[llvm] 5fe60e9 - [SDAG] Copy flags in convertMask when legalizing vselect/setcc (#180979)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Feb 12 00:38:43 PST 2026
Author: David Green
Date: 2026-02-12T08:38:37Z
New Revision: 5fe60e91e509b0ec574c80ac470a13fb0f277670
URL: https://github.com/llvm/llvm-project/commit/5fe60e91e509b0ec574c80ac470a13fb0f277670
DIFF: https://github.com/llvm/llvm-project/commit/5fe60e91e509b0ec574c80ac470a13fb0f277670.diff
LOG: [SDAG] Copy flags in convertMask when legalizing vselect/setcc (#180979)
This helps keep the flags around, making sure we correctly lower a
vector setcc with nnan. Fixes a regression from #180462.
Added:
Modified:
llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
llvm/test/CodeGen/AArch64/neon-compare-instructions.ll
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 51278a192fec9..aeb9d4d7bdc1d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -6822,7 +6822,8 @@ SDValue DAGTypeLegalizer::convertMask(SDValue InMask, EVT MaskVT,
ReplaceValueWith(InMask.getValue(1), Mask.getValue(1));
}
else
- Mask = DAG.getNode(InMask->getOpcode(), SDLoc(InMask), MaskVT, Ops);
+ Mask = DAG.getNode(InMask->getOpcode(), SDLoc(InMask), MaskVT, Ops,
+ InMask->getFlags());
// If MaskVT has smaller or bigger elements than ToMaskVT, a vector sign
// extend or truncate is needed.
diff --git a/llvm/test/CodeGen/AArch64/neon-compare-instructions.ll b/llvm/test/CodeGen/AArch64/neon-compare-instructions.ll
index f9f1d049b06d2..896b95ee1363b 100644
--- a/llvm/test/CodeGen/AArch64/neon-compare-instructions.ll
+++ b/llvm/test/CodeGen/AArch64/neon-compare-instructions.ll
@@ -4473,70 +4473,22 @@ define <4 x float> @vselect_oge_nnan(<4 x float> %a, <4 x float> %b, <4 x float>
}
define <4 x float> @vselect_olt_nnan(<4 x float> %a, <4 x float> %b, <4 x float> %d, <4 x float> %e) {
-; CHECK-SD-LABEL: vselect_olt_nnan:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: mov s4, v1.s[1]
-; CHECK-SD-NEXT: mov s5, v0.s[1]
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: mov s4, v1.s[2]
-; CHECK-SD-NEXT: mov s5, v0.s[2]
-; CHECK-SD-NEXT: csetm w8, lt
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov s1, v1.s[3]
-; CHECK-SD-NEXT: mov s0, v0.s[3]
-; CHECK-SD-NEXT: csetm w9, lt
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: fmov s6, w9
-; CHECK-SD-NEXT: mov v6.s[1], w8
-; CHECK-SD-NEXT: csetm w8, lt
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov v6.s[2], w8
-; CHECK-SD-NEXT: csetm w8, lt
-; CHECK-SD-NEXT: mov v6.s[3], w8
-; CHECK-SD-NEXT: mov v0.16b, v6.16b
-; CHECK-SD-NEXT: bsl v0.16b, v2.16b, v3.16b
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: vselect_olt_nnan:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: fcmgt v0.4s, v1.4s, v0.4s
-; CHECK-GI-NEXT: bsl v0.16b, v2.16b, v3.16b
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: vselect_olt_nnan:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcmgt v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b
+; CHECK-NEXT: ret
%c = fcmp nnan olt <4 x float> %a, %b
%r = select <4 x i1> %c, <4 x float> %d, <4 x float> %e
ret <4 x float> %r
}
define <4 x float> @vselect_ole_nnan(<4 x float> %a, <4 x float> %b, <4 x float> %d, <4 x float> %e) {
-; CHECK-SD-LABEL: vselect_ole_nnan:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: mov s4, v1.s[1]
