[PATCH] D118019: Implement correct cost for SVE bitcasts

[Alban Bridonneau via Phabricator via llvm-commits]

alban.bridonneau created this revision.
Herald added subscribers: hiraditya, tschuett.
alban.bridonneau requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

We have some bitcasts which we know will be simplified,
so their cost is zero.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D118019

Files:
  llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
  llvm/test/Analysis/CostModel/AArch64/sve-bitcast.ll


Index: llvm/test/Analysis/CostModel/AArch64/sve-bitcast.ll
===================================================================
--- llvm/test/Analysis/CostModel/AArch64/sve-bitcast.ll
+++ llvm/test/Analysis/CostModel/AArch64/sve-bitcast.ll
@@ -1,8 +1,29 @@
 ; RUN: opt -mtriple=aarch64-linux-gnu -mattr=+sve -cost-model -analyze < %s | FileCheck %s
 
+define <vscale x 2 x i64> @test_nxv2i64(<vscale x 2 x double> %a) {
+; CHECK-LABEL: test_nxv2i64
 ; CHECK: Found an estimated cost of 0 for instruction:   %b = bitcast <vscale x 2 x double> %a to <vscale x 2 x i64>
-
-define <vscale x 2 x i64> @foo(<vscale x 2 x double> %a, i32 %x) {
   %b = bitcast <vscale x 2 x double> %a to <vscale x 2 x i64>
   ret <vscale x 2 x i64> %b
 }
+
+define <vscale x 2 x half> @test_nxv2i16(<vscale x 2 x i16> %a) {
+; CHECK-LABEL: test_nxv2i16
+; CHECK: Found an estimated cost of 0 for instruction:   %b = bitcast <vscale x 2 x i16> %a to <vscale x 2 x half>
+  %b = bitcast <vscale x 2 x i16> %a to <vscale x 2 x half>
+  ret <vscale x 2 x half> %b
+}
+
+define <vscale x 4 x half> @test_nxv4i16(<vscale x 4 x i16> %a) {
+; CHECK-LABEL: test_nxv4i16
+; CHECK: Found an estimated cost of 0 for instruction:   %b = bitcast <vscale x 4 x i16> %a to <vscale x 4 x half>
+  %b = bitcast <vscale x 4 x i16> %a to <vscale x 4 x half>
+  ret <vscale x 4 x half> %b
+}
+
+define <vscale x 2 x float> @test_nxv2i32(<vscale x 2 x i32> %a) {
+; CHECK-LABEL: test_nxv2i32
+; CHECK: Found an estimated cost of 0 for instruction:   %b = bitcast <vscale x 2 x i32> %a to <vscale x 2 x float>
+  %b = bitcast <vscale x 2 x i32> %a to <vscale x 2 x float>
+  ret <vscale x 2 x float> %b
+}
Index: llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -1590,6 +1590,10 @@
     { ISD::FP_EXTEND, MVT::nxv4f64, MVT::nxv4f32, 2},
     { ISD::FP_EXTEND, MVT::nxv8f64, MVT::nxv8f32, 6},
 
+    // Bitcasts
+    { ISD::BITCAST, MVT::nxv2f16, MVT::nxv2i16, 0 },
+    { ISD::BITCAST, MVT::nxv4f16, MVT::nxv4i16, 0 },
+    { ISD::BITCAST, MVT::nxv2f32, MVT::nxv2i32, 0 },
   };
 
   if (const auto *Entry = ConvertCostTableLookup(ConversionTbl, ISD,


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D118019.402432.patch
Type: text/x-patch
Size: 2299 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220124/a64afa55/attachment.bin>