ARM cost model: Correct cost for some cheap float to integer conversions

Fix cost of some "cheap" cast instructions. Before this patch we used to
estimate for example:
  cost of 16 for instruction:   %r = fptoui <4 x float> %v0 to <4 x i16>

While we would emit:
  vcvt.s32.f32  q8, q8
  vmovn.i32 d16, q8
  vuzp.8  d16, d17

All other costs are left to the values assigned by the fallback logic. Theses
costs are mostly reasonable in the sense that they get progressively more
expensive as the instruction sequences emitted get longer.

radar://13434072

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177333 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMTargetTransformInfo.cpp b/lib/Target/ARM/ARMTargetTransformInfo.cpp
index 7a32ffb7878bab69caf6b979846fdf7dab56723f..38834030d2955bd992b3a7ea149cce3c2cab67ac 100644 (file)
--- a/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -224,12 +224,20 @@ unsigned ARMTTI::getCastInstrCost(unsigned Opcode, Type *Dst,
     { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i32, 1 },
     { ISD::FP_TO_SINT,  MVT::v4i32, MVT::v4f32, 1 },
     { ISD::FP_TO_UINT,  MVT::v4i32, MVT::v4f32, 1 },
+    { ISD::FP_TO_SINT,  MVT::v4i8, MVT::v4f32, 3 },
+    { ISD::FP_TO_UINT,  MVT::v4i8, MVT::v4f32, 3 },
+    { ISD::FP_TO_SINT,  MVT::v4i16, MVT::v4f32, 2 },
+    { ISD::FP_TO_UINT,  MVT::v4i16, MVT::v4f32, 2 },
 
     // Vector double <-> i32 conversions.
     { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
     { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
     { ISD::FP_TO_SINT,  MVT::v2i32, MVT::v2f64, 2 },
-    { ISD::FP_TO_UINT,  MVT::v2i32, MVT::v2f64, 2 }
+    { ISD::FP_TO_UINT,  MVT::v2i32, MVT::v2f64, 2 },
+    { ISD::FP_TO_SINT,  MVT::v8i16, MVT::v8f32, 4 },
+    { ISD::FP_TO_UINT,  MVT::v8i16, MVT::v8f32, 4 },
+    { ISD::FP_TO_SINT,  MVT::v16i16, MVT::v16f32, 8 },
+    { ISD::FP_TO_UINT,  MVT::v16i16, MVT::v16f32, 8 }
   };
 
   if (SrcTy.isVector() && ST->hasNEON()) {

diff --git a/test/Analysis/CostModel/ARM/cast.ll b/test/Analysis/CostModel/ARM/cast.ll
index 2b58f6ce38c54a13f03226fe55c84f37fa7241d2..88b18449471688e6a12fbba32e96b76d6bb43646 100644 (file)
--- a/test/Analysis/CostModel/ARM/cast.ll
+++ b/test/Analysis/CostModel/ARM/cast.ll
@@ -1,4 +1,4 @@
-; RUN: opt < %s  -cost-model -analyze -mtriple=thumbv7-apple-ios6.0.0 -mcpu=swift | FileCheck %s
+; RUN: opt < %s  -cost-model -analyze -mtriple=thumbv7-apple-ios6.0.0 -mcpu=cortex-a8 | FileCheck %s
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
 target triple = "thumbv7-apple-ios6.0.0"
 
@@ -190,6 +190,175 @@ define i32 @casts() {
   ; CHECK: cost of 16 {{.*}} fpext <16 x float
   %r89 = fpext <16 x float> undef to <16 x double>
 
+  ;; Floating point to integer vector casts.
+  ; CHECK: cost of 1 {{.*}} fptoui
+  %r90 = fptoui <2 x float> undef to <2 x i1>
+  ; CHECK: cost of 1 {{.*}} fptosi
+  %r91 = fptosi <2 x float> undef to <2 x i1>
+  ; CHECK: cost of 1 {{.*}} fptoui
+  %r92 = fptoui <2 x float> undef to <2 x i8>
+  ; CHECK: cost of 1 {{.*}} fptosi
+  %r93 = fptosi <2 x float> undef to <2 x i8>
+  ; CHECK: cost of 1 {{.*}} fptoui
+  %r94 = fptoui <2 x float> undef to <2 x i16>
+  ; CHECK: cost of 1 {{.*}} fptosi
+  %r95 = fptosi <2 x float> undef to <2 x i16>
+  ; CHECK: cost of 1 {{.*}} fptoui
+  %r96 = fptoui <2 x float> undef to <2 x i32>
+  ; CHECK: cost of 1 {{.*}} fptosi
+  %r97 = fptosi <2 x float> undef to <2 x i32>
+  ; CHECK: cost of 24 {{.*}} fptoui
+  %r98 = fptoui <2 x float> undef to <2 x i64>
+  ; CHECK: cost of 24 {{.*}} fptosi
+  %r99 = fptosi <2 x float> undef to <2 x i64>
+
+  ; CHECK: cost of 8 {{.*}} fptoui
+  %r100 = fptoui <2 x double> undef to <2 x i1>
+  ; CHECK: cost of 8 {{.*}} fptosi
+  %r101 = fptosi <2 x double> undef to <2 x i1>
+  ; CHECK: cost of 8 {{.*}} fptoui
+  %r102 = fptoui <2 x double> undef to <2 x i8>
+  ; CHECK: cost of 8 {{.*}} fptosi
+  %r103 = fptosi <2 x double> undef to <2 x i8>
+  ; CHECK: cost of 8 {{.*}} fptoui
+  %r104 = fptoui <2 x double> undef to <2 x i16>
+  ; CHECK: cost of 8 {{.*}} fptosi
+  %r105 = fptosi <2 x double> undef to <2 x i16>
+  ; CHECK: cost of 2 {{.*}} fptoui
+  %r106 = fptoui <2 x double> undef to <2 x i32>
+  ; CHECK: cost of 2 {{.*}} fptosi
+  %r107 = fptosi <2 x double> undef to <2 x i32>
+  ; CHECK: cost of 24 {{.*}} fptoui
+  %r108 = fptoui <2 x double> undef to <2 x i64>
+  ; CHECK: cost of 24 {{.*}} fptosi
+  %r109 = fptosi <2 x double> undef to <2 x i64>
+
+  ; CHECK: cost of 16 {{.*}} fptoui
+  %r110 = fptoui <4 x float> undef to <4 x i1>
+  ; CHECK: cost of 16 {{.*}} fptosi
+  %r111 = fptosi <4 x float> undef to <4 x i1>
+  ; CHECK: cost of 3 {{.*}} fptoui
+  %r112 = fptoui <4 x float> undef to <4 x i8>
+  ; CHECK: cost of 3 {{.*}} fptosi
+  %r113 = fptosi <4 x float> undef to <4 x i8>
+  ; CHECK: cost of 2 {{.*}} fptoui
+  %r114 = fptoui <4 x float> undef to <4 x i16>
+  ; CHECK: cost of 2 {{.*}} fptosi
+  %r115 = fptosi <4 x float> undef to <4 x i16>
+  ; CHECK: cost of 1 {{.*}} fptoui
+  %r116 = fptoui <4 x float> undef to <4 x i32>
+  ; CHECK: cost of 1 {{.*}} fptosi
+  %r117 = fptosi <4 x float> undef to <4 x i32>
+  ; CHECK: cost of 48 {{.*}} fptoui
+  %r118 = fptoui <4 x float> undef to <4 x i64>
+  ; CHECK: cost of 48 {{.*}} fptosi
+  %r119 = fptosi <4 x float> undef to <4 x i64>
+
+  ; CHECK: cost of 16 {{.*}} fptoui
+  %r120 = fptoui <4 x double> undef to <4 x i1>
+  ; CHECK: cost of 16 {{.*}} fptosi
+  %r121 = fptosi <4 x double> undef to <4 x i1>
+  ; CHECK: cost of 16 {{.*}} fptoui
+  %r122 = fptoui <4 x double> undef to <4 x i8>
+  ; CHECK: cost of 16 {{.*}} fptosi
+  %r123 = fptosi <4 x double> undef to <4 x i8>
+  ; CHECK: cost of 16 {{.*}} fptoui
+  %r124 = fptoui <4 x double> undef to <4 x i16>
+  ; CHECK: cost of 16 {{.*}} fptosi
+  %r125 = fptosi <4 x double> undef to <4 x i16>
+  ; CHECK: cost of 16 {{.*}} fptoui
+  %r126 = fptoui <4 x double> undef to <4 x i32>
+  ; CHECK: cost of 16 {{.*}} fptosi
+  %r127 = fptosi <4 x double> undef to <4 x i32>
+  ; CHECK: cost of 48 {{.*}} fptoui
+  %r128 = fptoui <4 x double> undef to <4 x i64>
+  ; CHECK: cost of 48 {{.*}} fptosi
+  %r129 = fptosi <4 x double> undef to <4 x i64>
+
+  ; CHECK: cost of 32 {{.*}} fptoui
+  %r130 = fptoui <8 x float> undef to <8 x i1>
+  ; CHECK: cost of 32 {{.*}} fptosi
+  %r131 = fptosi <8 x float> undef to <8 x i1>
+  ; CHECK: cost of 32 {{.*}} fptoui
+  %r132 = fptoui <8 x float> undef to <8 x i8>
+  ; CHECK: cost of 32 {{.*}} fptosi
+  %r133 = fptosi <8 x float> undef to <8 x i8>
+  ; CHECK: cost of 4 {{.*}} fptoui
+  %r134 = fptoui <8 x float> undef to <8 x i16>
+  ; CHECK: cost of 4 {{.*}} fptosi
+  %r135 = fptosi <8 x float> undef to <8 x i16>
+  ; CHECK: cost of 2 {{.*}} fptoui
+  %r136 = fptoui <8 x float> undef to <8 x i32>
+  ; CHECK: cost of 2 {{.*}} fptosi
+  %r137 = fptosi <8 x float> undef to <8 x i32>
+  ; CHECK: cost of 96 {{.*}} fptoui
+  %r138 = fptoui <8 x float> undef to <8 x i64>
+  ; CHECK: cost of 96 {{.*}} fptosi
+  %r139 = fptosi <8 x float> undef to <8 x i64>
+
+  ; CHECK: cost of 32 {{.*}} fptoui
+  %r140 = fptoui <8 x double> undef to <8 x i1>
+  ; CHECK: cost of 32 {{.*}} fptosi
+  %r141 = fptosi <8 x double> undef to <8 x i1>
+  ; CHECK: cost of 32 {{.*}} fptoui
+  %r142 = fptoui <8 x double> undef to <8 x i8>
+  ; CHECK: cost of 32 {{.*}} fptosi
+  %r143 = fptosi <8 x double> undef to <8 x i8>
+  ; CHECK: cost of 32 {{.*}} fptoui
+  %r144 = fptoui <8 x double> undef to <8 x i16>
+  ; CHECK: cost of 32 {{.*}} fptosi
+  %r145 = fptosi <8 x double> undef to <8 x i16>
+  ; CHECK: cost of 32 {{.*}} fptoui
+  %r146 = fptoui <8 x double> undef to <8 x i32>
+  ; CHECK: cost of 32 {{.*}} fptosi
+  %r147 = fptosi <8 x double> undef to <8 x i32>
+  ; CHECK: cost of 96 {{.*}} fptoui
+  %r148 = fptoui <8 x double> undef to <8 x i64>
+  ; CHECK: cost of 96 {{.*}} fptosi
+  %r149 = fptosi <8 x double> undef to <8 x i64>
+
+  ; CHECK: cost of 64 {{.*}} fptoui
+  %r150 = fptoui <16 x float> undef to <16 x i1>
+  ; CHECK: cost of 64 {{.*}} fptosi
+  %r151 = fptosi <16 x float> undef to <16 x i1>
+  ; CHECK: cost of 64 {{.*}} fptoui
+  %r152 = fptoui <16 x float> undef to <16 x i8>
+  ; CHECK: cost of 64 {{.*}} fptosi
+  %r153 = fptosi <16 x float> undef to <16 x i8>
+  ; CHECK: cost of 8 {{.*}} fptoui
+  %r154 = fptoui <16 x float> undef to <16 x i16>
+  ; CHECK: cost of 8 {{.*}} fptosi
+  %r155 = fptosi <16 x float> undef to <16 x i16>
+  ; CHECK: cost of 4 {{.*}} fptoui
+  %r156 = fptoui <16 x float> undef to <16 x i32>
+  ; CHECK: cost of 4 {{.*}} fptosi
+  %r157 = fptosi <16 x float> undef to <16 x i32>
+  ; CHECK: cost of 192 {{.*}} fptoui
+  %r158 = fptoui <16 x float> undef to <16 x i64>
+  ; CHECK: cost of 192 {{.*}} fptosi
+  %r159 = fptosi <16 x float> undef to <16 x i64>
+
+  ; CHECK: cost of 64 {{.*}} fptoui
+  %r160 = fptoui <16 x double> undef to <16 x i1>
+  ; CHECK: cost of 64 {{.*}} fptosi
+  %r161 = fptosi <16 x double> undef to <16 x i1>
+  ; CHECK: cost of 64 {{.*}} fptoui
+  %r162 = fptoui <16 x double> undef to <16 x i8>
+  ; CHECK: cost of 64 {{.*}} fptosi
+  %r163 = fptosi <16 x double> undef to <16 x i8>
+  ; CHECK: cost of 64 {{.*}} fptoui
+  %r164 = fptoui <16 x double> undef to <16 x i16>
+  ; CHECK: cost of 64 {{.*}} fptosi
+  %r165 = fptosi <16 x double> undef to <16 x i16>
+  ; CHECK: cost of 64 {{.*}} fptoui
+  %r166 = fptoui <16 x double> undef to <16 x i32>
+  ; CHECK: cost of 64 {{.*}} fptosi
+  %r167 = fptosi <16 x double> undef to <16 x i32>
+  ; CHECK: cost of 192 {{.*}} fptoui
+  %r168 = fptoui <16 x double> undef to <16 x i64>
+  ; CHECK: cost of 192 {{.*}} fptosi
+  %r169 = fptosi <16 x double> undef to <16 x i64>
+
   ;CHECK: cost of 0 {{.*}} ret
   ret i32 undef
 }