-; CHECK-SD-NEXT: mov s5, v0.s[1]
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: mov s4, v1.s[2]
-; CHECK-SD-NEXT: mov s5, v0.s[2]
-; CHECK-SD-NEXT: csetm w8, le
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov s1, v1.s[3]
-; CHECK-SD-NEXT: mov s0, v0.s[3]
-; CHECK-SD-NEXT: csetm w9, le
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: fmov s6, w9
-; CHECK-SD-NEXT: mov v6.s[1], w8
-; CHECK-SD-NEXT: csetm w8, le
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov v6.s[2], w8
-; CHECK-SD-NEXT: csetm w8, le
-; CHECK-SD-NEXT: mov v6.s[3], w8
-; CHECK-SD-NEXT: mov v0.16b, v6.16b
-; CHECK-SD-NEXT: bsl v0.16b, v2.16b, v3.16b
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: vselect_ole_nnan:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: fcmge v0.4s, v1.4s, v0.4s
-; CHECK-GI-NEXT: bsl v0.16b, v2.16b, v3.16b
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: vselect_ole_nnan:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcmge v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b
+; CHECK-NEXT: ret
%c = fcmp nnan ole <4 x float> %a, %b
%r = select <4 x i1> %c, <4 x float> %d, <4 x float> %e
ret <4 x float> %r
@@ -4633,25 +4585,7 @@ define <4 x float> @vselect_uge_nnan(<4 x float> %a, <4 x float> %b, <4 x float>
define <4 x float> @vselect_ult_nnan(<4 x float> %a, <4 x float> %b, <4 x float> %d, <4 x float> %e) {
; CHECK-SD-LABEL: vselect_ult_nnan:
; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: mov s4, v1.s[1]
-; CHECK-SD-NEXT: mov s5, v0.s[1]
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: mov s4, v1.s[2]
-; CHECK-SD-NEXT: mov s5, v0.s[2]
-; CHECK-SD-NEXT: csetm w8, lt
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov s1, v1.s[3]
-; CHECK-SD-NEXT: mov s0, v0.s[3]
-; CHECK-SD-NEXT: csetm w9, lt
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: fmov s6, w9
-; CHECK-SD-NEXT: mov v6.s[1], w8
-; CHECK-SD-NEXT: csetm w8, lt
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov v6.s[2], w8
-; CHECK-SD-NEXT: csetm w8, lt
-; CHECK-SD-NEXT: mov v6.s[3], w8
-; CHECK-SD-NEXT: mov v0.16b, v6.16b
+; CHECK-SD-NEXT: fcmgt v0.4s, v1.4s, v0.4s
; CHECK-SD-NEXT: bsl v0.16b, v2.16b, v3.16b
; CHECK-SD-NEXT: ret
;
@@ -4669,25 +4603,7 @@ define <4 x float> @vselect_ult_nnan(<4 x float> %a, <4 x float> %b, <4 x float>
define <4 x float> @vselect_ule_nnan(<4 x float> %a, <4 x float> %b, <4 x float> %d, <4 x float> %e) {
; CHECK-SD-LABEL: vselect_ule_nnan:
; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: mov s4, v1.s[1]
-; CHECK-SD-NEXT: mov s5, v0.s[1]
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: mov s4, v1.s[2]
-; CHECK-SD-NEXT: mov s5, v0.s[2]
-; CHECK-SD-NEXT: csetm w8, le
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov s1, v1.s[3]
-; CHECK-SD-NEXT: mov s0, v0.s[3]
-; CHECK-SD-NEXT: csetm w9, le
-; CHECK-SD-NEXT: fcmp s5, s4
-; CHECK-SD-NEXT: fmov s6, w9
-; CHECK-SD-NEXT: mov v6.s[1], w8
-; CHECK-SD-NEXT: csetm w8, le
-; CHECK-SD-NEXT: fcmp s0, s1
-; CHECK-SD-NEXT: mov v6.s[2], w8
-; CHECK-SD-NEXT: csetm w8, le
-; CHECK-SD-NEXT: mov v6.s[3], w8
-; CHECK-SD-NEXT: mov v0.16b, v6.16b
+; CHECK-SD-NEXT: fcmge v0.4s, v1.4s, v0.4s
; CHECK-SD-NEXT: bsl v0.16b, v2.16b, v3.16b
; CHECK-SD-NEXT: ret
;
More information about the llvm-commits
mailing list