[CostModel][X86] Fix AVX512BW vector shift costs for vXi16 types

We already have patterns in place to support 128/256-bit shifts without AVX512VL

llvm-svn: 292077

4 files changed, 28 insertions, 10 deletions

diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index dfc2a8d8c62..567ef743f6b 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -323,6 +323,14 @@ int X86TTIImpl::getArithmeticInstrCost(
       return LT.first * Entry->Cost;
 
   static const CostTblEntry AVX512BWCostTable[] = {
+    { ISD::SHL,   MVT::v8i16,      1 }, // vpsllvw
+    { ISD::SRL,   MVT::v8i16,      1 }, // vpsrlvw
+    { ISD::SRA,   MVT::v8i16,      1 }, // vpsravw
+
+    { ISD::SHL,   MVT::v16i16,     1 }, // vpsllvw
+    { ISD::SRL,   MVT::v16i16,     1 }, // vpsrlvw
+    { ISD::SRA,   MVT::v16i16,     1 }, // vpsravw
+
     { ISD::SHL,   MVT::v32i16,     1 }, // vpsllvw
     { ISD::SRL,   MVT::v32i16,     1 }, // vpsrlvw
     { ISD::SRA,   MVT::v32i16,     1 }, // vpsravw
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
index 811914b6666..a23b13fb2e2 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
@@ -98,7 +98,8 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; SSE41: Found an estimated cost of 14 for instruction:   %shift
 ; AVX: Found an estimated cost of 14 for instruction:   %shift
 ; AVX2: Found an estimated cost of 14 for instruction:   %shift
-; AVX512: Found an estimated cost of 14 for instruction:   %shift
+; AVX512F: Found an estimated cost of 14 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = ashr <8 x i16> %a, %b
   ret <8 x i16> %shift
@@ -110,7 +111,8 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
 ; AVX: Found an estimated cost of 28 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
-; AVX512: Found an estimated cost of 10 for instruction:   %shift
+; AVX512F: Found an estimated cost of 10 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = ashr <16 x i16> %a, %b
   ret <16 x i16> %shift
@@ -432,7 +434,8 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
 ; SSE41: Found an estimated cost of 14 for instruction:   %shift
 ; AVX: Found an estimated cost of 14 for instruction:   %shift
 ; AVX2: Found an estimated cost of 14 for instruction:   %shift
-; AVX512: Found an estimated cost of 14 for instruction:   %shift
+; AVX512F: Found an estimated cost of 14 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <8 x i16> %shift
@@ -444,7 +447,8 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
 ; AVX: Found an estimated cost of 28 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
-; AVX512: Found an estimated cost of 10 for instruction:   %shift
+; AVX512F: Found an estimated cost of 10 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <16 x i16> %shift
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
index f1cc2747c9d..546b2bb50f2 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
@@ -101,7 +101,8 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; SSE41: Found an estimated cost of 14 for instruction:   %shift
 ; AVX: Found an estimated cost of 14 for instruction:   %shift
 ; AVX2: Found an estimated cost of 14 for instruction:   %shift
-; AVX512: Found an estimated cost of 14 for instruction:   %shift
+; AVX512F: Found an estimated cost of 14 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = lshr <8 x i16> %a, %b
   ret <8 x i16> %shift
@@ -113,7 +114,8 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
 ; AVX: Found an estimated cost of 28 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
-; AVX512: Found an estimated cost of 10 for instruction:   %shift
+; AVX512F: Found an estimated cost of 10 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = lshr <16 x i16> %a, %b
   ret <16 x i16> %shift
@@ -437,7 +439,8 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
 ; SSE41: Found an estimated cost of 14 for instruction:   %shift
 ; AVX: Found an estimated cost of 14 for instruction:   %shift
 ; AVX2: Found an estimated cost of 14 for instruction:   %shift
-; AVX512: Found an estimated cost of 14 for instruction:   %shift
+; AVX512F: Found an estimated cost of 14 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <8 x i16> %shift
@@ -449,7 +452,8 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
 ; AVX: Found an estimated cost of 28 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
-; AVX512: Found an estimated cost of 10 for instruction:   %shift
+; AVX512F: Found an estimated cost of 10 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
   %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <16 x i16> %shift
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
index 5d0590a3542..90356f5ce8b 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
@@ -102,7 +102,8 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
 ; SSE41: Found an estimated cost of 14 for instruction:   %shift
 ; AVX: Found an estimated cost of 14 for instruction:   %shift
 ; AVX2: Found an estimated cost of 14 for instruction:   %shift
-; AVX512: Found an estimated cost of 14 for instruction:   %shift
+; AVX512F: Found an estimated cost of 14 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 1 for instruction:   %shift
   %shift = shl <8 x i16> %a, %b
   ret <8 x i16> %shift
@@ -114,7 +115,8 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
 ; AVX: Found an estimated cost of 28 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
-; AVX512: Found an estimated cost of 10 for instruction:   %shift
+; AVX512F: Found an estimated cost of 10 for instruction:   %shift
+; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 2 for instruction:   %shift
   %shift = shl <16 x i16> %a, %b
   ret <16 x i16> %shift