[llvm] r340766 - DAG: Check transformed type for forming fminnum/fmaxnum from vselect
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 27 11:11:31 PDT 2018
Author: arsenm
Date: Mon Aug 27 11:11:31 2018
New Revision: 340766
URL: http://llvm.org/viewvc/llvm-project?rev=340766&view=rev
Log:
DAG: Check transformed type for forming fminnum/fmaxnum from vselect
Follow up to r340655 to fix vector types which are split.
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/trunk/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=340766&r1=340765&r2=340766&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Mon Aug 27 11:11:31 2018
@@ -7040,6 +7040,7 @@ static SDValue combineMinNumMaxNum(const
if (!(LHS == True && RHS == False) && !(LHS == False && RHS == True))
return SDValue();
+ EVT TransformVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
switch (CC) {
case ISD::SETOLT:
case ISD::SETOLE:
@@ -7048,7 +7049,7 @@ static SDValue combineMinNumMaxNum(const
case ISD::SETULT:
case ISD::SETULE: {
unsigned Opcode = (LHS == True) ? ISD::FMINNUM : ISD::FMAXNUM;
- if (TLI.isOperationLegalOrCustom(Opcode, VT))
+ if (TLI.isOperationLegalOrCustom(Opcode, TransformVT))
return DAG.getNode(Opcode, DL, VT, LHS, RHS);
return SDValue();
}
@@ -7059,7 +7060,7 @@ static SDValue combineMinNumMaxNum(const
case ISD::SETUGT:
case ISD::SETUGE: {
unsigned Opcode = (LHS == True) ? ISD::FMAXNUM : ISD::FMINNUM;
- if (TLI.isOperationLegalOrCustom(Opcode, VT))
+ if (TLI.isOperationLegalOrCustom(Opcode, TransformVT))
return DAG.getNode(Opcode, DL, VT, LHS, RHS);
return SDValue();
}
Modified: llvm/trunk/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fmax_legacy.f16.ll?rev=340766&r1=340765&r2=340766&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fmax_legacy.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fmax_legacy.f16.ll Mon Aug 27 11:11:31 2018
@@ -153,11 +153,8 @@ define <3 x half> @test_fmax_legacy_ugt_
; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v3f16:
; GFX9-NNAN: ; %bb.0:
; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NNAN-NEXT: v_max_f16_sdwa v4, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_max_f16_e32 v0, v0, v2
-; GFX9-NNAN-NEXT: v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NNAN-NEXT: v_max_f16_e32 v1, v1, v3
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v0, v4, 16, v0
+; GFX9-NNAN-NEXT: v_pk_max_f16 v1, v1, v3
+; GFX9-NNAN-NEXT: v_pk_max_f16 v0, v0, v2
; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31]
;
; VI-SAFE-LABEL: test_fmax_legacy_ugt_v3f16:
@@ -414,23 +411,10 @@ define <8 x half> @test_fmax_legacy_ugt_
; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v8f16:
; GFX9-NNAN: ; %bb.0:
; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NNAN-NEXT: v_max_f16_sdwa v8, v3, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_max_f16_sdwa v9, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_max_f16_sdwa v10, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_max_f16_sdwa v11, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_max_f16_e32 v0, v0, v4
-; GFX9-NNAN-NEXT: v_mov_b32_e32 v4, 0xffff
-; GFX9-NNAN-NEXT: v_max_f16_e32 v3, v3, v7
-; GFX9-NNAN-NEXT: v_max_f16_e32 v2, v2, v6
-; GFX9-NNAN-NEXT: v_max_f16_e32 v1, v1, v5
-; GFX9-NNAN-NEXT: v_and_b32_e32 v0, v4, v0
-; GFX9-NNAN-NEXT: v_and_b32_e32 v1, v4, v1
-; GFX9-NNAN-NEXT: v_and_b32_e32 v2, v4, v2
-; GFX9-NNAN-NEXT: v_and_b32_e32 v3, v4, v3
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v0, v11, 16, v0
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v2, v9, 16, v2
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v3, v8, 16, v3
+; GFX9-NNAN-NEXT: v_pk_max_f16 v0, v0, v4
+; GFX9-NNAN-NEXT: v_pk_max_f16 v1, v1, v5
+; GFX9-NNAN-NEXT: v_pk_max_f16 v2, v2, v6
+; GFX9-NNAN-NEXT: v_pk_max_f16 v3, v3, v7
; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31]
;
; VI-SAFE-LABEL: test_fmax_legacy_ugt_v8f16:
Modified: llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.f16.ll?rev=340766&r1=340765&r2=340766&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.f16.ll Mon Aug 27 11:11:31 2018
@@ -154,11 +154,8 @@ define <3 x half> @test_fmin_legacy_ule_
; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v3f16:
; GFX9-NNAN: ; %bb.0:
; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NNAN-NEXT: v_min_f16_sdwa v4, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_min_f16_e32 v0, v0, v2
-; GFX9-NNAN-NEXT: v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NNAN-NEXT: v_min_f16_e32 v1, v1, v3
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v0, v4, 16, v0
+; GFX9-NNAN-NEXT: v_pk_min_f16 v1, v1, v3
+; GFX9-NNAN-NEXT: v_pk_min_f16 v0, v0, v2
; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31]
;
; VI-SAFE-LABEL: test_fmin_legacy_ule_v3f16:
@@ -415,23 +412,10 @@ define <8 x half> @test_fmin_legacy_ule_
; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v8f16:
; GFX9-NNAN: ; %bb.0:
; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NNAN-NEXT: v_min_f16_sdwa v8, v3, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_min_f16_sdwa v9, v2, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_min_f16_sdwa v10, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_min_f16_sdwa v11, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NNAN-NEXT: v_min_f16_e32 v0, v0, v4
-; GFX9-NNAN-NEXT: v_mov_b32_e32 v4, 0xffff
-; GFX9-NNAN-NEXT: v_min_f16_e32 v3, v3, v7
-; GFX9-NNAN-NEXT: v_min_f16_e32 v2, v2, v6
-; GFX9-NNAN-NEXT: v_min_f16_e32 v1, v1, v5
-; GFX9-NNAN-NEXT: v_and_b32_e32 v0, v4, v0
-; GFX9-NNAN-NEXT: v_and_b32_e32 v1, v4, v1
-; GFX9-NNAN-NEXT: v_and_b32_e32 v2, v4, v2
-; GFX9-NNAN-NEXT: v_and_b32_e32 v3, v4, v3
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v0, v11, 16, v0
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v2, v9, 16, v2
-; GFX9-NNAN-NEXT: v_lshl_or_b32 v3, v8, 16, v3
+; GFX9-NNAN-NEXT: v_pk_min_f16 v0, v0, v4
+; GFX9-NNAN-NEXT: v_pk_min_f16 v1, v1, v5
+; GFX9-NNAN-NEXT: v_pk_min_f16 v2, v2, v6
+; GFX9-NNAN-NEXT: v_pk_min_f16 v3, v3, v7
; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31]
;
; VI-SAFE-LABEL: test_fmin_legacy_ule_v8f16:
More information about the llvm-commits
mailing list