[x86] add fake scalar FP logic instructions to ReplaceableInstrs to save some bytes
We can replace "scalar" FP-bitwise-logic with other forms of bitwise-logic instructions.
Scalar SSE/AVX FP-logic instructions only exist in your imagination and/or the bowels of
compilers, but logically equivalent int, float, and double variants of bitwise-logic
instructions are reality in x86, and the float variant may be a shorter instruction
depending on which flavor (SSE or AVX) of vector ISA you have...so just prefer float all
the time.
This is a preliminary step towards solving PR6137:
https://llvm.org/bugs/show_bug.cgi?id=6137
Differential Revision:
https://reviews.llvm.org/D26712
llvm-svn: 287122
diff --git a/llvm/test/CodeGen/X86/fast-isel-select-sse.ll b/llvm/test/CodeGen/X86/fast-isel-select-sse.ll
index 033fc23..026732d 100644
--- a/llvm/test/CodeGen/X86/fast-isel-select-sse.ll
+++ b/llvm/test/CodeGen/X86/fast-isel-select-sse.ll
@@ -30,9 +30,9 @@
; SSE-LABEL: select_fcmp_oeq_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpeqsd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_oeq_f64:
@@ -71,10 +71,10 @@
; SSE-LABEL: select_fcmp_ogt_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpltsd %xmm0, %xmm1
-; SSE-NEXT: andpd %xmm1, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm1
-; SSE-NEXT: orpd %xmm2, %xmm1
-; SSE-NEXT: movapd %xmm1, %xmm0
+; SSE-NEXT: andps %xmm1, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm1
+; SSE-NEXT: orps %xmm2, %xmm1
+; SSE-NEXT: movaps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_ogt_f64:
@@ -113,10 +113,10 @@
; SSE-LABEL: select_fcmp_oge_f64:
; SSE: # BB#0:
; SSE-NEXT: cmplesd %xmm0, %xmm1
-; SSE-NEXT: andpd %xmm1, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm1
-; SSE-NEXT: orpd %xmm2, %xmm1
-; SSE-NEXT: movapd %xmm1, %xmm0
+; SSE-NEXT: andps %xmm1, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm1
+; SSE-NEXT: orps %xmm2, %xmm1
+; SSE-NEXT: movaps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_oge_f64:
@@ -154,9 +154,9 @@
; SSE-LABEL: select_fcmp_olt_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpltsd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_olt_f64:
@@ -194,9 +194,9 @@
; SSE-LABEL: select_fcmp_ole_f64:
; SSE: # BB#0:
; SSE-NEXT: cmplesd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_ole_f64:
@@ -234,9 +234,9 @@
; SSE-LABEL: select_fcmp_ord_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpordsd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_ord_f64:
@@ -274,9 +274,9 @@
; SSE-LABEL: select_fcmp_uno_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpunordsd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_uno_f64:
@@ -314,9 +314,9 @@
; SSE-LABEL: select_fcmp_ugt_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpnlesd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_ugt_f64:
@@ -354,9 +354,9 @@
; SSE-LABEL: select_fcmp_uge_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpnltsd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_uge_f64:
@@ -395,10 +395,10 @@
; SSE-LABEL: select_fcmp_ult_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpnlesd %xmm0, %xmm1
-; SSE-NEXT: andpd %xmm1, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm1
-; SSE-NEXT: orpd %xmm2, %xmm1
-; SSE-NEXT: movapd %xmm1, %xmm0
+; SSE-NEXT: andps %xmm1, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm1
+; SSE-NEXT: orps %xmm2, %xmm1
+; SSE-NEXT: movaps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_ult_f64:
@@ -437,10 +437,10 @@
; SSE-LABEL: select_fcmp_ule_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpnltsd %xmm0, %xmm1
-; SSE-NEXT: andpd %xmm1, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm1
-; SSE-NEXT: orpd %xmm2, %xmm1
-; SSE-NEXT: movapd %xmm1, %xmm0
+; SSE-NEXT: andps %xmm1, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm1
+; SSE-NEXT: orps %xmm2, %xmm1
+; SSE-NEXT: movaps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_ule_f64:
@@ -478,9 +478,9 @@
; SSE-LABEL: select_fcmp_une_f64:
; SSE: # BB#0:
; SSE-NEXT: cmpneqsd %xmm1, %xmm0
-; SSE-NEXT: andpd %xmm0, %xmm2
-; SSE-NEXT: andnpd %xmm3, %xmm0
-; SSE-NEXT: orpd %xmm2, %xmm0
+; SSE-NEXT: andps %xmm0, %xmm2
+; SSE-NEXT: andnps %xmm3, %xmm0
+; SSE-NEXT: orps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: select_fcmp_une_f64:
diff --git a/llvm/test/CodeGen/X86/fp-logic-replace.ll b/llvm/test/CodeGen/X86/fp-logic-replace.ll
index 8f2f649..47e0768 100644
--- a/llvm/test/CodeGen/X86/fp-logic-replace.ll
+++ b/llvm/test/CodeGen/X86/fp-logic-replace.ll
@@ -3,20 +3,20 @@
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX
; Test that we can replace "scalar" FP-bitwise-logic with the optimal instruction.
