[llvm] r339562 - [X86] Add tests showing missing div/rem 0, X -> 0 combines
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 13 06:29:54 PDT 2018
Author: rksimon
Date: Mon Aug 13 06:29:54 2018
New Revision: 339562
URL: http://llvm.org/viewvc/llvm-project?rev=339562&view=rev
Log:
[X86] Add tests showing missing div/rem 0, X -> 0 combines
Modified:
llvm/trunk/test/CodeGen/X86/combine-sdiv.ll
llvm/trunk/test/CodeGen/X86/combine-srem.ll
llvm/trunk/test/CodeGen/X86/combine-udiv.ll
llvm/trunk/test/CodeGen/X86/combine-urem.ll
Modified: llvm/trunk/test/CodeGen/X86/combine-sdiv.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/combine-sdiv.ll?rev=339562&r1=339561&r2=339562&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/combine-sdiv.ll (original)
+++ llvm/trunk/test/CodeGen/X86/combine-sdiv.ll Mon Aug 13 06:29:54 2018
@@ -106,6 +106,73 @@ define <4 x i32> @combine_vec_sdiv_by_mi
ret <4 x i32> %1
}
+; TODO fold (sdiv 0, x) -> 0
+define i32 @combine_sdiv_zero(i32 %x) {
+; CHECK-LABEL: combine_sdiv_zero:
+; CHECK: # %bb.0:
+; CHECK-NEXT: xorl %eax, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: idivl %edi
+; CHECK-NEXT: retq
+ %1 = sdiv i32 0, %x
+ ret i32 %1
+}
+
+define <4 x i32> @combine_vec_sdiv_zero(<4 x i32> %x) {
+; SSE-LABEL: combine_vec_sdiv_zero:
+; SSE: # %bb.0:
+; SSE-NEXT: pextrd $1, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %ecx
+; SSE-NEXT: movl %eax, %ecx
+; SSE-NEXT: movd %xmm0, %esi
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %esi
+; SSE-NEXT: movd %eax, %xmm1
+; SSE-NEXT: pinsrd $1, %ecx, %xmm1
+; SSE-NEXT: pextrd $2, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %ecx
+; SSE-NEXT: pinsrd $2, %eax, %xmm1
+; SSE-NEXT: pextrd $3, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %ecx
+; SSE-NEXT: pinsrd $3, %eax, %xmm1
+; SSE-NEXT: movdqa %xmm1, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: combine_vec_sdiv_zero:
+; AVX: # %bb.0:
+; AVX-NEXT: vpextrd $1, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %ecx
+; AVX-NEXT: movl %eax, %ecx
+; AVX-NEXT: vmovd %xmm0, %esi
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %esi
+; AVX-NEXT: vmovd %eax, %xmm1
+; AVX-NEXT: vpinsrd $1, %ecx, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $2, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %ecx
+; AVX-NEXT: vpinsrd $2, %eax, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $3, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %ecx
+; AVX-NEXT: vpinsrd $3, %eax, %xmm1, %xmm0
+; AVX-NEXT: retq
+ %1 = sdiv <4 x i32> zeroinitializer, %x
+ ret <4 x i32> %1
+}
+
; TODO fold (sdiv x, x) -> 1
define i32 @combine_sdiv_dupe(i32 %x) {
; CHECK-LABEL: combine_sdiv_dupe:
Modified: llvm/trunk/test/CodeGen/X86/combine-srem.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/combine-srem.ll?rev=339562&r1=339561&r2=339562&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/combine-srem.ll (original)
+++ llvm/trunk/test/CodeGen/X86/combine-srem.ll Mon Aug 13 06:29:54 2018
@@ -100,6 +100,74 @@ define <4 x i32> @combine_vec_srem_by_mi
ret <4 x i32> %1
}
+; TODO fold (srem 0, x) -> 0
+define i32 @combine_srem_zero(i32 %x) {
+; CHECK-LABEL: combine_srem_zero:
+; CHECK: # %bb.0:
+; CHECK-NEXT: xorl %eax, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: idivl %edi
+; CHECK-NEXT: movl %edx, %eax
+; CHECK-NEXT: retq
+ %1 = srem i32 0, %x
+ ret i32 %1
+}
+
+define <4 x i32> @combine_vec_srem_zero(<4 x i32> %x) {
+; SSE-LABEL: combine_vec_srem_zero:
+; SSE: # %bb.0:
+; SSE-NEXT: pextrd $1, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %ecx
+; SSE-NEXT: movl %edx, %ecx
+; SSE-NEXT: movd %xmm0, %esi
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %esi
+; SSE-NEXT: movd %edx, %xmm1
+; SSE-NEXT: pinsrd $1, %ecx, %xmm1
+; SSE-NEXT: pextrd $2, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %ecx
+; SSE-NEXT: pinsrd $2, %edx, %xmm1
+; SSE-NEXT: pextrd $3, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: idivl %ecx
+; SSE-NEXT: pinsrd $3, %edx, %xmm1
+; SSE-NEXT: movdqa %xmm1, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: combine_vec_srem_zero:
+; AVX: # %bb.0:
+; AVX-NEXT: vpextrd $1, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %ecx
+; AVX-NEXT: movl %edx, %ecx
+; AVX-NEXT: vmovd %xmm0, %esi
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %esi
+; AVX-NEXT: vmovd %edx, %xmm1
+; AVX-NEXT: vpinsrd $1, %ecx, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $2, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %ecx
+; AVX-NEXT: vpinsrd $2, %edx, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $3, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: idivl %ecx
