[llvm] r346683 - [CostModel][X86] Add SHLD/SHRD scalar funnel shift costs

Mon Nov 12 09:56:59 PST 2018

Author: rksimon
Date: Mon Nov 12 09:56:59 2018
New Revision: 346683

URL: http://llvm.org/viewvc/llvm-project?rev=346683&view=rev
Log:
[CostModel][X86] Add SHLD/SHRD scalar funnel shift costs

The costs match the typical reg-reg cases - the RMW case can be a lot slower but we don't model that at this level

Modified:
    llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp
    llvm/trunk/test/Analysis/CostModel/X86/fshl.ll
    llvm/trunk/test/Analysis/CostModel/X86/fshr.ll

Modified: llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp?rev=346683&r1=346682&r2=346683&view=diff
==============================================================================

--- llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp Mon Nov 12 09:56:59 2018
@@ -1848,12 +1848,16 @@ int X86TTIImpl::getIntrinsicInstrCost(In
     { ISD::FSQRT,      MVT::v4f32,  56 }, // Pentium III from http://www.agner.org/
   };
   static const CostTblEntry X64CostTbl[] = { // 64-bit targets
-    { ISD::BITREVERSE, MVT::i64,    14 }
+    { ISD::BITREVERSE, MVT::i64,    14 },
+    { X86ISD::SHLD,    MVT::i64,     4 } 
   };
   static const CostTblEntry X86CostTbl[] = { // 32 or 64-bit targets
     { ISD::BITREVERSE, MVT::i32,    14 },
     { ISD::BITREVERSE, MVT::i16,    14 },
-    { ISD::BITREVERSE, MVT::i8,     11 }
+    { ISD::BITREVERSE, MVT::i8,     11 },
+    { X86ISD::SHLD,    MVT::i32,     4 },
+    { X86ISD::SHLD,    MVT::i16,     4 },
+    { X86ISD::SHLD,    MVT::i8,      4 }
   };
 
   unsigned ISD = ISD::DELETED_NODE;
@@ -1875,6 +1879,11 @@ int X86TTIImpl::getIntrinsicInstrCost(In
   case Intrinsic::cttz:
     ISD = ISD::CTTZ;
     break;
+  case Intrinsic::fshl:
+  case Intrinsic::fshr:
+    // SHRD has same costs so don't duplicate.
+    ISD = X86ISD::SHLD;
+    break;
   case Intrinsic::sqrt:
     ISD = ISD::FSQRT;
     break;

Modified: llvm/trunk/test/Analysis/CostModel/X86/fshl.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/fshl.ll?rev=346683&r1=346682&r2=346683&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/fshl.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/fshl.ll Mon Nov 12 09:56:59 2018
@@ -22,38 +22,38 @@ target triple = "x86_64-apple-macosx10.8
 
 define void @var_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512, i64 %b64, <2 x i64> %b128, <4 x i64> %b256, <8 x i64> %b512, i64 %c64, <2 x i64> %c128, <4 x i64> %c256, <8 x i64> %c512) {
 ; CHECK-LABEL: 'var_funnel_i64'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i64'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i64'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i64'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i64'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I64    = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
@@ -65,38 +65,38 @@ define void @var_funnel_i64(i64 %a64, <2
 
 define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %b32, <4 x i32> %b128, <8 x i32> %b256, <16 x i32> %b512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
 ; CHECK-LABEL: 'var_funnel_i32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i32'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i32'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i32'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i32'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I32   = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
@@ -108,38 +108,38 @@ define void @var_funnel_i32(i32 %a32, <4
 
 define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %b16, <8 x i16> %b128, <16 x i16> %b256, <32 x i16> %b512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
 ; CHECK-LABEL: 'var_funnel_i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i16'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i16'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i16'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i16'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I16    = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
@@ -151,38 +151,38 @@ define void @var_funnel_i16(i16 %a16, <8
 
 define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512, i8 %b8, <16 x i8> %b128, <32 x i8> %b256, <64 x i8> %b512, i8 %c8, <16 x i8> %c128, <32 x i8> %c256, <64 x i8> %c512) {
 ; CHECK-LABEL: 'var_funnel_i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i8'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i8'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i8'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i8'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
@@ -199,38 +199,38 @@ define void @var_funnel_i8(i8 %a8, <16 x
 
