[llvm] r294845 - [X86][3DNow!] Add tests showing missed commutation opportunities.
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sat Feb 11 05:00:32 PST 2017
Author: rksimon
Date: Sat Feb 11 07:00:32 2017
New Revision: 294845
URL: http://llvm.org/viewvc/llvm-project?rev=294845&view=rev
Log:
[X86][3DNow!] Add tests showing missed commutation opportunities.
Added:
llvm/trunk/test/CodeGen/X86/commute-3dnow.ll
Added: llvm/trunk/test/CodeGen/X86/commute-3dnow.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/commute-3dnow.ll?rev=294845&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/commute-3dnow.ll (added)
+++ llvm/trunk/test/CodeGen/X86/commute-3dnow.ll Sat Feb 11 07:00:32 2017
@@ -0,0 +1,222 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+mmx,+3dnow | FileCheck %s --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+mmx,+3dnow | FileCheck %s --check-prefix=X64
+
+; FIXME - missed commutation opportunities.
+
+define void @commute_m_pfadd(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pfadd:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pfadd (%eax), %mm0
+; X32-NEXT: pfadd %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pfadd:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pfadd (%rsi), %mm0
+; X64-NEXT: pfadd %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pfadd(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pfadd(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pfadd(x86_mmx, x86_mmx)
+
+define void @commute_m_pfsub(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pfsub:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pfsub (%eax), %mm0
+; X32-NEXT: pfsub %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pfsub:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pfsub (%rsi), %mm0
+; X64-NEXT: pfsub %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pfsub(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pfsub(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pfsub(x86_mmx, x86_mmx)
+
+define void @commute_m_pfsubr(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pfsubr:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pfsubr (%eax), %mm0
+; X32-NEXT: pfsubr %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pfsubr:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pfsubr (%rsi), %mm0
+; X64-NEXT: pfsubr %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pfsubr(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pfsubr(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pfsubr(x86_mmx, x86_mmx)
+
+define void @commute_m_pfmul(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pfmul:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pfmul (%eax), %mm0
+; X32-NEXT: pfmul %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pfmul:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pfmul (%rsi), %mm0
+; X64-NEXT: pfmul %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pfmul(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pfmul(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pfmul(x86_mmx, x86_mmx)
+
+define void @commute_m_pfcmpeq(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pfcmpeq:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pfcmpeq (%eax), %mm0
+; X32-NEXT: pfcmpeq %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pfcmpeq:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pfcmpeq (%rsi), %mm0
+; X64-NEXT: pfcmpeq %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pfcmpeq(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pfcmpeq(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pfcmpeq(x86_mmx, x86_mmx)
+
+define void @commute_m_pavgusb(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pavgusb:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pavgusb (%eax), %mm0
+; X32-NEXT: pavgusb %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pavgusb:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pavgusb (%rsi), %mm0
+; X64-NEXT: pavgusb %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pavgusb(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pavgusb(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pavgusb(x86_mmx, x86_mmx)
+
+define void @commute_m_pmulhrw(x86_mmx *%a0, x86_mmx *%a1, x86_mmx *%a2) nounwind {
+; X32-LABEL: commute_m_pmulhrw:
+; X32: # BB#0:
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movq (%edx), %mm0
+; X32-NEXT: movq (%ecx), %mm1
+; X32-NEXT: pmulhrw (%eax), %mm0
+; X32-NEXT: pmulhrw %mm0, %mm1
+; X32-NEXT: movq %mm1, (%ecx)
+; X32-NEXT: retl
+;
+; X64-LABEL: commute_m_pmulhrw:
+; X64: # BB#0:
+; X64-NEXT: movq (%rdi), %mm0
+; X64-NEXT: movq (%rdx), %mm1
+; X64-NEXT: pmulhrw (%rsi), %mm0
+; X64-NEXT: pmulhrw %mm0, %mm1
+; X64-NEXT: movq %mm1, (%rdx)
+; X64-NEXT: retq
+ %1 = load x86_mmx, x86_mmx* %a0
+ %2 = load x86_mmx, x86_mmx* %a1
+ %3 = load x86_mmx, x86_mmx* %a2
+ %4 = tail call x86_mmx @llvm.x86.3dnow.pmulhrw(x86_mmx %1, x86_mmx %2)
+ %5 = tail call x86_mmx @llvm.x86.3dnow.pmulhrw(x86_mmx %3, x86_mmx %4)
+ store x86_mmx %5, x86_mmx* %a2
+ ret void
+}
+declare x86_mmx @llvm.x86.3dnow.pmulhrw(x86_mmx, x86_mmx)
More information about the llvm-commits
mailing list