[llvm] AMDGPU: Disable AGPR allocation in VGPR MFMA tests (PR #150873)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Jul 27 21:47:14 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Matt Arsenault (arsenm)
<details>
<summary>Changes</summary>
The intent of these tests broke at some point; these were supposed
to test both selection paths but the agpr and vgpr versions of the
test functions were both selecting to the AGPR version. Explicitly
disable AGPR usage with the attribute.
---
Patch is 226.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/150873.diff
5 Files Affected:
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll (+117-147)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll (+893-1101)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.16x16x128.f8f6f4.ll (+137-155)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.ll (+281-349)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll (+170-170)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll
index d358837452eab..8081a15b53bb7 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll
@@ -252,62 +252,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd(<8 x bfloat> %arg
; GCN-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24
; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GCN-NEXT: v_mov_b32_e32 v8, 0
+; GCN-NEXT: v_mov_b32_e32 v44, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT: v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT: v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT: v_accvgpr_write_b32 a31, s23
-; GCN-NEXT: v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT: v_accvgpr_write_b32 a30, s22
-; GCN-NEXT: v_accvgpr_write_b32 a29, s21
-; GCN-NEXT: v_accvgpr_write_b32 a28, s20
-; GCN-NEXT: v_accvgpr_write_b32 a27, s19
-; GCN-NEXT: v_accvgpr_write_b32 a26, s18
-; GCN-NEXT: v_accvgpr_write_b32 a25, s17
-; GCN-NEXT: v_accvgpr_write_b32 a24, s16
-; GCN-NEXT: v_accvgpr_write_b32 a23, s15
-; GCN-NEXT: v_accvgpr_write_b32 a22, s14
-; GCN-NEXT: v_accvgpr_write_b32 a21, s13
-; GCN-NEXT: v_accvgpr_write_b32 a20, s12
-; GCN-NEXT: v_accvgpr_write_b32 a19, s11
-; GCN-NEXT: v_accvgpr_write_b32 a18, s10
-; GCN-NEXT: v_accvgpr_write_b32 a17, s9
-; GCN-NEXT: v_accvgpr_write_b32 a16, s8
-; GCN-NEXT: v_mov_b32_e32 v10, s20
-; GCN-NEXT: v_mov_b32_e32 v11, s21
-; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[16:31]
-; GCN-NEXT: v_mov_b32_e32 v12, s22
-; GCN-NEXT: v_mov_b32_e32 v13, s23
-; GCN-NEXT: v_mov_b32_e32 v0, s16
-; GCN-NEXT: v_mov_b32_e32 v1, s17
-; GCN-NEXT: v_mov_b32_e32 v2, s18
-; GCN-NEXT: v_mov_b32_e32 v3, s19
-; GCN-NEXT: global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
-; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT: v_mov_b64_e32 v[34:35], s[26:27]
+; GCN-NEXT: v_mov_b64_e32 v[32:33], s[24:25]
+; GCN-NEXT: v_mov_b64_e32 v[38:39], s[30:31]
+; GCN-NEXT: v_mov_b64_e32 v[30:31], s[22:23]
+; GCN-NEXT: v_mov_b64_e32 v[36:37], s[28:29]
+; GCN-NEXT: v_mov_b64_e32 v[28:29], s[20:21]
+; GCN-NEXT: v_mov_b64_e32 v[26:27], s[18:19]
+; GCN-NEXT: v_mov_b64_e32 v[24:25], s[16:17]
+; GCN-NEXT: v_mov_b64_e32 v[22:23], s[14:15]
+; GCN-NEXT: v_mov_b64_e32 v[20:21], s[12:13]
+; GCN-NEXT: v_mov_b64_e32 v[18:19], s[10:11]
+; GCN-NEXT: v_mov_b64_e32 v[16:17], s[8:9]
+; GCN-NEXT: v_mov_b32_e32 v40, s20
+; GCN-NEXT: v_mov_b32_e32 v41, s21
+; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[32:35], v[36:39], v[16:31]
+; GCN-NEXT: v_mov_b32_e32 v42, s22
+; GCN-NEXT: v_mov_b32_e32 v43, s23
+; GCN-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_nop 2