-; Scalar x86 FP-logic instructions only exist in your imagination and/or the bowels
+; Scalar x86 FP-logic instructions only exist in your imagination and/or the bowels
; of compilers, but float and double variants of FP-logic instructions are reality
-; and float may be a shorter instruction depending on which flavor of vector ISA
-; you have...so just prefer float all the time, ok? Yay, x86!
+; and float may be a shorter instruction depending on which flavor of vector ISA
+; you have...so just prefer float all the time, ok? Yay, x86!
define double @FsANDPSrr(double %x, double %y) {
; SSE-LABEL: FsANDPSrr:
; SSE: # BB#0:
-; SSE-NEXT: andpd %xmm1, %xmm0
+; SSE-NEXT: andps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: FsANDPSrr:
; AVX: # BB#0:
-; AVX-NEXT: vandpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vandps %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
%bc1 = bitcast double %x to i64
@@ -56,12 +56,12 @@
define double @FsORPSrr(double %x, double %y) {
; SSE-LABEL: FsORPSrr:
; SSE: # BB#0:
-; SSE-NEXT: orpd %xmm1, %xmm0
+; SSE-NEXT: orps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: FsORPSrr:
; AVX: # BB#0:
-; AVX-NEXT: vorpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vorps %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
%bc1 = bitcast double %x to i64
@@ -74,12 +74,12 @@
define double @FsXORPSrr(double %x, double %y) {
; SSE-LABEL: FsXORPSrr:
; SSE: # BB#0:
-; SSE-NEXT: xorpd %xmm1, %xmm0
+; SSE-NEXT: xorps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: FsXORPSrr:
; AVX: # BB#0:
-; AVX-NEXT: vxorpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vxorps %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
%bc1 = bitcast double %x to i64
diff --git a/llvm/test/CodeGen/X86/fp-logic.ll b/llvm/test/CodeGen/X86/fp-logic.ll
index 40bab00..2c6698f 100644
--- a/llvm/test/CodeGen/X86/fp-logic.ll
+++ b/llvm/test/CodeGen/X86/fp-logic.ll
@@ -223,7 +223,7 @@
define double @doubles(double %x, double %y) {
; CHECK-LABEL: doubles:
; CHECK: # BB#0:
-; CHECK-NEXT: andpd %xmm1, %xmm0
+; CHECK-NEXT: andps %xmm1, %xmm0
; CHECK-NEXT: retq
;
%bc1 = bitcast double %x to i64
diff --git a/llvm/test/CodeGen/X86/fp-select-cmp-and.ll b/llvm/test/CodeGen/X86/fp-select-cmp-and.ll
index e012809..c9c8922 100644
--- a/llvm/test/CodeGen/X86/fp-select-cmp-and.ll
+++ b/llvm/test/CodeGen/X86/fp-select-cmp-and.ll
@@ -5,7 +5,7 @@
; CHECK-LABEL: test1:
; CHECK: # BB#0:
; CHECK-NEXT: cmpltsd %xmm2, %xmm0
-; CHECK-NEXT: andpd %xmm1, %xmm0
+; CHECK-NEXT: andps %xmm1, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp olt double %a, %eps
@@ -17,7 +17,7 @@
; CHECK-LABEL: test2:
; CHECK: # BB#0:
; CHECK-NEXT: cmplesd %xmm2, %xmm0
-; CHECK-NEXT: andpd %xmm1, %xmm0
+; CHECK-NEXT: andps %xmm1, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp ole double %a, %eps
@@ -29,8 +29,8 @@
; CHECK-LABEL: test3:
; CHECK: # BB#0:
; CHECK-NEXT: cmpltsd %xmm0, %xmm2
-; CHECK-NEXT: andpd %xmm1, %xmm2
-; CHECK-NEXT: movapd %xmm2, %xmm0
