[llvm] r360512 - [X86] Don't emit MOVNTDQA loads from fast-isel without SSE4.1.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Fri May 10 21:19:33 PDT 2019
Author: ctopper
Date: Fri May 10 21:19:33 2019
New Revision: 360512
URL: http://llvm.org/viewvc/llvm-project?rev=360512&view=rev
Log:
[X86] Don't emit MOVNTDQA loads from fast-isel without SSE4.1.
We were checking for SSE4.1 for FP types, but not integer 128-bit types.
Fixes PR41837.
Modified:
llvm/trunk/lib/Target/X86/X86FastISel.cpp
llvm/trunk/test/CodeGen/X86/fast-isel-nontemporal.ll
Modified: llvm/trunk/lib/Target/X86/X86FastISel.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86FastISel.cpp?rev=360512&r1=360511&r2=360512&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86FastISel.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86FastISel.cpp Fri May 10 21:19:33 2019
@@ -396,7 +396,7 @@ bool X86FastISel::X86FastEmitLoad(EVT VT
case MVT::v2i64:
case MVT::v8i16:
case MVT::v16i8:
- if (IsNonTemporal && Alignment >= 16)
+ if (IsNonTemporal && Alignment >= 16 && HasSSE41)
Opc = HasVLX ? X86::VMOVNTDQAZ128rm :
HasAVX ? X86::VMOVNTDQArm : X86::MOVNTDQArm;
else if (Alignment >= 16)
Modified: llvm/trunk/test/CodeGen/X86/fast-isel-nontemporal.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fast-isel-nontemporal.ll?rev=360512&r1=360511&r2=360512&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/fast-isel-nontemporal.ll (original)
+++ llvm/trunk/test/CodeGen/X86/fast-isel-nontemporal.ll Fri May 10 21:19:33 2019
@@ -300,10 +300,20 @@ entry:
}
define <16 x i8> @test_load_nt16xi8(<16 x i8>* nocapture %ptr) {
-; SSE-LABEL: test_load_nt16xi8:
-; SSE: # %bb.0: # %entry
-; SSE-NEXT: movntdqa (%rdi), %xmm0
-; SSE-NEXT: retq
+; SSE2-LABEL: test_load_nt16xi8:
+; SSE2: # %bb.0: # %entry
+; SSE2-NEXT: movdqa (%rdi), %xmm0
+; SSE2-NEXT: retq
+;
+; SSE4A-LABEL: test_load_nt16xi8:
+; SSE4A: # %bb.0: # %entry
+; SSE4A-NEXT: movdqa (%rdi), %xmm0
+; SSE4A-NEXT: retq
+;
+; SSE41-LABEL: test_load_nt16xi8:
+; SSE41: # %bb.0: # %entry
+; SSE41-NEXT: movntdqa (%rdi), %xmm0
+; SSE41-NEXT: retq
;
; AVX-LABEL: test_load_nt16xi8:
; AVX: # %bb.0: # %entry
@@ -320,10 +330,20 @@ entry:
}
define <8 x i16> @test_load_nt8xi16(<8 x i16>* nocapture %ptr) {
-; SSE-LABEL: test_load_nt8xi16:
-; SSE: # %bb.0: # %entry
-; SSE-NEXT: movntdqa (%rdi), %xmm0
-; SSE-NEXT: retq
+; SSE2-LABEL: test_load_nt8xi16:
+; SSE2: # %bb.0: # %entry
+; SSE2-NEXT: movdqa (%rdi), %xmm0
+; SSE2-NEXT: retq
+;
+; SSE4A-LABEL: test_load_nt8xi16:
+; SSE4A: # %bb.0: # %entry
+; SSE4A-NEXT: movdqa (%rdi), %xmm0
+; SSE4A-NEXT: retq
+;
+; SSE41-LABEL: test_load_nt8xi16:
+; SSE41: # %bb.0: # %entry
+; SSE41-NEXT: movntdqa (%rdi), %xmm0
+; SSE41-NEXT: retq
;
; AVX-LABEL: test_load_nt8xi16:
; AVX: # %bb.0: # %entry
@@ -340,10 +360,20 @@ entry:
}
define <4 x i32> @test_load_nt4xi32(<4 x i32>* nocapture %ptr) {
-; SSE-LABEL: test_load_nt4xi32:
-; SSE: # %bb.0: # %entry
-; SSE-NEXT: movntdqa (%rdi), %xmm0
-; SSE-NEXT: retq
+; SSE2-LABEL: test_load_nt4xi32:
+; SSE2: # %bb.0: # %entry
+; SSE2-NEXT: movdqa (%rdi), %xmm0
+; SSE2-NEXT: retq
+;
+; SSE4A-LABEL: test_load_nt4xi32:
+; SSE4A: # %bb.0: # %entry
+; SSE4A-NEXT: movdqa (%rdi), %xmm0
+; SSE4A-NEXT: retq
+;
+; SSE41-LABEL: test_load_nt4xi32:
+; SSE41: # %bb.0: # %entry
+; SSE41-NEXT: movntdqa (%rdi), %xmm0
+; SSE41-NEXT: retq
;
; AVX-LABEL: test_load_nt4xi32:
; AVX: # %bb.0: # %entry
@@ -360,10 +390,20 @@ entry:
}
define <2 x i64> @test_load_nt2xi64(<2 x i64>* nocapture %ptr) {
-; SSE-LABEL: test_load_nt2xi64:
-; SSE: # %bb.0: # %entry
-; SSE-NEXT: movntdqa (%rdi), %xmm0
-; SSE-NEXT: retq
+; SSE2-LABEL: test_load_nt2xi64:
+; SSE2: # %bb.0: # %entry
+; SSE2-NEXT: movdqa (%rdi), %xmm0
+; SSE2-NEXT: retq
+;
+; SSE4A-LABEL: test_load_nt2xi64:
+; SSE4A: # %bb.0: # %entry
+; SSE4A-NEXT: movdqa (%rdi), %xmm0
+; SSE4A-NEXT: retq
+;
+; SSE41-LABEL: test_load_nt2xi64:
+; SSE41: # %bb.0: # %entry
+; SSE41-NEXT: movntdqa (%rdi), %xmm0
+; SSE41-NEXT: retq
;
; AVX-LABEL: test_load_nt2xi64:
; AVX: # %bb.0: # %entry
More information about the llvm-commits
mailing list