+; GCN-NEXT: v_mov_b32_e32 v16, s16
+; GCN-NEXT: v_mov_b32_e32 v17, s17
+; GCN-NEXT: v_mov_b32_e32 v18, s18
+; GCN-NEXT: v_mov_b32_e32 v19, s19
+; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_nop 0
-; GCN-NEXT: v_mov_b32_e32 v0, s12
-; GCN-NEXT: v_mov_b32_e32 v1, s13
-; GCN-NEXT: v_mov_b32_e32 v2, s14
-; GCN-NEXT: v_mov_b32_e32 v3, s15
-; GCN-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT: v_mov_b32_e32 v16, s12
+; GCN-NEXT: v_mov_b32_e32 v17, s13
+; GCN-NEXT: v_mov_b32_e32 v18, s14
+; GCN-NEXT: v_mov_b32_e32 v19, s15
+; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_nop 0
-; GCN-NEXT: v_mov_b32_e32 v0, s8
-; GCN-NEXT: v_mov_b32_e32 v1, s9
-; GCN-NEXT: v_mov_b32_e32 v2, s10
-; GCN-NEXT: v_mov_b32_e32 v3, s11
-; GCN-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; GCN-NEXT: v_mov_b32_e32 v16, s8
+; GCN-NEXT: v_mov_b32_e32 v17, s9
+; GCN-NEXT: v_mov_b32_e32 v18, s10
+; GCN-NEXT: v_mov_b32_e32 v19, s11
+; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_endpgm
%result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0)
@@ -322,62 +315,55 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd__flags(<8 x bfloa
; GCN-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24
; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4
-; GCN-NEXT: v_mov_b32_e32 v8, 0
+; GCN-NEXT: v_mov_b32_e32 v44, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT: v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT: v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT: v_accvgpr_write_b32 a31, s23
-; GCN-NEXT: v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT: v_accvgpr_write_b32 a30, s22
-; GCN-NEXT: v_accvgpr_write_b32 a29, s21
-; GCN-NEXT: v_accvgpr_write_b32 a28, s20
-; GCN-NEXT: v_accvgpr_write_b32 a27, s19
-; GCN-NEXT: v_accvgpr_write_b32 a26, s18
-; GCN-NEXT: v_accvgpr_write_b32 a25, s17
-; GCN-NEXT: v_accvgpr_write_b32 a24, s16
-; GCN-NEXT: v_accvgpr_write_b32 a23, s15
-; GCN-NEXT: v_accvgpr_write_b32 a22, s14
-; GCN-NEXT: v_accvgpr_write_b32 a21, s13
-; GCN-NEXT: v_accvgpr_write_b32 a20, s12
-; GCN-NEXT: v_accvgpr_write_b32 a19, s11
-; GCN-NEXT: v_accvgpr_write_b32 a18, s10
-; GCN-NEXT: v_accvgpr_write_b32 a17, s9
-; GCN-NEXT: v_accvgpr_write_b32 a16, s8
-; GCN-NEXT: v_mov_b32_e32 v10, s20
-; GCN-NEXT: v_mov_b32_e32 v11, s21
-; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[16:31] cbsz:1 abid:2 blgp:3
-; GCN-NEXT: v_mov_b32_e32 v12, s22
-; GCN-NEXT: v_mov_b32_e32 v13, s23
-; GCN-NEXT: v_mov_b32_e32 v0, s16
-; GCN-NEXT: v_mov_b32_e32 v1, s17
-; GCN-NEXT: v_mov_b32_e32 v2, s18
-; GCN-NEXT: v_mov_b32_e32 v3, s19
-; GCN-NEXT: global_store_dwordx4 v8, v[10:13], s[0:1] offset:48 sc0 sc1
-; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT: v_mov_b64_e32 v[34:35], s[26:27]
+; GCN-NEXT: v_mov_b64_e32 v[32:33], s[24:25]
+; GCN-NEXT: v_mov_b64_e32 v[38:39], s[30:31]
+; GCN-NEXT: v_mov_b64_e32 v[30:31], s[22:23]
+; GCN-NEXT: v_mov_b64_e32 v[36:37], s[28:29]
+; GCN-NEXT: v_mov_b64_e32 v[28:29], s[20:21]
+; GCN-NEXT: v_mov_b64_e32 v[26:27], s[18:19]
+; GCN-NEXT: v_mov_b64_e32 v[24:25], s[16:17]
+; GCN-NEXT: v_mov_b64_e32 v[22:23], s[14:15]
+; GCN-NEXT: v_mov_b64_e32 v[20:21], s[12:13]
+; GCN-NEXT: v_mov_b64_e32 v[18:19], s[10:11]
+; GCN-NEXT: v_mov_b64_e32 v[16:17], s[8:9]
+; GCN-NEXT: v_mov_b32_e32 v40, s20
+; GCN-NEXT: v_mov_b32_e32 v41, s21
+; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3
+; GCN-NEXT: v_mov_b32_e32 v42, s22
+; GCN-NEXT: v_mov_b32_e32 v43, s23
