[X86][AVX1] Account for cost of extract/insert of 256-bit shifts

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@303023 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp
index 4a5c7e395fd5526aa524bd05faa66efd50368a6b..80e18161a94bb25cf4d71e1558f5d3b4285fbacc 100644 (file)
--- a/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -589,28 +589,28 @@ int X86TTIImpl::getArithmeticInstrCost(
       return LT.first * Entry->Cost;
 
   static const CostTblEntry SSE41CostTable[] = {
-    { ISD::SHL,  MVT::v16i8,    11 }, // pblendvb sequence.
-    { ISD::SHL,  MVT::v32i8,  2*11 }, // pblendvb sequence.
-    { ISD::SHL,  MVT::v8i16,    14 }, // pblendvb sequence.
-    { ISD::SHL,  MVT::v16i16, 2*14 }, // pblendvb sequence.
-    { ISD::SHL,  MVT::v4i32,     4 }, // pslld/paddd/cvttps2dq/pmulld
-    { ISD::SHL,  MVT::v8i32,   2*4 }, // pslld/paddd/cvttps2dq/pmulld
-
-    { ISD::SRL,  MVT::v16i8,    12 }, // pblendvb sequence.
-    { ISD::SRL,  MVT::v32i8,  2*12 }, // pblendvb sequence.
-    { ISD::SRL,  MVT::v8i16,    14 }, // pblendvb sequence.
-    { ISD::SRL,  MVT::v16i16, 2*14 }, // pblendvb sequence.
-    { ISD::SRL,  MVT::v4i32,    11 }, // Shift each lane + blend.
-    { ISD::SRL,  MVT::v8i32,  2*11 }, // Shift each lane + blend.
-
-    { ISD::SRA,  MVT::v16i8,    24 }, // pblendvb sequence.
-    { ISD::SRA,  MVT::v32i8,  2*24 }, // pblendvb sequence.
-    { ISD::SRA,  MVT::v8i16,    14 }, // pblendvb sequence.
-    { ISD::SRA,  MVT::v16i16, 2*14 }, // pblendvb sequence.
-    { ISD::SRA,  MVT::v4i32,    12 }, // Shift each lane + blend.
-    { ISD::SRA,  MVT::v8i32,  2*12 }, // Shift each lane + blend.
-
-    { ISD::MUL,  MVT::v4i32,     1 }  // pmulld
+    { ISD::SHL,  MVT::v16i8,      11 }, // pblendvb sequence.
+    { ISD::SHL,  MVT::v32i8,  2*11+2 }, // pblendvb sequence + split.
+    { ISD::SHL,  MVT::v8i16,      14 }, // pblendvb sequence.
+    { ISD::SHL,  MVT::v16i16, 2*14+2 }, // pblendvb sequence + split.
+    { ISD::SHL,  MVT::v4i32,       4 }, // pslld/paddd/cvttps2dq/pmulld
+    { ISD::SHL,  MVT::v8i32,   2*4+2 }, // pslld/paddd/cvttps2dq/pmulld + split
+
+    { ISD::SRL,  MVT::v16i8,      12 }, // pblendvb sequence.
+    { ISD::SRL,  MVT::v32i8,  2*12+2 }, // pblendvb sequence + split.
+    { ISD::SRL,  MVT::v8i16,      14 }, // pblendvb sequence.
+    { ISD::SRL,  MVT::v16i16, 2*14+2 }, // pblendvb sequence + split.
+    { ISD::SRL,  MVT::v4i32,      11 }, // Shift each lane + blend.
+    { ISD::SRL,  MVT::v8i32,  2*11+2 }, // Shift each lane + blend + split.
+
+    { ISD::SRA,  MVT::v16i8,      24 }, // pblendvb sequence.
+    { ISD::SRA,  MVT::v32i8,  2*24+2 }, // pblendvb sequence + split.
