
@@ -10261,6 +10262,7 @@ end_uniform_if(isel_context* ctx, if_context* ic, bool logical_else)
    ctx->cf_info.parent_loop.has_divergent_branch = false;
    ctx->cf_info.had_divergent_discard |= ic->had_divergent_discard_then;
    ctx->cf_info.parent_loop.has_divergent_continue |= ic->has_divergent_continue_then;
+   ctx->cf_info.exec.combine(ic->exec_old);
 
    /** emit endif merge block */
    if (ic->cond.id())
@@ -10279,8 +10281,6 @@ end_empty_exec_skip(isel_context* ctx)
       begin_uniform_if_else(ctx, &ctx->cf_info.empty_exec_skip, false);
       end_uniform_if(ctx, &ctx->cf_info.empty_exec_skip, false);
       ctx->cf_info.skipping_empty_exec = false;
-
-      ctx->cf_info.exec.combine(ctx->cf_info.empty_exec_skip.exec_old);
    }
 }
 
@@ -10333,8 +10333,6 @@ begin_empty_exec_skip(isel_context* ctx, nir_instr* after_instr, nir_block* bloc
 
    begin_uniform_if_then(ctx, &ctx->cf_info.empty_exec_skip, Temp());
    ctx->cf_info.skipping_empty_exec = true;
-
-   ctx->cf_info.empty_exec_skip.exec_old = ctx->cf_info.exec;
    ctx->cf_info.exec = exec_info();
 
    ctx->program->should_repair_ssa = true;
diff --git a/src/amd/compiler/tests/test_isel.cpp b/src/amd/compiler/tests/test_isel.cpp
index c6514a01fee..7322f253df8 100644
--- a/src/amd/compiler/tests/test_isel.cpp
+++ b/src/amd/compiler/tests/test_isel.cpp
@@ -1185,6 +1185,61 @@ BEGIN_TEST(isel.cf.empty_exec.uniform_if)
    finish_isel_test();
 END_TEST
 
+/*
+ * if (divergent) {
+ *    if (uniform) {
+ *       terminate_if
+ *       // exec potentially empty
+ *    } else {
+ *    }
+ *    // exec potentially empty
+ * }
+ */
+BEGIN_TEST(isel.cf.empty_exec.nested_uniform_if)
+   if (!setup_nir_cs(GFX11))
+      return;
+
+   //>> BB0
+   //>> s2: %_ = p_unit_test 0
+   //>> p_cbranch_z %_
+   nir_push_if(nb, nir_unit_test_divergent_amd(nb, 1, 1, .base = 0));
+   {
+      //>> BB1
+      //>> s2: %_ = p_unit_test 1
+      //>> p_cbranch_z %_:scc
+      nir_push_if(nb, nir_unit_test_uniform_amd(nb, 1, 1, .base = 1));
+      {
+         //>> BB2
+         //>> s2: %_ = p_unit_test 2
+         //>> p_discard_if %_
+         nir_terminate_if(nb, nir_unit_test_divergent_amd(nb, 1, 1, .base = 2));
+
+         //>> p_cbranch_z %0:exec rarely_taken
+         //>> p_unit_test 3, %_
+         nir_unit_test_amd(nb, nir_undef(nb, 1, 32), .base = 3);
+      }
+      nir_push_else(nb, NULL);
+      {
+         //>> BB6
+         //>> p_unit_test 4, %_
+         nir_unit_test_amd(nb, nir_undef(nb, 1, 32), .base = 4);
+      }
+      nir_pop_if(nb, NULL);
+
+      //>> BB7
+      //>> p_cbranch_z %0:exec rarely_taken
+      //>> p_unit_test 5, %_
+      nir_unit_test_amd(nb, nir_undef(nb, 1, 32), .base = 5);
+   }
+   nir_pop_if(nb, NULL);
+   //>> BB15
+   //! /* logical preds: BB10, BB13, / linear preds: BB13, BB14, / kind: uniform, top-level, merge, */
+   //>> p_unit_test 6, %_
+   nir_unit_test_amd(nb, nir_undef(nb, 1, 32), .base = 6);
+
+   finish_isel_test();
+END_TEST
+
 /*
  * if (divergent) {
  *   terminate_if