+; GCN-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_nop 2
+; GCN-NEXT: v_mov_b32_e32 v16, s16
+; GCN-NEXT: v_mov_b32_e32 v17, s17
+; GCN-NEXT: v_mov_b32_e32 v18, s18
+; GCN-NEXT: v_mov_b32_e32 v19, s19
+; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_nop 0
-; GCN-NEXT: v_mov_b32_e32 v0, s12
-; GCN-NEXT: v_mov_b32_e32 v1, s13
-; GCN-NEXT: v_mov_b32_e32 v2, s14
-; GCN-NEXT: v_mov_b32_e32 v3, s15
-; GCN-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT: v_mov_b32_e32 v16, s12
+; GCN-NEXT: v_mov_b32_e32 v17, s13
+; GCN-NEXT: v_mov_b32_e32 v18, s14
+; GCN-NEXT: v_mov_b32_e32 v19, s15
+; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_nop 0
-; GCN-NEXT: v_mov_b32_e32 v0, s8
-; GCN-NEXT: v_mov_b32_e32 v1, s9
-; GCN-NEXT: v_mov_b32_e32 v2, s10
-; GCN-NEXT: v_mov_b32_e32 v3, s11
-; GCN-NEXT: global_store_dwordx4 v8, v[0:3], s[0:1] sc0 sc1
+; GCN-NEXT: v_mov_b32_e32 v16, s8
+; GCN-NEXT: v_mov_b32_e32 v17, s9
+; GCN-NEXT: v_mov_b32_e32 v18, s10
+; GCN-NEXT: v_mov_b32_e32 v19, s11
+; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[8:11], s[0:1] offset:32 sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[12:15], s[0:1] offset:48 sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[0:3], s[0:1] sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: global_store_dwordx4 v8, a[4:7], s[0:1] offset:16 sc0 sc1
+; GCN-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_endpgm
%result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 1, i32 2, i32 3)
@@ -393,35 +379,27 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd_mac(<8 x bfloat>
; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4
; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT: v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT: v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT: v_accvgpr_write_b32 a0, s8
-; GCN-NEXT: v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT: v_accvgpr_write_b32 a1, s9
-; GCN-NEXT: v_accvgpr_write_b32 a2, s10
-; GCN-NEXT: v_accvgpr_write_b32 a3, s11
-; GCN-NEXT: v_accvgpr_write_b32 a4, s12
-; GCN-NEXT: v_accvgpr_write_b32 a5, s13
-; GCN-NEXT: v_accvgpr_write_b32 a6, s14
-; GCN-NEXT: v_accvgpr_write_b32 a7, s15
-; GCN-NEXT: v_accvgpr_write_b32 a8, s16
-; GCN-NEXT: v_accvgpr_write_b32 a9, s17
-; GCN-NEXT: v_accvgpr_write_b32 a10, s18
-; GCN-NEXT: v_accvgpr_write_b32 a11, s19
-; GCN-NEXT: v_accvgpr_write_b32 a12, s20
-; GCN-NEXT: v_accvgpr_write_b32 a13, s21
-; GCN-NEXT: v_accvgpr_write_b32 a14, s22
-; GCN-NEXT: v_accvgpr_write_b32 a15, s23
+; GCN-NEXT: v_mov_b64_e32 v[16:17], s[24:25]
+; GCN-NEXT: v_mov_b64_e32 v[18:19], s[26:27]
+; GCN-NEXT: v_mov_b64_e32 v[20:21], s[28:29]
+; GCN-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
+; GCN-NEXT: v_mov_b64_e32 v[22:23], s[30:31]
+; GCN-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
+; GCN-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
+; GCN-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
+; GCN-NEXT: v_mov_b64_e32 v[8:9], s[16:17]
+; GCN-NEXT: v_mov_b64_e32 v[10:11], s[18:19]
+; GCN-NEXT: v_mov_b64_e32 v[12:13], s[20:21]
+; GCN-NEXT: v_mov_b64_e32 v[14:15], s[22:23]
; GCN-NEXT: s_nop 1
-; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[0:15]
-; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[16:19], v[20:23], v[0:15]
+; GCN-NEXT: v_mov_b32_e32 v16, 0
; GCN-NEXT: s_nop 7
; GCN-NEXT: s_nop 2
-; GCN-NEXT: global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; GCN-NEXT: global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GCN-NEXT: global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GCN-NEXT: global_store_dwordx4 v0, a[0:3], s[0:1]