+    { ISD::SRA,  MVT::v8i16,      14 }, // pblendvb sequence.
+    { ISD::SRA,  MVT::v16i16, 2*14+2 }, // pblendvb sequence + split.
+    { ISD::SRA,  MVT::v4i32,      12 }, // Shift each lane + blend.
+    { ISD::SRA,  MVT::v8i32,  2*12+2 }, // Shift each lane + blend + split.
+
+    { ISD::MUL,  MVT::v4i32,       1 }  // pmulld
   };
 
   if (ST->hasSSE41())
@@ -620,33 +620,33 @@ int X86TTIImpl::getArithmeticInstrCost(
   static const CostTblEntry SSE2CostTable[] = {
     // We don't correctly identify costs of casts because they are marked as
     // custom.
-    { ISD::SHL,  MVT::v16i8,    26 }, // cmpgtb sequence.
-    { ISD::SHL,  MVT::v8i16,    32 }, // cmpgtb sequence.
-    { ISD::SHL,  MVT::v4i32,   2*5 }, // We optimized this using mul.
-    { ISD::SHL,  MVT::v2i64,     4 }, // splat+shuffle sequence.
-    { ISD::SHL,  MVT::v4i64,   2*4 }, // splat+shuffle sequence.
-
-    { ISD::SRL,  MVT::v16i8,    26 }, // cmpgtb sequence.
-    { ISD::SRL,  MVT::v8i16,    32 }, // cmpgtb sequence.
-    { ISD::SRL,  MVT::v4i32,    16 }, // Shift each lane + blend.
-    { ISD::SRL,  MVT::v2i64,     4 }, // splat+shuffle sequence.
-    { ISD::SRL,  MVT::v4i64,   2*4 }, // splat+shuffle sequence.
-
-    { ISD::SRA,  MVT::v16i8,    54 }, // unpacked cmpgtb sequence.
-    { ISD::SRA,  MVT::v8i16,    32 }, // cmpgtb sequence.
-    { ISD::SRA,  MVT::v4i32,    16 }, // Shift each lane + blend.
-    { ISD::SRA,  MVT::v2i64,    12 }, // srl/xor/sub sequence.
-    { ISD::SRA,  MVT::v4i64,  2*12 }, // srl/xor/sub sequence.
-
-    { ISD::MUL,  MVT::v16i8,    12 }, // extend/pmullw/trunc sequence.
-    { ISD::MUL,  MVT::v8i16,     1 }, // pmullw
-    { ISD::MUL,  MVT::v4i32,     6 }, // 3*pmuludq/4*shuffle
-    { ISD::MUL,  MVT::v2i64,     8 }, // 3*pmuludq/3*shift/2*add
-
-    { ISD::FDIV, MVT::f32,      23 }, // Pentium IV from http://www.agner.org/
-    { ISD::FDIV, MVT::v4f32,    39 }, // Pentium IV from http://www.agner.org/
-    { ISD::FDIV, MVT::f64,      38 }, // Pentium IV from http://www.agner.org/
-    { ISD::FDIV, MVT::v2f64,    69 }, // Pentium IV from http://www.agner.org/
+    { ISD::SHL,  MVT::v16i8,      26 }, // cmpgtb sequence.
+    { ISD::SHL,  MVT::v8i16,      32 }, // cmpgtb sequence.
+    { ISD::SHL,  MVT::v4i32,     2*5 }, // We optimized this using mul.
+    { ISD::SHL,  MVT::v2i64,       4 }, // splat+shuffle sequence.
+    { ISD::SHL,  MVT::v4i64,   2*4+2 }, // splat+shuffle sequence + split.
+
+    { ISD::SRL,  MVT::v16i8,      26 }, // cmpgtb sequence.
+    { ISD::SRL,  MVT::v8i16,      32 }, // cmpgtb sequence.
+    { ISD::SRL,  MVT::v4i32,      16 }, // Shift each lane + blend.
