[X86][XOP] Added support for VPMADCSWD 'extend+hadd' IFMA patterns

VPMADCSWD act as VPADDD( VPMADDWD( x, y ), z ) - multiply+extend+hadd and add to v4i32 accumulator

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292021 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrXOP.td b/lib/Target/X86/X86InstrXOP.td
index 85945e7794ffb782293dcf5f45a79d873e63534b..10acdca2df749415ff1f7a3b90c9ecd9b1ce8070 100644 (file)
--- a/lib/Target/X86/X86InstrXOP.td
+++ b/lib/Target/X86/X86InstrXOP.td
@@ -199,6 +199,9 @@ let Predicates = [HasXOP] in {
   def : Pat<(v2i64 (add (X86pmuldq (v4i32 VR128:$src1), (v4i32 VR128:$src2)),
                         (v2i64 VR128:$src3))),
             (VPMACSDQLrr VR128:$src1, VR128:$src2, VR128:$src3)>;
+  def : Pat<(v4i32 (add (X86vpmaddwd (v8i16 VR128:$src1), (v8i16 VR128:$src2)),
+                        (v4i32 VR128:$src3))),
+            (VPMADCSWDrr VR128:$src1, VR128:$src2, VR128:$src3)>;
 }
 
 // Instruction where second source can be memory, third must be imm8

diff --git a/test/CodeGen/X86/xop-ifma.ll b/test/CodeGen/X86/xop-ifma.ll
index a362dcdf257f0ffa03e995c2da16ee14c05aec54..83291095b876dfab23bafdca9a8fdd19ad9a3f35 100644 (file)
--- a/test/CodeGen/X86/xop-ifma.ll
+++ b/test/CodeGen/X86/xop-ifma.ll
@@ -118,8 +118,7 @@ define <2 x i64> @test_pmuldq_hi_v4i32_add_v2i64(<4 x i32> %a0, <4 x i32> %a1, <
 define <4 x i32> @test_pmaddwd_v8i16_add_v4i32(<8 x i16> %a0, <8 x i16> %a1, <4 x i32> %a2) {
 ; XOP-LABEL: test_pmaddwd_v8i16_add_v4i32:
 ; XOP:       # BB#0:
-; XOP-NEXT:    vpmaddwd %xmm1, %xmm0, %xmm0
-; XOP-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
+; XOP-NEXT:    vpmadcswd %xmm2, %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
   %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> %a1)
   %2 = add <4 x i32> %1, %a2