+; CHECK-NEXT: andps %xmm1, %xmm2
+; CHECK-NEXT: movaps %xmm2, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp ogt double %a, %eps
@@ -42,8 +42,8 @@
; CHECK-LABEL: test4:
; CHECK: # BB#0:
; CHECK-NEXT: cmplesd %xmm0, %xmm2
-; CHECK-NEXT: andpd %xmm1, %xmm2
-; CHECK-NEXT: movapd %xmm2, %xmm0
+; CHECK-NEXT: andps %xmm1, %xmm2
+; CHECK-NEXT: movaps %xmm2, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp oge double %a, %eps
@@ -55,7 +55,7 @@
; CHECK-LABEL: test5:
; CHECK: # BB#0:
; CHECK-NEXT: cmpltsd %xmm2, %xmm0
-; CHECK-NEXT: andnpd %xmm1, %xmm0
+; CHECK-NEXT: andnps %xmm1, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp olt double %a, %eps
@@ -67,7 +67,7 @@
; CHECK-LABEL: test6:
; CHECK: # BB#0:
; CHECK-NEXT: cmplesd %xmm2, %xmm0
-; CHECK-NEXT: andnpd %xmm1, %xmm0
+; CHECK-NEXT: andnps %xmm1, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp ole double %a, %eps
@@ -79,8 +79,8 @@
; CHECK-LABEL: test7:
; CHECK: # BB#0:
; CHECK-NEXT: cmpltsd %xmm0, %xmm2
-; CHECK-NEXT: andnpd %xmm1, %xmm2
-; CHECK-NEXT: movapd %xmm2, %xmm0
+; CHECK-NEXT: andnps %xmm1, %xmm2
+; CHECK-NEXT: movaps %xmm2, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp ogt double %a, %eps
@@ -92,8 +92,8 @@
; CHECK-LABEL: test8:
; CHECK: # BB#0:
; CHECK-NEXT: cmplesd %xmm0, %xmm2
-; CHECK-NEXT: andnpd %xmm1, %xmm2
-; CHECK-NEXT: movapd %xmm2, %xmm0
+; CHECK-NEXT: andnps %xmm1, %xmm2
+; CHECK-NEXT: movaps %xmm2, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp oge double %a, %eps
@@ -220,10 +220,10 @@
; CHECK-LABEL: test18:
; CHECK: # BB#0:
; CHECK-NEXT: cmplesd %xmm0, %xmm3
-; CHECK-NEXT: andpd %xmm3, %xmm2
-; CHECK-NEXT: andnpd %xmm1, %xmm3
-; CHECK-NEXT: orpd %xmm2, %xmm3
-; CHECK-NEXT: movapd %xmm3, %xmm0
+; CHECK-NEXT: andps %xmm3, %xmm2
+; CHECK-NEXT: andnps %xmm1, %xmm3
+; CHECK-NEXT: orps %xmm2, %xmm3
+; CHECK-NEXT: movaps %xmm3, %xmm0
; CHECK-NEXT: retq
;
%cmp = fcmp oge double %a, %eps
diff --git a/llvm/test/CodeGen/X86/sse-minmax.ll b/llvm/test/CodeGen/X86/sse-minmax.ll
index c7a170d..b9d9e31 100644
--- a/llvm/test/CodeGen/X86/sse-minmax.ll
+++ b/llvm/test/CodeGen/X86/sse-minmax.ll
@@ -84,11 +84,11 @@
define double @oge(double %x, double %y) {
; STRICT-LABEL: oge:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm1, %xmm2
; STRICT-NEXT: cmplesd %xmm0, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm0
-; STRICT-NEXT: andnpd %xmm1, %xmm2
-; STRICT-NEXT: orpd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm0
+; STRICT-NEXT: andnps %xmm1, %xmm2
+; STRICT-NEXT: orps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: oge:
@@ -104,12 +104,12 @@
define double @ole(double %x, double %y) {
; STRICT-LABEL: ole:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm0, %xmm2
+; STRICT-NEXT: movaps %xmm0, %xmm2
; STRICT-NEXT: cmplesd %xmm1, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm0
-; STRICT-NEXT: andnpd %xmm1, %xmm2
-; STRICT-NEXT: orpd %xmm0, %xmm2
-; STRICT-NEXT: movapd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm0