+    { ISD::SRL,  MVT::v2i64,       4 }, // splat+shuffle sequence.
+    { ISD::SRL,  MVT::v4i64,   2*4+2 }, // splat+shuffle sequence + split.
+
+    { ISD::SRA,  MVT::v16i8,      54 }, // unpacked cmpgtb sequence.
+    { ISD::SRA,  MVT::v8i16,      32 }, // cmpgtb sequence.
+    { ISD::SRA,  MVT::v4i32,      16 }, // Shift each lane + blend.
+    { ISD::SRA,  MVT::v2i64,      12 }, // srl/xor/sub sequence.
+    { ISD::SRA,  MVT::v4i64,  2*12+2 }, // srl/xor/sub sequence+split.
+
+    { ISD::MUL,  MVT::v16i8,      12 }, // extend/pmullw/trunc sequence.
+    { ISD::MUL,  MVT::v8i16,       1 }, // pmullw
+    { ISD::MUL,  MVT::v4i32,       6 }, // 3*pmuludq/4*shuffle
+    { ISD::MUL,  MVT::v2i64,       8 }, // 3*pmuludq/3*shift/2*add
+
+    { ISD::FDIV, MVT::f32,        23 }, // Pentium IV from http://www.agner.org/
+    { ISD::FDIV, MVT::v4f32,      39 }, // Pentium IV from http://www.agner.org/
+    { ISD::FDIV, MVT::f64,        38 }, // Pentium IV from http://www.agner.org/
+    { ISD::FDIV, MVT::v2f64,      69 }, // Pentium IV from http://www.agner.org/
 
     // It is not a good idea to vectorize division. We have to scalarize it and
     // in the process we will often end up having to spilling regular

diff --git a/test/Analysis/CostModel/X86/vshift-ashr-cost.ll b/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
index 9b3b58a42b1247b34e9a07d20a22e89ec789cacc..eabc2330ddc6648c6439e375a668f68c2f077e1f 100644 (file)
--- a/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
+++ b/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
@@ -33,7 +33,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v4i64':
 ; SSE2: Found an estimated cost of 24 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -45,7 +45,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v8i64':
 ; SSE2: Found an estimated cost of 48 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 8 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 12 for instruction:   %shift
@@ -70,7 +70,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v8i32':
 ; SSE2: Found an estimated cost of 32 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 6 for instruction:   %shift
@@ -83,7 +83,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v16i32':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 12 for instruction:   %shift
@@ -109,7 +109,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v16i16':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
-; AVX: Found an estimated cost of 28 for instruction:   %shift
+; AVX: Found an estimated cost of 30 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -122,7 +122,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v32i16':
 ; SSE2: Found an estimated cost of 128 for instruction:   %shift
 ; SSE41: Found an estimated cost of 56 for instruction:   %shift
-; AVX: Found an estimated cost of 56 for instruction:   %shift
+; AVX: Found an estimated cost of 60 for instruction:   %shift
 ; AVX2: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -147,7 +147,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v32i8':
 ; SSE2: Found an estimated cost of 108 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 50 for instruction:   %shift
 ; AVX2: Found an estimated cost of 24 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 24 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 24 for instruction:   %shift
@@ -160,7 +160,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v64i8':
 ; SSE2: Found an estimated cost of 216 for instruction:   %shift
 ; SSE41: Found an estimated cost of 96 for instruction:   %shift
-; AVX: Found an estimated cost of 96 for instruction:   %shift
+; AVX: Found an estimated cost of 100 for instruction:   %shift
 ; AVX2: Found an estimated cost of 48 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 48 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 24 for instruction:   %shift
@@ -322,7 +322,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
 ; CHECK: 'Cost Model Analysis' for function 'splatvar_shift_v32i8':
 ; SSE2: Found an estimated cost of 108 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 50 for instruction:   %shift
 ; AVX2: Found an estimated cost of 24 for instruction:   %shift
 ; AVX512: Found an estimated cost of 24 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -336,7 +336,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
 ; CHECK: 'Cost Model Analysis' for function 'splatvar_shift_v64i8':
 ; SSE2: Found an estimated cost of 216 for instruction:   %shift
 ; SSE41: Found an estimated cost of 96 for instruction:   %shift
-; AVX: Found an estimated cost of 96 for instruction:   %shift
+; AVX: Found an estimated cost of 100 for instruction:   %shift
 ; AVX2: Found an estimated cost of 48 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 48 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 24 for instruction:   %shift
@@ -367,7 +367,7 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v4i64':
 ; SSE2: Found an estimated cost of 24 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 4 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -379,7 +379,7 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v8i64':
 ; SSE2: Found an estimated cost of 48 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 8 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOP: Found an estimated cost of 12 for instruction:   %shift
@@ -404,7 +404,7 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v8i32':
 ; SSE2: Found an estimated cost of 32 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 6 for instruction:   %shift
@@ -417,7 +417,7 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v16i32':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 12 for instruction:   %shift
@@ -443,7 +443,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v16i16':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
-; AVX: Found an estimated cost of 28 for instruction:   %shift
+; AVX: Found an estimated cost of 30 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -456,7 +456,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v32i16':
 ; SSE2: Found an estimated cost of 128 for instruction:   %shift
 ; SSE41: Found an estimated cost of 56 for instruction:   %shift
-; AVX: Found an estimated cost of 56 for instruction:   %shift
+; AVX: Found an estimated cost of 60 for instruction:   %shift
 ; AVX2: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -481,7 +481,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v32i8':
 ; SSE2: Found an estimated cost of 108 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 50 for instruction:   %shift
 ; AVX2: Found an estimated cost of 24 for instruction:   %shift
 ; AVX512: Found an estimated cost of 24 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -493,7 +493,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v64i8':
 ; SSE2: Found an estimated cost of 216 for instruction:   %shift
 ; SSE41: Found an estimated cost of 96 for instruction:   %shift
-; AVX: Found an estimated cost of 96 for instruction:   %shift
+; AVX: Found an estimated cost of 100 for instruction:   %shift
 ; AVX2: Found an estimated cost of 48 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 48 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 24 for instruction:   %shift

diff --git a/test/Analysis/CostModel/X86/vshift-lshr-cost.ll b/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
index d5c5d23efdfd7827de9e3347f7e29f8e29fd6d05..6e890369d677ecd27ba61a4d9c71d57920c1559b 100644 (file)
--- a/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
+++ b/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
@@ -34,7 +34,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v4i64':
 ; SSE2: Found an estimated cost of 8 for instruction:   %shift
 ; SSE41: Found an estimated cost of 8 for instruction:   %shift
-; AVX: Found an estimated cost of 8 for instruction:   %shift
+; AVX: Found an estimated cost of 10 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 6 for instruction:   %shift
@@ -47,7 +47,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v8i64':
 ; SSE2: Found an estimated cost of 16 for instruction:   %shift
 ; SSE41: Found an estimated cost of 16 for instruction:   %shift
-; AVX: Found an estimated cost of 16 for instruction:   %shift
+; AVX: Found an estimated cost of 20 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 12 for instruction:   %shift
@@ -73,7 +73,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v8i32':
 ; SSE2: Found an estimated cost of 32 for instruction:   %shift
 ; SSE41: Found an estimated cost of 22 for instruction:   %shift
-; AVX: Found an estimated cost of 22 for instruction:   %shift
+; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 6 for instruction:   %shift