+; AVX-NEXT: vpinsrd $3, %edx, %xmm1, %xmm0
+; AVX-NEXT: retq
+ %1 = srem <4 x i32> zeroinitializer, %x
+ ret <4 x i32> %1
+}
+
; TODO fold (srem x, x) -> 0
define i32 @combine_srem_dupe(i32 %x) {
; CHECK-LABEL: combine_srem_dupe:
Modified: llvm/trunk/test/CodeGen/X86/combine-udiv.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/combine-udiv.ll?rev=339562&r1=339561&r2=339562&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/combine-udiv.ll (original)
+++ llvm/trunk/test/CodeGen/X86/combine-udiv.ll Mon Aug 13 06:29:54 2018
@@ -76,6 +76,73 @@ define <4 x i32> @combine_vec_udiv_by_mi
ret <4 x i32> %1
}
+; TODO fold (udiv 0, x) -> 0
+define i32 @combine_udiv_zero(i32 %x) {
+; CHECK-LABEL: combine_udiv_zero:
+; CHECK: # %bb.0:
+; CHECK-NEXT: xorl %eax, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %edi
+; CHECK-NEXT: retq
+ %1 = udiv i32 0, %x
+ ret i32 %1
+}
+
+define <4 x i32> @combine_vec_udiv_zero(<4 x i32> %x) {
+; SSE-LABEL: combine_vec_udiv_zero:
+; SSE: # %bb.0:
+; SSE-NEXT: pextrd $1, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %ecx
+; SSE-NEXT: movl %eax, %ecx
+; SSE-NEXT: movd %xmm0, %esi
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %esi
+; SSE-NEXT: movd %eax, %xmm1
+; SSE-NEXT: pinsrd $1, %ecx, %xmm1
+; SSE-NEXT: pextrd $2, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %ecx
+; SSE-NEXT: pinsrd $2, %eax, %xmm1
+; SSE-NEXT: pextrd $3, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %ecx
+; SSE-NEXT: pinsrd $3, %eax, %xmm1
+; SSE-NEXT: movdqa %xmm1, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: combine_vec_udiv_zero:
+; AVX: # %bb.0:
+; AVX-NEXT: vpextrd $1, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %ecx
+; AVX-NEXT: movl %eax, %ecx
+; AVX-NEXT: vmovd %xmm0, %esi
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %esi
+; AVX-NEXT: vmovd %eax, %xmm1
+; AVX-NEXT: vpinsrd $1, %ecx, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $2, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %ecx
+; AVX-NEXT: vpinsrd $2, %eax, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $3, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %ecx
+; AVX-NEXT: vpinsrd $3, %eax, %xmm1, %xmm0
+; AVX-NEXT: retq
+ %1 = udiv <4 x i32> zeroinitializer, %x
+ ret <4 x i32> %1
+}
+
; TODO fold (udiv x, x) -> 1
define i32 @combine_udiv_dupe(i32 %x) {
; CHECK-LABEL: combine_udiv_dupe:
Modified: llvm/trunk/test/CodeGen/X86/combine-urem.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/combine-urem.ll?rev=339562&r1=339561&r2=339562&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/combine-urem.ll (original)
+++ llvm/trunk/test/CodeGen/X86/combine-urem.ll Mon Aug 13 06:29:54 2018
@@ -89,6 +89,74 @@ define <4 x i32> @combine_vec_urem_by_mi
ret <4 x i32> %1
}
+; TODO fold (urem 0, x) -> 0
+define i32 @combine_urem_zero(i32 %x) {
+; CHECK-LABEL: combine_urem_zero:
+; CHECK: # %bb.0:
+; CHECK-NEXT: xorl %eax, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %edi
+; CHECK-NEXT: movl %edx, %eax
+; CHECK-NEXT: retq
+ %1 = urem i32 0, %x
+ ret i32 %1
+}
+
+define <4 x i32> @combine_vec_urem_zero(<4 x i32> %x) {
+; SSE-LABEL: combine_vec_urem_zero:
+; SSE: # %bb.0:
+; SSE-NEXT: pextrd $1, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %ecx
+; SSE-NEXT: movl %edx, %ecx
+; SSE-NEXT: movd %xmm0, %esi
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %esi
+; SSE-NEXT: movd %edx, %xmm1
+; SSE-NEXT: pinsrd $1, %ecx, %xmm1
+; SSE-NEXT: pextrd $2, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %ecx
+; SSE-NEXT: pinsrd $2, %edx, %xmm1
+; SSE-NEXT: pextrd $3, %xmm0, %ecx
+; SSE-NEXT: xorl %eax, %eax
+; SSE-NEXT: xorl %edx, %edx
+; SSE-NEXT: divl %ecx
+; SSE-NEXT: pinsrd $3, %edx, %xmm1
+; SSE-NEXT: movdqa %xmm1, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: combine_vec_urem_zero:
+; AVX: # %bb.0:
+; AVX-NEXT: vpextrd $1, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %ecx
+; AVX-NEXT: movl %edx, %ecx
+; AVX-NEXT: vmovd %xmm0, %esi
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %esi
+; AVX-NEXT: vmovd %edx, %xmm1
+; AVX-NEXT: vpinsrd $1, %ecx, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $2, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %ecx
+; AVX-NEXT: vpinsrd $2, %edx, %xmm1, %xmm1
+; AVX-NEXT: vpextrd $3, %xmm0, %ecx
+; AVX-NEXT: xorl %eax, %eax
+; AVX-NEXT: xorl %edx, %edx
+; AVX-NEXT: divl %ecx
+; AVX-NEXT: vpinsrd $3, %edx, %xmm1, %xmm0
+; AVX-NEXT: retq
+ %1 = urem <4 x i32> zeroinitializer, %x
+ ret <4 x i32> %1
+}
+
; TODO fold (urem x, x) -> 0
define i32 @combine_urem_dupe(i32 %x) {
; CHECK-LABEL: combine_urem_dupe:
More information about the llvm-commits
mailing list