+; STRICT-NEXT: andnps %xmm1, %xmm2
+; STRICT-NEXT: orps %xmm0, %xmm2
+; STRICT-NEXT: movaps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: ole:
@@ -125,12 +125,12 @@
define double @oge_inverse(double %x, double %y) {
; STRICT-LABEL: oge_inverse:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm1, %xmm2
; STRICT-NEXT: cmplesd %xmm0, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm2
-; STRICT-NEXT: orpd %xmm1, %xmm2
-; STRICT-NEXT: movapd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm1
+; STRICT-NEXT: andnps %xmm0, %xmm2
+; STRICT-NEXT: orps %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: oge_inverse:
@@ -152,12 +152,12 @@
define double @ole_inverse(double %x, double %y) {
; STRICT-LABEL: ole_inverse:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm0, %xmm2
+; STRICT-NEXT: movaps %xmm0, %xmm2
; STRICT-NEXT: cmplesd %xmm1, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm2
-; STRICT-NEXT: orpd %xmm1, %xmm2
-; STRICT-NEXT: movapd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm1
+; STRICT-NEXT: andnps %xmm0, %xmm2
+; STRICT-NEXT: orps %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: ole_inverse:
@@ -257,7 +257,7 @@
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm1, %xmm1
; STRICT-NEXT: cmplesd %xmm0, %xmm1
-; STRICT-NEXT: andpd %xmm1, %xmm0
+; STRICT-NEXT: andps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: oge_x:
@@ -275,10 +275,10 @@
; STRICT-LABEL: ole_x:
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm2, %xmm2
-; STRICT-NEXT: movapd %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm0, %xmm1
; STRICT-NEXT: cmplesd %xmm2, %xmm1
-; STRICT-NEXT: andpd %xmm0, %xmm1
-; STRICT-NEXT: movapd %xmm1, %xmm0
+; STRICT-NEXT: andps %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: ole_x:
@@ -297,8 +297,8 @@
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm1, %xmm1
; STRICT-NEXT: cmplesd %xmm0, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm1
-; STRICT-NEXT: movapd %xmm1, %xmm0
+; STRICT-NEXT: andnps %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: oge_inverse_x:
@@ -323,10 +323,10 @@
; STRICT-LABEL: ole_inverse_x:
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm2, %xmm2
-; STRICT-NEXT: movapd %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm0, %xmm1
; STRICT-NEXT: cmplesd %xmm2, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm1
-; STRICT-NEXT: movapd %xmm1, %xmm0
+; STRICT-NEXT: andnps %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: ole_inverse_x:
@@ -350,12 +350,12 @@
define double @ugt(double %x, double %y) {
; STRICT-LABEL: ugt:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm0, %xmm2
+; STRICT-NEXT: movaps %xmm0, %xmm2
; STRICT-NEXT: cmpnlesd %xmm1, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm0
-; STRICT-NEXT: andnpd %xmm1, %xmm2
-; STRICT-NEXT: orpd %xmm0, %xmm2
-; STRICT-NEXT: movapd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm0
+; STRICT-NEXT: andnps %xmm1, %xmm2
+; STRICT-NEXT: orps %xmm0, %xmm2