@@ -86,7 +86,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v16i32':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 44 for instruction:   %shift
-; AVX: Found an estimated cost of 44 for instruction:   %shift
+; AVX: Found an estimated cost of 48 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 12 for instruction:   %shift
@@ -112,7 +112,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v16i16':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
-; AVX: Found an estimated cost of 28 for instruction:   %shift
+; AVX: Found an estimated cost of 30 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -125,7 +125,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v32i16':
 ; SSE2: Found an estimated cost of 128 for instruction:   %shift
 ; SSE41: Found an estimated cost of 56 for instruction:   %shift
-; AVX: Found an estimated cost of 56 for instruction:   %shift
+; AVX: Found an estimated cost of 60 for instruction:   %shift
 ; AVX2: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -150,7 +150,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v32i8':
 ; SSE2: Found an estimated cost of 52 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 11 for instruction:   %shift
 ; AVX512: Found an estimated cost of 11 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -162,7 +162,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v64i8':
 ; SSE2: Found an estimated cost of 104 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 11 for instruction:   %shift
@@ -326,7 +326,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
 ; CHECK: 'Cost Model Analysis' for function 'splatvar_shift_v32i8':
 ; SSE2: Found an estimated cost of 52 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 11 for instruction:   %shift
 ; AVX512: Found an estimated cost of 11 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -340,7 +340,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
 ; CHECK: 'Cost Model Analysis' for function 'splatvar_shift_v64i8':
 ; SSE2: Found an estimated cost of 104 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 11 for instruction:   %shift
@@ -372,7 +372,7 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v4i64':
 ; SSE2: Found an estimated cost of 8 for instruction:   %shift
 ; SSE41: Found an estimated cost of 8 for instruction:   %shift
-; AVX: Found an estimated cost of 8 for instruction:   %shift
+; AVX: Found an estimated cost of 10 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 6 for instruction:   %shift
@@ -385,7 +385,7 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v8i64':
 ; SSE2: Found an estimated cost of 16 for instruction:   %shift
 ; SSE41: Found an estimated cost of 16 for instruction:   %shift
-; AVX: Found an estimated cost of 16 for instruction:   %shift
+; AVX: Found an estimated cost of 20 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 12 for instruction:   %shift
@@ -411,7 +411,7 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v8i32':
 ; SSE2: Found an estimated cost of 32 for instruction:   %shift
 ; SSE41: Found an estimated cost of 22 for instruction:   %shift
-; AVX: Found an estimated cost of 22 for instruction:   %shift
+; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 6 for instruction:   %shift
@@ -424,7 +424,7 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v16i32':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 44 for instruction:   %shift
-; AVX: Found an estimated cost of 44 for instruction:   %shift
+; AVX: Found an estimated cost of 48 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 12 for instruction:   %shift
@@ -450,7 +450,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v16i16':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
-; AVX: Found an estimated cost of 28 for instruction:   %shift
+; AVX: Found an estimated cost of 30 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -463,7 +463,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v32i16':
 ; SSE2: Found an estimated cost of 128 for instruction:   %shift
 ; SSE41: Found an estimated cost of 56 for instruction:   %shift
-; AVX: Found an estimated cost of 56 for instruction:   %shift
+; AVX: Found an estimated cost of 60 for instruction:   %shift
 ; AVX2: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -488,7 +488,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v32i8':
