[llvm] 723a53c - [AMDGPU] Avoid constant bus limitation on V_BFE GISel pattern

[via llvm-commits]

Author: pvanhout
Date: 2023-03-15T15:01:33+01:00
New Revision: 723a53caaf93ff47d3a74324ace1ab05de994c12

URL: https://github.com/llvm/llvm-project/commit/723a53caaf93ff47d3a74324ace1ab05de994c12
DIFF: https://github.com/llvm/llvm-project/commit/723a53caaf93ff47d3a74324ace1ab05de994c12.diff

LOG: [AMDGPU] Avoid constant bus limitation on V_BFE GISel pattern

For D141247 - if that pattern was used by GISel it could cause constant bus limitation failures.
Just use inline immediates instead of S_MOV to avoid the issue.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D146131

Added: 
    llvm/test/CodeGen/AMDGPU/GlobalISel/v_bfe_i32.ll

Modified: 
    llvm/lib/Target/AMDGPU/VOP3Instructions.td

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/VOP3Instructions.td b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
index 70e50111790bb..65db790fedf1c 100644
--- a/llvm/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
@@ -263,7 +263,7 @@ let SchedRW = [Write64Bit] in {
 
 def : GCNPat<
   (i32 (DivergentUnaryFrag<sext> i16:$src)),
-  (i32 (V_BFE_I32_e64 $src, (S_MOV_B32 (i32 0)), (S_MOV_B32 (i32 0x10))))
+  (i32 (V_BFE_I32_e64 i16:$src, (i32 0), (i32 0x10)))
 >;
 
 let isReMaterializable = 1 in {

diff  --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/v_bfe_i32.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/v_bfe_i32.ll
new file mode 100644
index 0000000000000..a07c760576350
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/v_bfe_i32.ll
@@ -0,0 +1,25 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd-amdhsa --global-isel -verify-machineinstrs < %s | FileCheck --check-prefix=PREGFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa --global-isel -mcpu=hawaii -verify-machineinstrs < %s | FileCheck --check-prefix=PREGFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa --global-isel -mcpu=fiji -verify-machineinstrs < %s | FileCheck --check-prefix=PREGFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa --global-isel -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck --check-prefix=PREGFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa --global-isel -mcpu=gfx1030 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10PLUS %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa --global-isel -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10PLUS %s
+
+define i32 @check_v_bfe(i16 %a) {
+; PREGFX9-LABEL: check_v_bfe:
+; PREGFX9:       ; %bb.0: ; %entry
+; PREGFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; PREGFX9-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; PREGFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10PLUS-LABEL: check_v_bfe:
+; GFX10PLUS:       ; %bb.0: ; %entry
+; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10PLUS-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10PLUS-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX10PLUS-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %res = sext i16 %a to i32
+  ret i32 %res
+}