From: Craig Topper <craig.topper@intel.com>
Date: Tue, 4 Jun 2019 21:47:50 +0000 (+0000)
Subject: [X86] Add avx512bw to the avx512 machine-combiner-int-vec.ll to ensure we use zmm... 
X-Git-Tag: llvmorg-10-init~3758
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=1648cb17e40e2579a1a54df35639002f2acf7fe5;p=platform%2Fupstream%2Fllvm.git

[X86] Add avx512bw to the avx512 machine-combiner-int-vec.ll to ensure we use zmm for v32i16/v64i8.

llvm-svn: 362552
---

diff --git a/llvm/test/CodeGen/X86/machine-combiner-int-vec.ll b/llvm/test/CodeGen/X86/machine-combiner-int-vec.ll
index 6a1385a..52c4cd6 100644
--- a/llvm/test/CodeGen/X86/machine-combiner-int-vec.ll
+++ b/llvm/test/CodeGen/X86/machine-combiner-int-vec.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=sse2 -machine-combiner-verify-pattern-order=true < %s | FileCheck %s --check-prefix=SSE
 ; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=avx2 -machine-combiner-verify-pattern-order=true < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
-; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=avx512vl -machine-combiner-verify-pattern-order=true < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX512
+; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=avx512vl,avx512bw -machine-combiner-verify-pattern-order=true < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX512
 
 ; Verify that 128-bit vector logical ops are reassociated.
 
@@ -1710,15 +1710,22 @@ define <64 x i8> @reassociate_umax_v64i8(<64 x i8> %x0, <64 x i8> %x1, <64 x i8>
 ; SSE-NEXT:    pmaxub {{[0-9]+}}(%rsp), %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_umax_v64i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpmaxub %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpmaxub %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpmaxub %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpmaxub %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_umax_v64i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpmaxub %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpmaxub %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpmaxub %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpmaxub %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_umax_v64i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpmaxub %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpmaxub %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <64 x i8> %x0, %x1
   %t1 = icmp ugt <64 x i8> %x2, %t0
@@ -1770,15 +1777,22 @@ define <32 x i16> @reassociate_umax_v32i16(<32 x i16> %x0, <32 x i16> %x1, <32 x
 ; SSE-NEXT:    pxor %xmm4, %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_umax_v32i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpmaxuw %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpmaxuw %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpmaxuw %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpmaxuw %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_umax_v32i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpmaxuw %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpmaxuw %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpmaxuw %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpmaxuw %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_umax_v32i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddw %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpmaxuw %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpmaxuw %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <32 x i16> %x0, %x1
   %t1 = icmp ugt <32 x i16> %x2, %t0
@@ -2129,15 +2143,22 @@ define <64 x i8> @reassociate_smax_v64i8(<64 x i8> %x0, <64 x i8> %x1, <64 x i8>
 ; SSE-NEXT:    por %xmm8, %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_smax_v64i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpmaxsb %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpmaxsb %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpmaxsb %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpmaxsb %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_smax_v64i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpmaxsb %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpmaxsb %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpmaxsb %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpmaxsb %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_smax_v64i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpmaxsb %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpmaxsb %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <64 x i8> %x0, %x1
   %t1 = icmp sgt <64 x i8> %x2, %t0
@@ -2164,15 +2185,22 @@ define <32 x i16> @reassociate_smax_v32i16(<32 x i16> %x0, <32 x i16> %x1, <32 x
 ; SSE-NEXT:    pmaxsw {{[0-9]+}}(%rsp), %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_smax_v32i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpmaxsw %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpmaxsw %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpmaxsw %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpmaxsw %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_smax_v32i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpmaxsw %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpmaxsw %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpmaxsw %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpmaxsw %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_smax_v32i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddw %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpmaxsw %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpmaxsw %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <32 x i16> %x0, %x1
   %t1 = icmp sgt <32 x i16> %x2, %t0
@@ -2447,15 +2475,22 @@ define <64 x i8> @reassociate_umin_v64i8(<64 x i8> %x0, <64 x i8> %x1, <64 x i8>
 ; SSE-NEXT:    pminub {{[0-9]+}}(%rsp), %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_umin_v64i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpminub %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpminub %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpminub %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpminub %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_umin_v64i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpminub %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpminub %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpminub %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpminub %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_umin_v64i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpminub %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpminub %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <64 x i8> %x0, %x1
   %t1 = icmp ult <64 x i8> %x2, %t0
@@ -2507,15 +2542,22 @@ define <32 x i16> @reassociate_umin_v32i16(<32 x i16> %x0, <32 x i16> %x1, <32 x
 ; SSE-NEXT:    pxor %xmm4, %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_umin_v32i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpminuw %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpminuw %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpminuw %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpminuw %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_umin_v32i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpminuw %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpminuw %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpminuw %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpminuw %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_umin_v32i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddw %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpminuw %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpminuw %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <32 x i16> %x0, %x1
   %t1 = icmp ult <32 x i16> %x2, %t0
@@ -2863,15 +2905,22 @@ define <64 x i8> @reassociate_smin_v64i8(<64 x i8> %x0, <64 x i8> %x1, <64 x i8>
 ; SSE-NEXT:    por %xmm8, %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_smin_v64i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpminsb %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpminsb %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpminsb %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpminsb %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_smin_v64i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpminsb %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpminsb %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpminsb %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpminsb %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_smin_v64i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpminsb %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpminsb %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <64 x i8> %x0, %x1
   %t1 = icmp slt <64 x i8> %x2, %t0
@@ -2898,15 +2947,22 @@ define <32 x i16> @reassociate_smin_v32i16(<32 x i16> %x0, <32 x i16> %x1, <32 x
 ; SSE-NEXT:    pminsw {{[0-9]+}}(%rsp), %xmm3
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: reassociate_smin_v32i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
-; AVX-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
-; AVX-NEXT:    vpminsw %ymm1, %ymm5, %ymm1
-; AVX-NEXT:    vpminsw %ymm0, %ymm4, %ymm0
-; AVX-NEXT:    vpminsw %ymm0, %ymm6, %ymm0
-; AVX-NEXT:    vpminsw %ymm1, %ymm7, %ymm1
-; AVX-NEXT:    retq
+; AVX2-LABEL: reassociate_smin_v32i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddw %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpminsw %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpminsw %ymm0, %ymm4, %ymm0
+; AVX2-NEXT:    vpminsw %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpminsw %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: reassociate_smin_v32i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpaddw %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpminsw %zmm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpminsw %zmm0, %zmm3, %zmm0
+; AVX512-NEXT:    retq
 
   %t0 = add <32 x i16> %x0, %x1
   %t1 = icmp slt <32 x i16> %x2, %t0