[llvm] r292021 - [X86][XOP] Added support for VPMADCSWD 'extend+hadd' IFMA patterns

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Sat Jan 14 12:52:13 2017
New Revision: 292021

URL: http://llvm.org/viewvc/llvm-project?rev=292021&view=rev
Log:
[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

Modified:
    llvm/trunk/lib/Target/X86/X86InstrXOP.td
    llvm/trunk/test/CodeGen/X86/xop-ifma.ll

Modified: llvm/trunk/lib/Target/X86/X86InstrXOP.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrXOP.td?rev=292021&r1=292020&r2=292021&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrXOP.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrXOP.td Sat Jan 14 12:52:13 2017
@@ -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

Modified: llvm/trunk/test/CodeGen/X86/xop-ifma.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/xop-ifma.ll?rev=292021&r1=292020&r2=292021&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/xop-ifma.ll (original)
+++ llvm/trunk/test/CodeGen/X86/xop-ifma.ll Sat Jan 14 12:52:13 2017
@@ -118,8 +118,7 @@ define <2 x i64> @test_pmuldq_hi_v4i32_a
 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