Mesa (staging/21.3): aco: consider pseudo-instructions reading exec in needs_exec_mask()

[GitLab Mirror]

Module: Mesa
Branch: staging/21.3
Commit: dc7bccdb545c60a11743fb37f4671d67d02c7031
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=dc7bccdb545c60a11743fb37f4671d67d02c7031

Author: Rhys Perry <pendingchaos02 at gmail.com>
Date:   Fri Nov 12 13:46:17 2021 +0000

aco: consider pseudo-instructions reading exec in needs_exec_mask()

No matter the format, this should return true if the instruction has an
exec operand.

Otherwise, eliminate_useless_exec_writes_in_block() could remove an exec
write in a block if it's successor begins with:
 s2: %3737:s[8-9] = p_parallelcopy %0:exec
 s2: %0:exec,  s1: %3738:scc = s_wqm_b64 %3737:s[8-9]

Totals from 3 (0.00% of 150170) affected shaders (GFX10.3):
CodeSize: 23184 -> 23204 (+0.09%)
Instrs: 4143 -> 4148 (+0.12%)
Latency: 98379 -> 98382 (+0.00%)
Copies: 172 -> 175 (+1.74%)
Branches: 95 -> 97 (+2.11%)

Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
Reviewed-by: Daniel Schürmann <daniel at schuermann.dev>
Fixes: bc130497472 ("aco: Eliminate useless exec writes in jump threading.")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5620
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13776>
(cherry picked from commit d89461208b79a0c5f577caf6a98dab6976c12251)

---

 .pick_status.json           |  2 +-
 src/amd/compiler/aco_ir.cpp | 26 +++++++++++++-------------
 2 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/.pick_status.json b/.pick_status.json
index bb32bf5abe8..10e43531f20 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -967,7 +967,7 @@
         "description": "aco: consider pseudo-instructions reading exec in needs_exec_mask()",
         "nominated": true,
         "nomination_type": 1,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": "bc130497472cb4ec4ec60695ed99b169d6681118"
     },
diff --git a/src/amd/compiler/aco_ir.cpp b/src/amd/compiler/aco_ir.cpp
index dcc85a92e2f..5923fea9589 100644
--- a/src/amd/compiler/aco_ir.cpp
+++ b/src/amd/compiler/aco_ir.cpp
@@ -525,12 +525,18 @@ get_reduction_identity(ReduceOp op, unsigned idx)
 bool
 needs_exec_mask(const Instruction* instr)
 {
-   if (instr->isSALU() || instr->isBranch())
+   if (instr->isVALU()) {
+      return instr->opcode != aco_opcode::v_readlane_b32 &&
+             instr->opcode != aco_opcode::v_readlane_b32_e64 &&
+             instr->opcode != aco_opcode::v_writelane_b32 &&
+             instr->opcode != aco_opcode::v_writelane_b32_e64;
+   }
+
+   if (instr->isVMEM() || instr->isFlatLike())
+      return true;
+
+   if (instr->isSALU() || instr->isBranch() || instr->isSMEM() || instr->isBarrier())
       return instr->reads_exec();
-   if (instr->isSMEM())
-      return false;
-   if (instr->isBarrier())
-      return false;
 
    if (instr->isPseudo()) {
       switch (instr->opcode) {
@@ -543,22 +549,16 @@ needs_exec_mask(const Instruction* instr)
             if (def.getTemp().type() == RegType::vgpr)
                return true;
          }
-         return false;
+         return instr->reads_exec();
       case aco_opcode::p_spill:
       case aco_opcode::p_reload:
       case aco_opcode::p_logical_start:
       case aco_opcode::p_logical_end:
-      case aco_opcode::p_startpgm: return false;
+      case aco_opcode::p_startpgm: return instr->reads_exec();
       default: break;
       }
    }
 
-   if (instr->opcode == aco_opcode::v_readlane_b32 ||
-       instr->opcode == aco_opcode::v_readlane_b32_e64 ||
-       instr->opcode == aco_opcode::v_writelane_b32 ||
-       instr->opcode == aco_opcode::v_writelane_b32_e64)
-      return false;
-
    return true;
 }