 define void @var_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512, i64 %c64, <2 x i64> %c128, <4 x i64> %c256, <8 x i64> %c512) {
 ; CHECK-LABEL: 'var_rotate_i64'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i64'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i64'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i64'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i64'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I64    = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
@@ -242,38 +242,38 @@ define void @var_rotate_i64(i64 %a64, <2
 
 define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
 ; CHECK-LABEL: 'var_rotate_i32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i32'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i32'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i32'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i32'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I32   = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
@@ -285,38 +285,38 @@ define void @var_rotate_i32(i32 %a32, <4
 
 define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
 ; CHECK-LABEL: 'var_rotate_i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i16'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i16'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i16'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i16'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I16    = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
@@ -328,38 +328,38 @@ define void @var_rotate_i16(i16 %a16, <8
 
 define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512, i8 %c8, <16 x i8> %c128, <32 x i8> %c256, <64 x i8> %c512) {
 ; CHECK-LABEL: 'var_rotate_i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i8'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i8'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i8'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i8'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)

Modified: llvm/trunk/test/Analysis/CostModel/X86/fshr.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/fshr.ll?rev=346683&r1=346682&r2=346683&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/fshr.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/fshr.ll Mon Nov 12 09:56:59 2018
@@ -22,38 +22,38 @@ target triple = "x86_64-apple-macosx10.8
 
 define void @var_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512, i64 %b64, <2 x i64> %b128, <4 x i64> %b256, <8 x i64> %b512, i64 %c64, <2 x i64> %c128, <4 x i64> %c256, <8 x i64> %c512) {
 ; CHECK-LABEL: 'var_funnel_i64'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i64'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i64'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i64'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i64'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I64    = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
@@ -65,38 +65,38 @@ define void @var_funnel_i64(i64 %a64, <2
 
 define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %b32, <4 x i32> %b128, <8 x i32> %b256, <16 x i32> %b512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
 ; CHECK-LABEL: 'var_funnel_i32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i32'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i32'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i32'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i32'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I32   = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
@@ -108,38 +108,38 @@ define void @var_funnel_i32(i32 %a32, <4
 
 define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %b16, <8 x i16> %b128, <16 x i16> %b256, <32 x i16> %b512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
 ; CHECK-LABEL: 'var_funnel_i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i16'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i16'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i16'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i16'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I16    = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
@@ -151,38 +151,38 @@ define void @var_funnel_i16(i16 %a16, <8
 
 define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512, i8 %b8, <16 x i8> %b128, <32 x i8> %b256, <64 x i8> %b512, i8 %c8, <16 x i8> %c128, <32 x i8> %c256, <64 x i8> %c512) {
 ; CHECK-LABEL: 'var_funnel_i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_funnel_i8'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_funnel_i8'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_funnel_i8'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_funnel_i8'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 512 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
@@ -199,38 +199,38 @@ define void @var_funnel_i8(i8 %a8, <16 x
 
 define void @var_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512, i64 %c64, <2 x i64> %c128, <4 x i64> %c256, <8 x i64> %c512) {
 ; CHECK-LABEL: 'var_rotate_i64'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i64'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i64'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i64'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i64'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I64    = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
@@ -242,38 +242,38 @@ define void @var_rotate_i64(i64 %a64, <2
 
 define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
 ; CHECK-LABEL: 'var_rotate_i32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i32'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i32'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i32'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i32'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I32   = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
@@ -285,38 +285,38 @@ define void @var_rotate_i32(i32 %a32, <4
 
 define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
 ; CHECK-LABEL: 'var_rotate_i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i16'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i16'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i16'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i16'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I16    = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
@@ -328,38 +328,38 @@ define void @var_rotate_i16(i16 %a16, <8
 
 define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512, i8 %c8, <16 x i8> %c128, <32 x i8> %c256, <64 x i8> %c512) {
 ; CHECK-LABEL: 'var_rotate_i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'var_rotate_i8'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; GLM-LABEL: 'var_rotate_i8'
-; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BDVER2-LABEL: 'var_rotate_i8'
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; BDVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; BDVER2-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; BDVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; BTVER2-LABEL: 'var_rotate_i8'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)


