[CostModel][X86] Updated vXi64 ASHR costs on AVX512 targets now that D28604 has landed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292023 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp
index 5715d826862e9f88e5d0f9ba35ed0b432695c350..dfc2a8d8c62f3c3a52619f9c19712e2468bd09f3 100644 (file)
--- a/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -207,6 +207,10 @@ int X86TTIImpl::getArithmeticInstrCost(
   }
 
   static const CostTblEntry AVX512UniformConstCostTable[] = {
+    { ISD::SRA,  MVT::v2i64,   1 },
+    { ISD::SRA,  MVT::v4i64,   1 },
+    { ISD::SRA,  MVT::v8i64,   1 },
+
     { ISD::SDIV, MVT::v16i32, 15 }, // vpmuldq sequence
     { ISD::UDIV, MVT::v16i32, 15 }, // vpmuludq sequence
   };
@@ -347,8 +351,12 @@ int X86TTIImpl::getArithmeticInstrCost(
     { ISD::SHL,     MVT::v16i32,     1 },
     { ISD::SRL,     MVT::v16i32,     1 },
     { ISD::SRA,     MVT::v16i32,     1 },
+
     { ISD::SHL,     MVT::v8i64,      1 },
     { ISD::SRL,     MVT::v8i64,      1 },
+
+    { ISD::SRA,     MVT::v2i64,      1 },
+    { ISD::SRA,     MVT::v4i64,      1 },
     { ISD::SRA,     MVT::v8i64,      1 },
 
     { ISD::MUL,     MVT::v32i8,     13 }, // extend/pmullw/trunc sequence.

diff --git a/test/Analysis/CostModel/X86/vshift-ashr-cost.ll b/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
index 13519018d957086024009ae9dd2797ce2f375d49..7ff1130ae5454c17796e6351bd6720db8fe5bbba 100644 (file)
--- a/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
+++ b/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
@@ -23,7 +23,7 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
 ; SSE41: Found an estimated cost of 12 for instruction:   %shift
 ; AVX: Found an estimated cost of 12 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = ashr <2 x i64> %a, %b
   ret <2 x i64> %shift
@@ -35,7 +35,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
 ; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = ashr <4 x i64> %a, %b
   ret <4 x i64> %shift
@@ -183,7 +183,7 @@ define <2 x i64> @splatvar_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
 ; SSE41: Found an estimated cost of 12 for instruction:   %shift
 ; AVX: Found an estimated cost of 12 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %splat = shufflevector <2 x i64> %b, <2 x i64> undef, <2 x i32> zeroinitializer
   %shift = ashr <2 x i64> %a, %splat
@@ -196,7 +196,7 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
 ; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %splat = shufflevector <4 x i64> %b, <4 x i64> undef, <4 x i32> zeroinitializer
   %shift = ashr <4 x i64> %a, %splat
@@ -352,7 +352,7 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
 ; SSE41: Found an estimated cost of 12 for instruction:   %shift
 ; AVX: Found an estimated cost of 12 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = ashr <2 x i64> %a, <i64 1, i64 7>
   ret <2 x i64> %shift
@@ -364,7 +364,7 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
 ; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
   ret <4 x i64> %shift
@@ -509,7 +509,7 @@ define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) {
 ; SSE41: Found an estimated cost of 4 for instruction:   %shift
 ; AVX: Found an estimated cost of 4 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = ashr <2 x i64> %a, <i64 7, i64 7>
   ret <2 x i64> %shift
@@ -521,7 +521,7 @@ define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) {
 ; SSE41: Found an estimated cost of 8 for instruction:   %shift
 ; AVX: Found an estimated cost of 8 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
-; AVX512: Found an estimated cost of 4 for instruction:   %shift
+; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = ashr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
   ret <4 x i64> %shift