+; GCN-NEXT: global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GCN-NEXT: global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GCN-NEXT: global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GCN-NEXT: global_store_dwordx4 v16, v[0:3], s[0:1]
; GCN-NEXT: s_endpgm
%result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0)
store <16 x float> %result, ptr addrspace(1) %out
@@ -435,40 +413,32 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd_mac_flags(<8 x bf
; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4
; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: v_mov_b64_e32 v[0:1], s[24:25]
-; GCN-NEXT: v_mov_b64_e32 v[2:3], s[26:27]
-; GCN-NEXT: v_mov_b64_e32 v[4:5], s[28:29]
-; GCN-NEXT: v_accvgpr_write_b32 a0, s8
-; GCN-NEXT: v_mov_b64_e32 v[6:7], s[30:31]
-; GCN-NEXT: v_accvgpr_write_b32 a1, s9
-; GCN-NEXT: v_accvgpr_write_b32 a2, s10
-; GCN-NEXT: v_accvgpr_write_b32 a3, s11
-; GCN-NEXT: v_accvgpr_write_b32 a4, s12
-; GCN-NEXT: v_accvgpr_write_b32 a5, s13
-; GCN-NEXT: v_accvgpr_write_b32 a6, s14
-; GCN-NEXT: v_accvgpr_write_b32 a7, s15
-; GCN-NEXT: v_accvgpr_write_b32 a8, s16
-; GCN-NEXT: v_accvgpr_write_b32 a9, s17
-; GCN-NEXT: v_accvgpr_write_b32 a10, s18
-; GCN-NEXT: v_accvgpr_write_b32 a11, s19
-; GCN-NEXT: v_accvgpr_write_b32 a12, s20
-; GCN-NEXT: v_accvgpr_write_b32 a13, s21
-; GCN-NEXT: v_accvgpr_write_b32 a14, s22
-; GCN-NEXT: v_accvgpr_write_b32 a15, s23
+; GCN-NEXT: v_mov_b64_e32 v[16:17], s[24:25]
+; GCN-NEXT: v_mov_b64_e32 v[18:19], s[26:27]
+; GCN-NEXT: v_mov_b64_e32 v[20:21], s[28:29]
+; GCN-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
+; GCN-NEXT: v_mov_b64_e32 v[22:23], s[30:31]
+; GCN-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
+; GCN-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
+; GCN-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
+; GCN-NEXT: v_mov_b64_e32 v[8:9], s[16:17]
+; GCN-NEXT: v_mov_b64_e32 v[10:11], s[18:19]
+; GCN-NEXT: v_mov_b64_e32 v[12:13], s[20:21]
+; GCN-NEXT: v_mov_b64_e32 v[14:15], s[22:23]
; GCN-NEXT: s_nop 1
-; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[0:15], v[0:3], v[4:7], a[0:15] cbsz:3 abid:2 blgp:1
-; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[16:19], v[20:23], v[0:15] cbsz:3 abid:2 blgp:1
+; GCN-NEXT: v_mov_b32_e32 v16, 0
; GCN-NEXT: s_nop 7
; GCN-NEXT: s_nop 2
-; GCN-NEXT: global_store_dwordx4 v0, a[12:15], s[0:1] offset:48
-; GCN-NEXT: global_store_dwordx4 v0, a[8:11], s[0:1] offset:32
-; GCN-NEXT: global_store_dwordx4 v0, a[4:7], s[0:1] offset:16
-; GCN-NEXT: global_store_dwordx4 v0, a[0:3], s[0:1]
+; GCN-NEXT: global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GCN-NEXT: global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GCN-NEXT: global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GCN-NEXT: global_store_dwordx4 v16, v[0:3], s[0:1]
; GCN-NEXT: s_endpgm
%result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 3, i32 2, i32 1)
store <16 x float> %result, ptr addrspace(1) %out
ret void
}
-attributes #0 = { "amdgpu-flat-work-group-size"="512,512" }
+attributes #0 = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-agpr-alloc"="0,0" }
attributes #1 = { "amdgpu-flat-work-group-size"="1,64" }
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
index 21465beb21de7..d81ec1c61634b 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
@@ -141,20 +141,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp
; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34
; SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54
; SDAG-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24
-; SDAG-NEXT: v_mov_b32_e32 v8, 0
+; SDAG-NEXT: v_mov_b32_e32 v12, 0
; SDAG-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
-; SDAG-NEXT: v_accvgpr_write_b32 a0, s0
+; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
-; SDAG-NEXT: v_accvgpr_write_b32 a1, s1
-; SDAG-NEXT: v_accvgpr_write_b32 a2, s2
-; SDAG-NEXT: v_accvgpr_write_b32 a3, s3
+; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
; SDAG-NEXT: s_nop 1
-; SDAG-NEXT: v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3]
+; SDAG-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11]
; SDAG-NEXT: s_nop 7
-; SDAG-NEXT: global_store_dwordx4 v8, a[0:3], s[6:7]
+; SDAG-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7]
; SDAG-NEXT: s_endpgm
;
; GISEL-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd:
@@ -166,16 +164,14 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp
; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
-; GISEL-NEXT: v_accvgpr_write_b32 a0, s0
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
-; GISEL-NEXT: v_accvgpr_write_b32 a1, s1
-; GISEL-NEXT: v_accvgpr_write_b32 a2, s2
-; GISEL-NEXT: v_accvgpr_write_b32 a3, s3
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
; GISEL-NEXT: s_nop 1
-; GISEL-NEXT: v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3]
-; GISEL-NEXT: v_mov_b32_e32 v0, 0
+; GISEL-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11]
+; GISEL-NEXT: v_mov_b32_e32 v4, 0
; GISEL-NEXT: s_nop 6
-; GISEL-NEXT: global_store_dwordx4 v0, a[0:3], s[6:7]
+; GISEL-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7]
; GISEL-NEXT: s_endpgm
;
; HEURRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd:
@@ -183,20 +179,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp
; HEURRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34
; HEURRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54
; HEURRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24
-; HEURRC-NEXT: v_mov_b32_e32 v8, 0
+; HEURRC-NEXT: v_mov_b32_e32 v12, 0
; HEURRC-NEXT: s_waitcnt lgkmcnt(0)
; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
-; HEURRC-NEXT: v_accvgpr_write_b32 a0, s0
+; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3]
; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
-; HEURRC-NEXT: v_accvgpr_write_b32 a1, s1
-; HEURRC-NEXT: v_accvgpr_write_b32 a2, s2
-; HEURRC-NEXT: v_accvgpr_write_b32 a3, s3
+; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1]
; HEURRC-NEXT: s_nop 1
-; HEURRC-NEXT: v_mfma_f32_16x16x32_f16 a[0:3], v[0:3], v[4:7], a[0:3]
+; HEURRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11]
; HEURRC-NEXT: s_nop 7
-; HEURRC-NEXT: global_store_dwordx4 v8, a[0:3], s[6:7]
+; HEURRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7]
; HEURRC-NEXT: s_endpgm
;
; VGPRRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd:
@@ -266,20 +260,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr
; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34
; SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54
; SDAG-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24
-; SDAG-NEXT: v_mov_b32_e32 v8, 0
+; SDAG-NEXT: v_mov_b32_e32 v12, 0
; SDAG-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
-; SDAG-NEXT: v_accvgpr_write_b...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/150873
More information about the llvm-commits
mailing list