+; STRICT-NEXT: movaps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: ugt:
@@ -371,11 +371,11 @@
define double @ult(double %x, double %y) {
; STRICT-LABEL: ult:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm1, %xmm2
; STRICT-NEXT: cmpnlesd %xmm0, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm0
-; STRICT-NEXT: andnpd %xmm1, %xmm2
-; STRICT-NEXT: orpd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm0
+; STRICT-NEXT: andnps %xmm1, %xmm2
+; STRICT-NEXT: orps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: ult:
@@ -391,12 +391,12 @@
define double @ugt_inverse(double %x, double %y) {
; STRICT-LABEL: ugt_inverse:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm0, %xmm2
+; STRICT-NEXT: movaps %xmm0, %xmm2
; STRICT-NEXT: cmpnlesd %xmm1, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm2
-; STRICT-NEXT: orpd %xmm1, %xmm2
-; STRICT-NEXT: movapd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm1
+; STRICT-NEXT: andnps %xmm0, %xmm2
+; STRICT-NEXT: orps %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: ugt_inverse:
@@ -418,12 +418,12 @@
define double @ult_inverse(double %x, double %y) {
; STRICT-LABEL: ult_inverse:
; STRICT: # BB#0:
-; STRICT-NEXT: movapd %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm1, %xmm2
; STRICT-NEXT: cmpnlesd %xmm0, %xmm2
-; STRICT-NEXT: andpd %xmm2, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm2
-; STRICT-NEXT: orpd %xmm1, %xmm2
-; STRICT-NEXT: movapd %xmm2, %xmm0
+; STRICT-NEXT: andps %xmm2, %xmm1
+; STRICT-NEXT: andnps %xmm0, %xmm2
+; STRICT-NEXT: orps %xmm1, %xmm2
+; STRICT-NEXT: movaps %xmm2, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: ult_inverse:
@@ -524,10 +524,10 @@
; STRICT-LABEL: ugt_x:
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm2, %xmm2
-; STRICT-NEXT: movapd %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm0, %xmm1
; STRICT-NEXT: cmpnlesd %xmm2, %xmm1
-; STRICT-NEXT: andpd %xmm0, %xmm1
-; STRICT-NEXT: movapd %xmm1, %xmm0
+; STRICT-NEXT: andps %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: ugt_x:
@@ -546,7 +546,7 @@
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm1, %xmm1
; STRICT-NEXT: cmpnlesd %xmm0, %xmm1
-; STRICT-NEXT: andpd %xmm1, %xmm0
+; STRICT-NEXT: andps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; RELAX-LABEL: ult_x:
@@ -564,10 +564,10 @@
; STRICT-LABEL: ugt_inverse_x:
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm2, %xmm2
-; STRICT-NEXT: movapd %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm0, %xmm1
; STRICT-NEXT: cmpnlesd %xmm2, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm1
-; STRICT-NEXT: movapd %xmm1, %xmm0
+; STRICT-NEXT: andnps %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: ugt_inverse_x:
@@ -593,8 +593,8 @@
; STRICT: # BB#0:
; STRICT-NEXT: xorps %xmm1, %xmm1
; STRICT-NEXT: cmpnlesd %xmm0, %xmm1
-; STRICT-NEXT: andnpd %xmm0, %xmm1
-; STRICT-NEXT: movapd %xmm1, %xmm0
+; STRICT-NEXT: andnps %xmm0, %xmm1
+; STRICT-NEXT: movaps %xmm1, %xmm0
; STRICT-NEXT: retq
;
; UNSAFE-LABEL: ult_inverse_x: