X86: add costs for 64-bit vector ext/trunc & rebalance

The most important part of this is probably adding any cost at all for
operations like zext <8 x i8> to <8 x i32>. Before they were being
recorded as extremely costly (24, I believe) which made LLVM fall back
on a 4-wide vectorisation of a loop.

It also rebalances the values for sext, zext and trunc. Lacking any
other sane metric that might work across CPU microarchitectures I went
for instructions. This seems to be in reasonable accord with the rest
of the table (sitofp, ...) though no doubt at least one value is
sub-optimal for some bizarre reason.

Finally, separate AVX and AVX2 values are provided where appropriate.
The CodeGen is quite different in many cases.

rdar://problem/15981990

llvm-svn: 200928

2 files changed, 133 insertions, 37 deletions

diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 781be2fddd9..207a7685c59 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -411,16 +411,58 @@ unsigned X86TTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
     return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
 
   static const TypeConversionCostTblEntry<MVT::SimpleValueType>
+  AVX2ConversionTbl[] = {
+    { ISD::SIGN_EXTEND, MVT::v16i16, MVT::v16i8,  1 },
+    { ISD::ZERO_EXTEND, MVT::v16i16, MVT::v16i8,  1 },
+    { ISD::SIGN_EXTEND, MVT::v8i32,  MVT::v8i1,   3 },
+    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i1,   3 },
+    { ISD::SIGN_EXTEND, MVT::v8i32,  MVT::v8i8,   3 },
+    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i8,   3 },
+    { ISD::SIGN_EXTEND, MVT::v8i32,  MVT::v8i16,  1 },
+    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i16,  1 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i1,   3 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i1,   3 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i8,   3 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i8,   3 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i16,  3 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i16,  3 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i32,  1 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i32,  1 },
+
+    { ISD::TRUNCATE,    MVT::v4i8,   MVT::v4i64,  2 },
+    { ISD::TRUNCATE,    MVT::v4i16,  MVT::v4i64,  2 },
+    { ISD::TRUNCATE,    MVT::v4i32,  MVT::v4i64,  2 },
+    { ISD::TRUNCATE,    MVT::v8i8,   MVT::v8i32,  2 },
+    { ISD::TRUNCATE,    MVT::v8i16,  MVT::v8i32,  2 },
+    { ISD::TRUNCATE,    MVT::v8i32,  MVT::v8i64,  4 },
+  };
+
+  static const TypeConversionCostTblEntry<MVT::SimpleValueType>
   AVXConversionTbl[] = {
-    { ISD::SIGN_EXTEND, MVT::v16i16, MVT::v16i8, 1 },
-    { ISD::ZERO_EXTEND, MVT::v16i16, MVT::v16i8, 1 },
-    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
-    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
-    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
-    { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
-    { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64, 1 },
-    { ISD::TRUNCATE,    MVT::v8i16, MVT::v8i32, 1 },
-    { ISD::TRUNCATE,    MVT::v16i8, MVT::v16i16, 2 },
+    { ISD::SIGN_EXTEND, MVT::v16i16, MVT::v16i8, 4 },
+    { ISD::ZERO_EXTEND, MVT::v16i16, MVT::v16i8, 4 },
+    { ISD::SIGN_EXTEND, MVT::v8i32,  MVT::v8i1,  7 },
+    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i1,  4 },
+    { ISD::SIGN_EXTEND, MVT::v8i32,  MVT::v8i8,  7 },
+    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i8,  4 },
+    { ISD::SIGN_EXTEND, MVT::v8i32,  MVT::v8i16, 4 },
+    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i16, 4 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i1,  6 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i1,  4 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i8,  6 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i8,  4 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i16, 6 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i16, 3 },
+    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i32, 4 },
+    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i32, 4 },
+
+    { ISD::TRUNCATE,    MVT::v4i8,  MVT::v4i64,  4 },
+    { ISD::TRUNCATE,    MVT::v4i16, MVT::v4i64,  4 },
+    { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64,  4 },
+    { ISD::TRUNCATE,    MVT::v8i8,  MVT::v8i32,  4 },
+    { ISD::TRUNCATE,    MVT::v8i16, MVT::v8i32,  5 },
+    { ISD::TRUNCATE,    MVT::v16i8, MVT::v16i16, 4 },
+    { ISD::TRUNCATE,    MVT::v8i32, MVT::v8i64,  9 },
 
     { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i1,  8 },
     { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i8,  8 },
@@ -450,14 +492,15 @@ unsigned X86TTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
 
     { ISD::FP_TO_SINT,  MVT::v8i8,  MVT::v8f32, 1 },
     { ISD::FP_TO_SINT,  MVT::v4i8,  MVT::v4f32, 1 },
-    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i1,  6 },
-    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i1,  9 },
-    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i1,  8 },
-    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i8,  6 },
-    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 6 },
-    { ISD::TRUNCATE,    MVT::v8i32, MVT::v8i64, 3 },
   };
 
+  if (ST->hasAVX2()) {
+    int Idx = ConvertCostTableLookup(AVX2ConversionTbl, ISD,
+                                     DstTy.getSimpleVT(), SrcTy.getSimpleVT());
+    if (Idx != -1)
+      return AVX2ConversionTbl[Idx].Cost;
+  }
+
   if (ST->hasAVX()) {
     int Idx = ConvertCostTableLookup(AVXConversionTbl, ISD, DstTy.getSimpleVT(),
                                      SrcTy.getSimpleVT());
diff --git a/llvm/test/Analysis/CostModel/X86/cast.ll b/llvm/test/Analysis/CostModel/X86/cast.ll
index f3c1283c7e3..7f97b176f7c 100644
--- a/llvm/test/Analysis/CostModel/X86/cast.ll
+++ b/llvm/test/Analysis/CostModel/X86/cast.ll
@@ -1,10 +1,11 @@
-; RUN: opt < %s  -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
+; RUN: opt < %s  -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=core-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX2
+; RUN: opt < %s  -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.8.0"
 
 define i32 @add(i32 %arg) {
-
+; CHECK-LABEL: for function 'add'
   ; -- Same size registeres --
   ;CHECK: cost of 1 {{.*}} zext
   %A = zext <4 x i1> undef to <4 x i32>
@@ -33,57 +34,106 @@ define i32 @add(i32 %arg) {
 }
 
 define i32 @zext_sext(<8 x i1> %in) {
-  ;CHECK: cost of 6 {{.*}} zext
+; CHECK-AVX2-LABEL: for function 'zext_sext'
+; CHECK-AVX-LABEL: for function 'zext_sext'
+  ;CHECK-AVX2: cost of 3 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
   %Z = zext <8 x i1> %in to <8 x i32>
-  ;CHECK: cost of 9 {{.*}} sext
+  ;CHECK-AVX2: cost of 3 {{.*}} sext
+  ;CHECK-AVX: cost of 7 {{.*}} sext
   %S = sext <8 x i1> %in to <8 x i32>
 
-  ;CHECK: cost of 1 {{.*}} zext
+  ;CHECK-AVX2: cost of 1 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
   %A1 = zext <16 x i8> undef to <16 x i16>
-  ;CHECK: cost of 1 {{.*}} sext
+  ;CHECK-AVX2: cost of 1 {{.*}} sext
+  ;CHECK-AVX: cost of 4 {{.*}} sext
   %A2 = sext <16 x i8> undef to <16 x i16>
-  ;CHECK: cost of 1 {{.*}} sext
+  ;CHECK-AVX2: cost of 1 {{.*}} sext
+  ;CHECK-AVX: cost of 4 {{.*}} sext
   %A = sext <8 x i16> undef to <8 x i32>
-  ;CHECK: cost of 1 {{.*}} zext
+  ;CHECK-AVX2: cost of 1 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
   %B = zext <8 x i16> undef to <8 x i32>
-  ;CHECK: cost of 1 {{.*}} sext
+  ;CHECK-AVX2: cost of 1 {{.*}} sext
+  ;CHECK-AVX: cost of 4 {{.*}} sext
   %C = sext <4 x i32> undef to <4 x i64>
-  ;CHECK: cost of 6 {{.*}} sext
-  %C1 = sext <4 x i8> undef to <4 x i64>
-  ;CHECK: cost of 6 {{.*}} sext
-  %C2 = sext <4 x i16> undef to <4 x i64>
 
-  ;CHECK: cost of 1 {{.*}} zext
+  ;CHECK-AVX2: cost of 3 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
+  %C.v8i8.z = zext <8 x i8> undef to <8 x i32>
+  ;CHECK-AVX2: cost of 3 {{.*}} sext
+  ;CHECK-AVX: cost of 7 {{.*}} sext
+  %C.v8i8.s = sext <8 x i8> undef to <8 x i32>
+  ;CHECK-AVX2: cost of 3 {{.*}} zext
+  ;CHECK-AVX: cost of 3 {{.*}} zext
+  %C.v4i16.z = zext <4 x i16> undef to <4 x i64>
+  ;CHECK-AVX2: cost of 3 {{.*}} sext
+  ;CHECK-AVX: cost of 6 {{.*}} sext
+  %C.v4i16.s = sext <4 x i16> undef to <4 x i64>
+
+  ;CHECK-AVX2: cost of 3 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
+  %C.v4i8.z = zext <4 x i8> undef to <4 x i64>
+  ;CHECK-AVX2: cost of 3 {{.*}} sext
+  ;CHECK-AVX: cost of 6 {{.*}} sext
+  %C.v4i8.s = sext <4 x i8> undef to <4 x i64>
+
+  ;CHECK-AVX2: cost of 1 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
   %D = zext <4 x i32> undef to <4 x i64>
 
-  ;CHECK: cost of 1 {{.*}} trunc
+  ;CHECK-AVX2: cost of 2 {{.*}} trunc
+  ;CHECK-AVX: cost of 4 {{.*}} trunc
   %E = trunc <4 x i64> undef to <4 x i32>
-  ;CHECK: cost of 1 {{.*}} trunc
+  ;CHECK-AVX2: cost of 2 {{.*}} trunc
+  ;CHECK-AVX: cost of 5 {{.*}} trunc
   %F = trunc <8 x i32> undef to <8 x i16>
-  ;CHECK: cost of 2 {{.*}} trunc
+  ;CHECK-AVX2: cost of 4 {{.*}} trunc
+  ;CHECK-AVX: cost of 4 {{.*}} trunc
   %F1 = trunc <16 x i16> undef to <16 x i8>
-
-  ;CHECK: cost of 3 {{.*}} trunc
+  ;CHECK-AVX2: cost of 2 {{.*}} trunc
+  ;CHECK-AVX: cost of 4 {{.*}} trunc
+  %F2 = trunc <8 x i32> undef to <8 x i8>
+  ;CHECK-AVX2: cost of 2 {{.*}} trunc
+  ;CHECK-AVX: cost of 4 {{.*}} trunc
+  %F3 = trunc <4 x i64> undef to <4 x i8>
+
+  ;CHECK-AVX2: cost of 4 {{.*}} trunc
+  ;CHECK-AVX: cost of 9 {{.*}} trunc
   %G = trunc <8 x i64> undef to <8 x i32>
 
   ret i32 undef
 }
 
 define i32 @masks8(<8 x i1> %in) {
-  ;CHECK: cost of 6 {{.*}} zext
+; CHECK-AVX2-LABEL: for function 'masks8'
+; CHECK-AVX-LABEL: for function 'masks8'
+
+  ;CHECK-AVX2: cost of 3 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
   %Z = zext <8 x i1> %in to <8 x i32>
-  ;CHECK: cost of 9 {{.*}} sext
+  ;CHECK-AVX2: cost of 3 {{.*}} sext
+  ;CHECK-AVX: cost of 7 {{.*}} sext
   %S = sext <8 x i1> %in to <8 x i32>
   ret i32 undef
 }
 
 define i32 @masks4(<4 x i1> %in) {
-  ;CHECK: cost of 8 {{.*}} sext
+; CHECK-AVX2-LABEL: for function 'masks4'
+; CHECK-AVX-LABEL: for function 'masks4'
+
+  ;CHECK-AVX2: cost of 3 {{.*}} zext
+  ;CHECK-AVX: cost of 4 {{.*}} zext
+  %Z = zext <4 x i1> %in to <4 x i64>
+  ;CHECK-AVX2: cost of 3 {{.*}} sext
+  ;CHECK-AVX: cost of 6 {{.*}} sext
   %S = sext <4 x i1> %in to <4 x i64>
   ret i32 undef
 }
 
 define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
+; CHECK-LABEL: for function 'sitofp4'
   ; CHECK: cost of 3 {{.*}} sitofp
   %A1 = sitofp <4 x i1> %a to <4 x float>
   ; CHECK: cost of 3 {{.*}} sitofp
@@ -107,6 +157,7 @@ define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 }
 
 define void @sitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
+; CHECK-LABEL: for function 'sitofp8'
   ; CHECK: cost of 8 {{.*}} sitofp
   %A1 = sitofp <8 x i1> %a to <8 x float>
 
@@ -122,6 +173,7 @@ define void @sitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
 }
 
 define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
+; CHECK-LABEL: for function 'uitofp4'
   ; CHECK: cost of 7 {{.*}} uitofp
   %A1 = uitofp <4 x i1> %a to <4 x float>
   ; CHECK: cost of 7 {{.*}} uitofp
@@ -145,6 +197,7 @@ define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 }
 
 define void @uitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
+; CHECK-LABEL: for function 'uitofp8'
   ; CHECK: cost of 6 {{.*}} uitofp
   %A1 = uitofp <8 x i1> %a to <8 x float>