 ; SSE2: Found an estimated cost of 52 for instruction:   %shift
 ; SSE41: Found an estimated cost of 24 for instruction:   %shift
-; AVX: Found an estimated cost of 24 for instruction:   %shift
+; AVX: Found an estimated cost of 26 for instruction:   %shift
 ; AVX2: Found an estimated cost of 11 for instruction:   %shift
 ; AVX512: Found an estimated cost of 11 for instruction:   %shift
 ; XOP: Found an estimated cost of 6 for instruction:   %shift
@@ -500,7 +500,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v64i8':
 ; SSE2: Found an estimated cost of 104 for instruction:   %shift
 ; SSE41: Found an estimated cost of 48 for instruction:   %shift
-; AVX: Found an estimated cost of 48 for instruction:   %shift
+; AVX: Found an estimated cost of 52 for instruction:   %shift
 ; AVX2: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 11 for instruction:   %shift

diff --git a/test/Analysis/CostModel/X86/vshift-shl-cost.ll b/test/Analysis/CostModel/X86/vshift-shl-cost.ll
index ddcd2ecc727751f651764cf0a4ed21f27f511ab0..5e604bb7983e3ee1c2795aa5c350b7adccb67e32 100644 (file)
--- a/test/Analysis/CostModel/X86/vshift-shl-cost.ll
+++ b/test/Analysis/CostModel/X86/vshift-shl-cost.ll
@@ -35,7 +35,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v4i64':
 ; SSE2: Found an estimated cost of 8 for instruction:   %shift
 ; SSE41: Found an estimated cost of 8 for instruction:   %shift
-; AVX: Found an estimated cost of 8 for instruction:   %shift
+; AVX: Found an estimated cost of 10 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 4 for instruction:   %shift
@@ -48,7 +48,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v8i64':
 ; SSE2: Found an estimated cost of 16 for instruction:   %shift
 ; SSE41: Found an estimated cost of 16 for instruction:   %shift
-; AVX: Found an estimated cost of 16 for instruction:   %shift
+; AVX: Found an estimated cost of 20 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 8 for instruction:   %shift
@@ -74,7 +74,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v8i32':
 ; SSE2: Found an estimated cost of 20 for instruction:   %shift
 ; SSE41: Found an estimated cost of 8 for instruction:   %shift
-; AVX: Found an estimated cost of 8 for instruction:   %shift
+; AVX: Found an estimated cost of 10 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 4 for instruction:   %shift
@@ -87,7 +87,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v16i32':
 ; SSE2: Found an estimated cost of 40 for instruction:   %shift
 ; SSE41: Found an estimated cost of 16 for instruction:   %shift
-; AVX: Found an estimated cost of 16 for instruction:   %shift
+; AVX: Found an estimated cost of 20 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 8 for instruction:   %shift
@@ -113,7 +113,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v16i16':
 ; SSE2: Found an estimated cost of 64 for instruction:   %shift
 ; SSE41: Found an estimated cost of 28 for instruction:   %shift
-; AVX: Found an estimated cost of 28 for instruction:   %shift
+; AVX: Found an estimated cost of 30 for instruction:   %shift
 ; AVX2: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 10 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -126,7 +126,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v32i16':
 ; SSE2: Found an estimated cost of 128 for instruction:   %shift
 ; SSE41: Found an estimated cost of 56 for instruction:   %shift
-; AVX: Found an estimated cost of 56 for instruction:   %shift
+; AVX: Found an estimated cost of 60 for instruction:   %shift
 ; AVX2: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 20 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 1 for instruction:   %shift
@@ -151,7 +151,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v32i8':
 ; SSE2: Found an estimated cost of 52 for instruction:   %shift
 ; SSE41: Found an estimated cost of 22 for instruction:   %shift
-; AVX: Found an estimated cost of 22 for instruction:   %shift
+; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 11 for instruction:   %shift
 ; AVX512: Found an estimated cost of 11 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
@@ -163,7 +163,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
 ; CHECK: 'Cost Model Analysis' for function 'var_shift_v64i8':
 ; SSE2: Found an estimated cost of 104 for instruction:   %shift
 ; SSE41: Found an estimated cost of 44 for instruction:   %shift
-; AVX: Found an estimated cost of 44 for instruction:   %shift
+; AVX: Found an estimated cost of 48 for instruction:   %shift
 ; AVX2: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 11 for instruction:   %shift
@@ -327,7 +327,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
 ; CHECK: 'Cost Model Analysis' for function 'splatvar_shift_v32i8':
 ; SSE2: Found an estimated cost of 52 for instruction:   %shift
 ; SSE41: Found an estimated cost of 22 for instruction:   %shift
-; AVX: Found an estimated cost of 22 for instruction:   %shift
+; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 11 for instruction:   %shift
 ; AVX512: Found an estimated cost of 11 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
@@ -341,7 +341,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
 ; CHECK: 'Cost Model Analysis' for function 'splatvar_shift_v64i8':
 ; SSE2: Found an estimated cost of 104 for instruction:   %shift
 ; SSE41: Found an estimated cost of 44 for instruction:   %shift
-; AVX: Found an estimated cost of 44 for instruction:   %shift
+; AVX: Found an estimated cost of 48 for instruction:   %shift
 ; AVX2: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 11 for instruction:   %shift
@@ -373,7 +373,7 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v4i64':
 ; SSE2: Found an estimated cost of 8 for instruction:   %shift
 ; SSE41: Found an estimated cost of 8 for instruction:   %shift
-; AVX: Found an estimated cost of 8 for instruction:   %shift
+; AVX: Found an estimated cost of 10 for instruction:   %shift
 ; AVX2: Found an estimated cost of 1 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 4 for instruction:   %shift
@@ -386,7 +386,7 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v8i64':
 ; SSE2: Found an estimated cost of 16 for instruction:   %shift
 ; SSE41: Found an estimated cost of 16 for instruction:   %shift
-; AVX: Found an estimated cost of 16 for instruction:   %shift
+; AVX: Found an estimated cost of 20 for instruction:   %shift
 ; AVX2: Found an estimated cost of 2 for instruction:   %shift
 ; AVX512: Found an estimated cost of 1 for instruction:   %shift
 ; XOPAVX: Found an estimated cost of 8 for instruction:   %shift
@@ -489,7 +489,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v32i8':
 ; SSE2: Found an estimated cost of 52 for instruction:   %shift
 ; SSE41: Found an estimated cost of 22 for instruction:   %shift
-; AVX: Found an estimated cost of 22 for instruction:   %shift
+; AVX: Found an estimated cost of 24 for instruction:   %shift
 ; AVX2: Found an estimated cost of 11 for instruction:   %shift
 ; AVX512: Found an estimated cost of 11 for instruction:   %shift
 ; XOP: Found an estimated cost of 4 for instruction:   %shift
@@ -501,7 +501,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'constant_shift_v64i8':
 ; SSE2: Found an estimated cost of 104 for instruction:   %shift
 ; SSE41: Found an estimated cost of 44 for instruction:   %shift
-; AVX: Found an estimated cost of 44 for instruction:   %shift
+; AVX: Found an estimated cost of 48 for instruction:   %shift
 ; AVX2: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512F: Found an estimated cost of 22 for instruction:   %shift
 ; AVX512BW: Found an estimated cost of 11 for instruction:   %shift
@@ -794,7 +794,7 @@ define <4 x i64> @test8(<4 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'test8':
 ; SSE2: Found an estimated cost of 8 for instruction:   %shl
 ; SSE41: Found an estimated cost of 8 for instruction:   %shl
-; AVX: Found an estimated cost of 8 for instruction:   %shl
+; AVX: Found an estimated cost of 10 for instruction:   %shl
 ; AVX2: Found an estimated cost of 1 for instruction:   %shl
 ; XOPAVX: Found an estimated cost of 4 for instruction:   %shl
 ; XOPAVX2: Found an estimated cost of 1 for instruction:   %shl
@@ -842,7 +842,7 @@ define <8 x i64> @test11(<8 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'test11':
 ; SSE2: Found an estimated cost of 16 for instruction:   %shl
 ; SSE41: Found an estimated cost of 16 for instruction:   %shl
-; AVX: Found an estimated cost of 16 for instruction:   %shl
+; AVX: Found an estimated cost of 20 for instruction:   %shl
 ; AVX2: Found an estimated cost of 2 for instruction:   %shl
 ; XOPAVX: Found an estimated cost of 8 for instruction:   %shl
 ; XOPAVX2: Found an estimated cost of 2 for instruction:   %shl