From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Mon, 27 Aug 2018 18:11:31 +0000 (+0000)
Subject: DAG: Check transformed type for forming fminnum/fmaxnum from vselect
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=cea7c6969d777615b0a29584b1be4c90ea968944;p=platform%2Fupstream%2Fllvm.git

DAG: Check transformed type for forming fminnum/fmaxnum from vselect

Follow up to r340655 to fix vector types which are split.

llvm-svn: 340766
---

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 44fdbc1b..0addc50 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -7040,6 +7040,7 @@ static SDValue combineMinNumMaxNum(const SDLoc &DL, EVT VT, SDValue LHS,
   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 SDLoc &DL, EVT VT, SDValue LHS,
   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 SDLoc &DL, EVT VT, SDValue LHS,
   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();
   }
diff --git a/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll b/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
index 755a46b..4563ec4 100644
--- a/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
@@ -153,11 +153,8 @@ define <3 x half> @test_fmax_legacy_ugt_v3f16(<3 x half> %a, <3 x half> %b) #0 {
 ; 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_v8f16(<8 x half> %a, <8 x half> %b) #0 {
 ; 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:
diff --git a/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll b/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
index 9ad2bd4..58c70c0 100644
--- a/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
@@ -154,11 +154,8 @@ define <3 x half> @test_fmin_legacy_ule_v3f16(<3 x half> %a, <3 x half> %b) #0 {
 ; 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_v8f16(<8 x half> %a, <8 x half> %b) #0 {
 ; 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: