[x86] split tests for FP and integer horizontal math

These are similar patterns, but when you throw AVX512 onto the pile,
the number of variations explodes. For FP, we really don't care about
AVX1 vs. AVX2 for FP ops. There may be some superficial shuffle diffs,
but that's not what we're testing for here, so I removed those RUNs.

Separating by type also lets us specify 'sse3' for the FP file vs. 'ssse3'
for the integer file...because x86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350357 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/haddsub-undef.ll b/test/CodeGen/X86/haddsub-undef.ll
index 83782b191260ee54b72e74ba9805d060b1bec8fe..416ce9154652cba8fadeefbfadb31750a21494cc 100644 (file)
--- a/test/CodeGen/X86/haddsub-undef.ll
+++ b/test/CodeGen/X86/haddsub-undef.ll
@@ -1,10 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3              | FileCheck %s --check-prefixes=SSE,SSE-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3,fast-hops    | FileCheck %s --check-prefixes=SSE,SSE-FAST
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx                | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX1,AVX1-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx,fast-hops      | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX1,AVX1-FAST
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2               | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,fast-hops     | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse3               | FileCheck %s --check-prefixes=SSE,SSE-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse3,fast-hops     | FileCheck %s --check-prefixes=SSE,SSE-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx                | FileCheck %s --check-prefixes=AVX,AVX-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx,fast-hops      | FileCheck %s --check-prefixes=AVX,AVX-FAST
 
 ; Verify that we correctly fold horizontal binop even in the presence of UNDEFs.
 
@@ -316,141 +314,6 @@ define <8 x float> @test13_undef(<8 x float> %a, <8 x float> %b) {
   ret <8 x float> %vecinit4
 }
 
-define <8 x i32> @test14_undef(<8 x i32> %a, <8 x i32> %b) {
-; SSE-LABEL: test14_undef:
-; SSE:       # %bb.0:
-; SSE-NEXT:    phaddd %xmm2, %xmm0
-; SSE-NEXT:    retq
-;
-; AVX1-LABEL: test14_undef:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test14_undef:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    retq
-  %vecext = extractelement <8 x i32> %a, i32 0
-  %vecext1 = extractelement <8 x i32> %a, i32 1
-  %add = add i32 %vecext, %vecext1
-  %vecinit = insertelement <8 x i32> undef, i32 %add, i32 0
-  %vecext2 = extractelement <8 x i32> %b, i32 2
-  %vecext3 = extractelement <8 x i32> %b, i32 3
-  %add4 = add i32 %vecext2, %vecext3
-  %vecinit5 = insertelement <8 x i32> %vecinit, i32 %add4, i32 3
-  ret <8 x i32> %vecinit5
-}
-
-; integer horizontal adds instead of two scalar adds followed by vector inserts.
-define <8 x i32> @test15_undef(<8 x i32> %a, <8 x i32> %b) {
-; SSE-LABEL: test15_undef:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movd %xmm0, %eax
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE-NEXT:    movd %xmm0, %ecx
-; SSE-NEXT:    addl %eax, %ecx
-; SSE-NEXT:    movd %xmm3, %eax
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE-NEXT:    movd %xmm0, %edx
-; SSE-NEXT:    addl %eax, %edx
-; SSE-NEXT:    movd %ecx, %xmm0
-; SSE-NEXT:    movd %edx, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; SSE-NEXT:    retq
-;
-; AVX1-LABEL: test15_undef:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovd %xmm0, %eax
-; AVX1-NEXT:    vpextrd $1, %xmm0, %ecx
-; AVX1-NEXT:    addl %eax, %ecx
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
-; AVX1-NEXT:    vpextrd $1, %xmm0, %edx
-; AVX1-NEXT:    addl %eax, %edx
-; AVX1-NEXT:    vmovd %ecx, %xmm0
-; AVX1-NEXT:    vmovd %edx, %xmm1
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test15_undef:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vphaddd %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    retq
-  %vecext = extractelement <8 x i32> %a, i32 0
-  %vecext1 = extractelement <8 x i32> %a, i32 1
-  %add = add i32 %vecext, %vecext1
-  %vecinit = insertelement <8 x i32> undef, i32 %add, i32 0
-  %vecext2 = extractelement <8 x i32> %b, i32 4
-  %vecext3 = extractelement <8 x i32> %b, i32 5
-  %add4 = add i32 %vecext2, %vecext3
-  %vecinit5 = insertelement <8 x i32> %vecinit, i32 %add4, i32 6
-  ret <8 x i32> %vecinit5
-}
-
-define <8 x i32> @test16_undef(<8 x i32> %a, <8 x i32> %b) {
-; SSE-LABEL: test16_undef:
-; SSE:       # %bb.0:
-; SSE-NEXT:    phaddd %xmm0, %xmm0
-; SSE-NEXT:    retq
-;
-; AVX1-LABEL: test16_undef:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test16_undef:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vphaddd %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    retq
-  %vecext = extractelement <8 x i32> %a, i32 0
-  %vecext1 = extractelement <8 x i32> %a, i32 1
-  %add = add i32 %vecext, %vecext1
-  %vecinit = insertelement <8 x i32> undef, i32 %add, i32 0
-  %vecext2 = extractelement <8 x i32> %a, i32 2
-  %vecext3 = extractelement <8 x i32> %a, i32 3
-  %add4 = add i32 %vecext2, %vecext3
-  %vecinit5 = insertelement <8 x i32> %vecinit, i32 %add4, i32 1
-  ret <8 x i32> %vecinit5
-}
-
-define <8 x i32> @test17_undef(<8 x i32> %a, <8 x i32> %b) {
-; SSE-LABEL: test17_undef:
-; SSE:       # %bb.0:
-; SSE-NEXT:    phaddd %xmm1, %xmm0
-; SSE-NEXT:    retq
-;
-; AVX1-LABEL: test17_undef:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test17_undef:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    retq
-  %vecext = extractelement <8 x i32> %a, i32 0
-  %vecext1 = extractelement <8 x i32> %a, i32 1
-  %add1 = add i32 %vecext, %vecext1
-  %vecinit1 = insertelement <8 x i32> undef, i32 %add1, i32 0
-  %vecext2 = extractelement <8 x i32> %a, i32 2
-  %vecext3 = extractelement <8 x i32> %a, i32 3
-  %add2 = add i32 %vecext2, %vecext3
-  %vecinit2 = insertelement <8 x i32> %vecinit1, i32 %add2, i32 1
-  %vecext4 = extractelement <8 x i32> %a, i32 4
-  %vecext5 = extractelement <8 x i32> %a, i32 5
-  %add3 = add i32 %vecext4, %vecext5
-  %vecinit3 = insertelement <8 x i32> %vecinit2, i32 %add3, i32 2
-  %vecext6 = extractelement <8 x i32> %a, i32 6
-  %vecext7 = extractelement <8 x i32> %a, i32 7
-  %add4 = add i32 %vecext6, %vecext7
-  %vecinit4 = insertelement <8 x i32> %vecinit3, i32 %add4, i32 3
-  ret <8 x i32> %vecinit4
-}
-
 define <2 x double> @add_pd_003(<2 x double> %x) {
 ; SSE-SLOW-LABEL: add_pd_003:
 ; SSE-SLOW:       # %bb.0:
@@ -621,18 +484,23 @@ define <4 x float> @add_ps_007_2(<4 x float> %x) {
 ; SSE-FAST-NEXT:    haddps %xmm0, %xmm0
 ; SSE-FAST-NEXT:    retq
 ;
+; AVX-SLOW-LABEL: add_ps_007_2:
+; AVX-SLOW:       # %bb.0:
+; AVX-SLOW-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
+; AVX-SLOW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; AVX-SLOW-NEXT:    vaddps %xmm0, %xmm1, %xmm0
+; AVX-SLOW-NEXT:    retq
+;
+; AVX-FAST-LABEL: add_ps_007_2:
+; AVX-FAST:       # %bb.0:
+; AVX-FAST-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
+; AVX-FAST-NEXT:    retq
 ; AVX1-SLOW-LABEL: add_ps_007_2:
 ; AVX1-SLOW:       # %bb.0:
 ; AVX1-SLOW-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
 ; AVX1-SLOW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,1,1,3]
 ; AVX1-SLOW-NEXT:    vaddps %xmm0, %xmm1, %xmm0
 ; AVX1-SLOW-NEXT:    retq
-;
-; AVX-FAST-LABEL: add_ps_007_2:
-; AVX-FAST:       # %bb.0:
-; AVX-FAST-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
-; AVX-FAST-NEXT:    retq
-;
 ; AVX2-SLOW-LABEL: add_ps_007_2:
 ; AVX2-SLOW:       # %bb.0:
 ; AVX2-SLOW-NEXT:    vbroadcastss %xmm0, %xmm1
@@ -720,6 +588,19 @@ define <4 x float> @add_ps_018(<4 x float> %x) {
 ; SSE-FAST-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; SSE-FAST-NEXT:    retq
 ;
+; AVX-SLOW-LABEL: add_ps_018:
+; AVX-SLOW:       # %bb.0:
+; AVX-SLOW-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
+; AVX-SLOW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; AVX-SLOW-NEXT:    vaddps %xmm0, %xmm1, %xmm0
+; AVX-SLOW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX-SLOW-NEXT:    retq
+;
+; AVX-FAST-LABEL: add_ps_018:
+; AVX-FAST:       # %bb.0:
+; AVX-FAST-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
+; AVX-FAST-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX-FAST-NEXT:    retq
 ; AVX1-SLOW-LABEL: add_ps_018:
 ; AVX1-SLOW:       # %bb.0:
 ; AVX1-SLOW-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
@@ -727,13 +608,6 @@ define <4 x float> @add_ps_018(<4 x float> %x) {
 ; AVX1-SLOW-NEXT:    vaddps %xmm0, %xmm1, %xmm0
 ; AVX1-SLOW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; AVX1-SLOW-NEXT:    retq
-;
-; AVX-FAST-LABEL: add_ps_018:
-; AVX-FAST:       # %bb.0:
-; AVX-FAST-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
-; AVX-FAST-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX-FAST-NEXT:    retq
-;
 ; AVX2-SLOW-LABEL: add_ps_018:
 ; AVX2-SLOW:       # %bb.0:
 ; AVX2-SLOW-NEXT:    vbroadcastss %xmm0, %xmm1

diff --git a/test/CodeGen/X86/phaddsub-undef.ll b/test/CodeGen/X86/phaddsub-undef.ll
new file mode 100644 (file)
index 0000000..161d057
--- /dev/null
+++ b/test/CodeGen/X86/phaddsub-undef.ll
@@ -0,0 +1,145 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3              | FileCheck %s --check-prefixes=SSE,SSE-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3,fast-hops    | FileCheck %s --check-prefixes=SSE,SSE-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx                | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX1,AVX1-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx,fast-hops      | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX1,AVX1-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2               | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX2,AVX2-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,fast-hops     | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX2,AVX2-FAST
+
+; Verify that we correctly fold horizontal binop even in the presence of UNDEFs.
+
+define <8 x i32> @test14_undef(<8 x i32> %a, <8 x i32> %b) {
+; SSE-LABEL: test14_undef:
+; SSE:       # %bb.0:
+; SSE-NEXT:    phaddd %xmm2, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: test14_undef:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: test14_undef:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    retq
+  %vecext = extractelement <8 x i32> %a, i32 0
+  %vecext1 = extractelement <8 x i32> %a, i32 1
+  %add = add i32 %vecext, %vecext1
+  %vecinit = insertelement <8 x i32> undef, i32 %add, i32 0
+  %vecext2 = extractelement <8 x i32> %b, i32 2
+  %vecext3 = extractelement <8 x i32> %b, i32 3
+  %add4 = add i32 %vecext2, %vecext3
+  %vecinit5 = insertelement <8 x i32> %vecinit, i32 %add4, i32 3
+  ret <8 x i32> %vecinit5
+}
+
+; integer horizontal adds instead of two scalar adds followed by vector inserts.
+define <8 x i32> @test15_undef(<8 x i32> %a, <8 x i32> %b) {
+; SSE-LABEL: test15_undef:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movd %xmm0, %eax
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE-NEXT:    movd %xmm0, %ecx
+; SSE-NEXT:    addl %eax, %ecx
+; SSE-NEXT:    movd %xmm3, %eax
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
+; SSE-NEXT:    movd %xmm0, %edx
+; SSE-NEXT:    addl %eax, %edx
+; SSE-NEXT:    movd %ecx, %xmm0
+; SSE-NEXT:    movd %edx, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: test15_undef:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vpextrd $1, %xmm0, %ecx
+; AVX1-NEXT:    addl %eax, %ecx
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm0
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vpextrd $1, %xmm0, %edx
+; AVX1-NEXT:    addl %eax, %edx
+; AVX1-NEXT:    vmovd %ecx, %xmm0
+; AVX1-NEXT:    vmovd %edx, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: test15_undef:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vphaddd %ymm0, %ymm0, %ymm0
+; AVX2-NEXT:    retq
+  %vecext = extractelement <8 x i32> %a, i32 0
+  %vecext1 = extractelement <8 x i32> %a, i32 1
+  %add = add i32 %vecext, %vecext1
+  %vecinit = insertelement <8 x i32> undef, i32 %add, i32 0
+  %vecext2 = extractelement <8 x i32> %b, i32 4
+  %vecext3 = extractelement <8 x i32> %b, i32 5
+  %add4 = add i32 %vecext2, %vecext3
+  %vecinit5 = insertelement <8 x i32> %vecinit, i32 %add4, i32 6
+  ret <8 x i32> %vecinit5
+}
+
+define <8 x i32> @test16_undef(<8 x i32> %a, <8 x i32> %b) {
+; SSE-LABEL: test16_undef:
+; SSE:       # %bb.0:
+; SSE-NEXT:    phaddd %xmm0, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: test16_undef:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: test16_undef:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vphaddd %ymm0, %ymm0, %ymm0
+; AVX2-NEXT:    retq
+  %vecext = extractelement <8 x i32> %a, i32 0
+  %vecext1 = extractelement <8 x i32> %a, i32 1
+  %add = add i32 %vecext, %vecext1
+  %vecinit = insertelement <8 x i32> undef, i32 %add, i32 0
+  %vecext2 = extractelement <8 x i32> %a, i32 2
+  %vecext3 = extractelement <8 x i32> %a, i32 3
+  %add4 = add i32 %vecext2, %vecext3
+  %vecinit5 = insertelement <8 x i32> %vecinit, i32 %add4, i32 1
+  ret <8 x i32> %vecinit5
+}
+
+define <8 x i32> @test17_undef(<8 x i32> %a, <8 x i32> %b) {
+; SSE-LABEL: test17_undef:
+; SSE:       # %bb.0:
+; SSE-NEXT:    phaddd %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: test17_undef:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: test17_undef:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    retq
+  %vecext = extractelement <8 x i32> %a, i32 0
+  %vecext1 = extractelement <8 x i32> %a, i32 1
+  %add1 = add i32 %vecext, %vecext1
+  %vecinit1 = insertelement <8 x i32> undef, i32 %add1, i32 0
+  %vecext2 = extractelement <8 x i32> %a, i32 2
+  %vecext3 = extractelement <8 x i32> %a, i32 3
+  %add2 = add i32 %vecext2, %vecext3
+  %vecinit2 = insertelement <8 x i32> %vecinit1, i32 %add2, i32 1
+  %vecext4 = extractelement <8 x i32> %a, i32 4
+  %vecext5 = extractelement <8 x i32> %a, i32 5
+  %add3 = add i32 %vecext4, %vecext5
+  %vecinit3 = insertelement <8 x i32> %vecinit2, i32 %add3, i32 2
+  %vecext6 = extractelement <8 x i32> %a, i32 6
+  %vecext7 = extractelement <8 x i32> %a, i32 7
+  %add4 = add i32 %vecext6, %vecext7
+  %vecinit4 = insertelement <8 x i32> %vecinit3, i32 %add4, i32 3
+  ret <8 x i32> %vecinit4
+}
+