From 261f559da36636c9490ee0ae6aa1ffe0d9ca214f Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sat, 14 Jan 2017 18:08:54 +0000
Subject: [PATCH] [X86][XOP] Added support for VPMACSDQH/VPMACSDQL 'extension'
 IFMA patterns

VPMACSDQH/VPMACSDQL act as VPADDQ( VPMULDQ( x, y ), z ) - multiply+extending either the odd/even 4i32 input elements and adding to v2i64 accumulator

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292020 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrXOP.td |  9 ++++++++-
 test/CodeGen/X86/xop-ifma.ll  | 26 ++++++++++----------------
 2 files changed, 18 insertions(+), 17 deletions(-)

diff --git a/lib/Target/X86/X86InstrXOP.td b/lib/Target/X86/X86InstrXOP.td
index efd627d6b86..85945e7794f 100644
--- a/lib/Target/X86/X86InstrXOP.td
+++ b/lib/Target/X86/X86InstrXOP.td
@@ -184,7 +184,7 @@ let ExeDomain = SSEPackedInt in {
 }
 
 // IFMA patterns - for cases where we can safely ignore the overflow bits from
-// the multiply.
+// the multiply or easily match with existing intrinsics.
 let Predicates = [HasXOP] in {
   def : Pat<(v8i16 (add (mul (v8i16 VR128:$src1), (v8i16 VR128:$src2)),
                         (v8i16 VR128:$src3))),
@@ -192,6 +192,13 @@ let Predicates = [HasXOP] in {
   def : Pat<(v4i32 (add (mul (v4i32 VR128:$src1), (v4i32 VR128:$src2)),
                         (v4i32 VR128:$src3))),
             (VPMACSDDrr VR128:$src1, VR128:$src2, VR128:$src3)>;
+  def : Pat<(v2i64 (add (X86pmuldq (X86PShufd (v4i32 VR128:$src1), (i8 -11)),
+                                   (X86PShufd (v4i32 VR128:$src2), (i8 -11))),
+                        (v2i64 VR128:$src3))),
+            (VPMACSDQHrr VR128:$src1, VR128:$src2, VR128:$src3)>;
+  def : Pat<(v2i64 (add (X86pmuldq (v4i32 VR128:$src1), (v4i32 VR128:$src2)),
+                        (v2i64 VR128:$src3))),
+            (VPMACSDQLrr 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 088d6795814..a362dcdf257 100644
--- a/test/CodeGen/X86/xop-ifma.ll
+++ b/test/CodeGen/X86/xop-ifma.ll
@@ -67,18 +67,16 @@ define <8 x i32> @test_mul_v8i32_add_v8i32(<8 x i32> %a0, <8 x i32> %a1, <8 x i3
 define <4 x i64> @test_mulx_v4i32_add_v4i64(<4 x i32> %a0, <4 x i32> %a1, <4 x i64> %a2) {
 ; XOP-AVX1-LABEL: test_mulx_v4i32_add_v4i64:
 ; XOP-AVX1:       # BB#0:
-; XOP-AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
-; XOP-AVX1-NEXT:    vpmovsxdq %xmm3, %xmm3
+; XOP-AVX1-NEXT:    vpmovsxdq %xmm0, %xmm3
+; XOP-AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; XOP-AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
-; XOP-AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm1[2,3,0,1]
-; XOP-AVX1-NEXT:    vpmovsxdq %xmm4, %xmm4
+; XOP-AVX1-NEXT:    vpmovsxdq %xmm1, %xmm4
+; XOP-AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; XOP-AVX1-NEXT:    vpmovsxdq %xmm1, %xmm1
-; XOP-AVX1-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0
-; XOP-AVX1-NEXT:    vpmuldq %xmm4, %xmm3, %xmm1
-; XOP-AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; XOP-AVX1-NEXT:    vpaddq %xmm3, %xmm1, %xmm1
-; XOP-AVX1-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
-; XOP-AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; XOP-AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
+; XOP-AVX1-NEXT:    vpmacsdql %xmm5, %xmm1, %xmm0, %xmm0
+; XOP-AVX1-NEXT:    vpmacsdql %xmm2, %xmm4, %xmm3, %xmm1
+; XOP-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; XOP-AVX1-NEXT:    retq
 ;
 ; XOP-AVX2-LABEL: test_mulx_v4i32_add_v4i64:
@@ -98,8 +96,7 @@ define <4 x i64> @test_mulx_v4i32_add_v4i64(<4 x i32> %a0, <4 x i32> %a1, <4 x i
 define <2 x i64> @test_pmuldq_lo_v4i32_add_v2i64(<4 x i32> %a0, <4 x i32> %a1, <2 x i64> %a2) {
 ; XOP-LABEL: test_pmuldq_lo_v4i32_add_v2i64:
 ; XOP:       # BB#0:
-; XOP-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0
-; XOP-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
+; XOP-NEXT:    vpmacsdql %xmm2, %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
   %1 = call <2 x i64> @llvm.x86.sse41.pmuldq(<4 x i32> %a0, <4 x i32> %a1)
   %2 = add <2 x i64> %1, %a2
@@ -109,10 +106,7 @@ define <2 x i64> @test_pmuldq_lo_v4i32_add_v2i64(<4 x i32> %a0, <4 x i32> %a1, <
 define <2 x i64> @test_pmuldq_hi_v4i32_add_v2i64(<4 x i32> %a0, <4 x i32> %a1, <2 x i64> %a2) {
 ; XOP-LABEL: test_pmuldq_hi_v4i32_add_v2i64:
 ; XOP:       # BB#0:
-; XOP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; XOP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; XOP-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0
-; XOP-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
+; XOP-NEXT:    vpmacsdqh %xmm2, %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
   %1 = shufflevector <4 x i32> %a0, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 3, i32 undef>
   %2 = shufflevector <4 x i32> %a1, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 3, i32 undef>
-- 
2.40.0