[llvm] a4b7613 - [X86] Remove setOperationAction for FP_TO_SINT v8i16.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Tue Nov 12 23:00:18 PST 2019
Author: Craig Topper
Date: 2019-11-12T22:45:52-08:00
New Revision: a4b7613a49741e7faa284641b0a9830275371a24
URL: https://github.com/llvm/llvm-project/commit/a4b7613a49741e7faa284641b0a9830275371a24
DIFF: https://github.com/llvm/llvm-project/commit/a4b7613a49741e7faa284641b0a9830275371a24.diff
LOG: [X86] Remove setOperationAction for FP_TO_SINT v8i16.
This is no longer needed after widening legalization as we
custom legalize v8i8 ourselves.
Added entries to the cost model, but bumped the cost slightly
to account for the truncate shuffle that wasn't costed before.
Added:
Modified:
llvm/lib/Target/X86/X86ISelLowering.cpp
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
llvm/test/Analysis/CostModel/X86/fptosi.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 62577162f1d4..e7fe40379b41 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -937,14 +937,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::FP_TO_UINT, MVT::v2i16, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::v4i16, Custom);
- // By marking FP_TO_SINT v8i16 as Custom, will trick type legalization into
- // promoting v8i8 FP_TO_UINT into FP_TO_SINT. When the v8i16 FP_TO_SINT is
- // split again based on the input type, this will cause an AssertSExt i16 to
- // be emitted instead of an AssertZExt. This will allow packssdw followed by
- // packuswb to be used to truncate to v8i8. This is necessary since packusdw
- // isn't available until sse4.1.
- setOperationAction(ISD::FP_TO_SINT, MVT::v8i16, Custom);
-
setOperationAction(ISD::SINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::v2i32, Custom);
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index b4db1cc9bc17..0b3a5319baac 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -1576,6 +1576,9 @@ int X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
{ ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i64, 6 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v2i64, 15 },
+ { ISD::FP_TO_SINT, MVT::v4i16, MVT::v4f32, 2 },
+ { ISD::FP_TO_SINT, MVT::v2i16, MVT::v2f64, 2 },
+
{ ISD::FP_TO_SINT, MVT::v2i32, MVT::v2f64, 3 },
{ ISD::UINT_TO_FP, MVT::f64, MVT::i64, 6 },
diff --git a/llvm/test/Analysis/CostModel/X86/fptosi.ll b/llvm/test/Analysis/CostModel/X86/fptosi.ll
index 2f8ea1352aff..7583d6e60c80 100644
--- a/llvm/test/Analysis/CostModel/X86/fptosi.ll
+++ b/llvm/test/Analysis/CostModel/X86/fptosi.ll
@@ -92,28 +92,28 @@ define i32 @fptosi_double_i32(i32 %arg) {
define i32 @fptosi_double_i16(i32 %arg) {
; SSE-LABEL: 'fptosi_double_i16'
; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi double undef to i16
-; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'fptosi_double_i16'
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi double undef to i16
-; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'fptosi_double_i16'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi double undef to i16
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; BTVER2-LABEL: 'fptosi_double_i16'
; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi double undef to i16
-; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
@@ -235,28 +235,28 @@ define i32 @fptosi_float_i32(i32 %arg) {
define i32 @fptosi_float_i16(i32 %arg) {
; SSE-LABEL: 'fptosi_float_i16'
; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi float undef to i16
-; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'fptosi_float_i16'
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi float undef to i16
-; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'fptosi_float_i16'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi float undef to i16
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; BTVER2-LABEL: 'fptosi_float_i16'
; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi float undef to i16
-; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
More information about the llvm-commits
mailing list