RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 12:41:49 +01:00
Code Issues Projects Releases Wiki Activity

[AArch64] Consider instruction-level contract FMFs in combiner patterns.

Browse Source 
Currently, instruction level fast math flags are not considered when
generating patterns for the machine combiner.

This currently leads to some missed opportunities to generate FMAs in
combination with `#pragma clang fp contract (fast)`.

For example, when building the example below with -O3 for AArch64, no
FMADD is generated. If built with -O2 and the DAGCombiner is used
instead of the MachineCombiner for FMAs, an FMADD is generated.

With this patch, the same code is generated in both cases.

    float madd_contract(float a, float b, float c) {
    #pragma clang fp contract (fast)
      return (a * b) + c;
    }

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D84930
This commit is contained in:
Florian Hahn 2020-08-04 10:25:16 +01:00
parent b726ac8e42
commit beaa269335
3 changed files with 21 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Target/AArch64/AArch64InstrInfo.cpp
View File

@ -3861,7 +3861,7 @@ static bool isCombineInstrCandidate64(unsigned Opc) {
return false;
}
// FP Opcodes that can be combined with a FMUL
// FP Opcodes that can be combined with a FMUL.
static bool isCombineInstrCandidateFP(const MachineInstr &Inst) {
switch (Inst.getOpcode()) {
default:
@ -3883,8 +3883,12 @@ static bool isCombineInstrCandidateFP(const MachineInstr &Inst) {
case AArch64::FSUBv2f64:
case AArch64::FSUBv4f32:
TargetOptions Options = Inst.getParent()->getParent()->getTarget().Options;
return (Options.UnsafeFPMath ||
Options.AllowFPOpFusion == FPOpFusion::Fast);
// We can fuse FADD/FSUB with FMUL, if fusion is either allowed globally by
// the target options or if FADD/FSUB has the contract fast-math flag.
return Options.UnsafeFPMath ||
Options.AllowFPOpFusion == FPOpFusion::Fast ||
Inst.getFlag(MachineInstr::FmContract);
return true;
}
return false;
}

32
test/CodeGen/AArch64/machine-combiner-instr-fmf.mir
View File

@ -6,8 +6,7 @@
# CHECK: [[C:%.*]]:fpr32 = COPY $s2
# CHECK-NEXT: [[B:%.*]]:fpr32 = COPY $s1
# CHECK-NEXT: [[A:%.*]]:fpr32 = COPY $s0
# CHECK-NEXT: [[MUL:%.*]]:fpr32 = nnan ninf nsz arcp contract afn reassoc FMULSrr [[B]], [[A]]
# CHECK-NEXT: fpr32 = nnan ninf nsz arcp contract afn reassoc FADDSrr killed [[MUL]], [[C]]
# CHECK-NEXT: :fpr32 = FMADDSrrr [[B]], [[A]], [[C]]
---
name: scalar_fmadd_fast
alignment: 4
@ -46,8 +45,7 @@ body: |
# CHECK: [[C:%.*]]:fpr32 = COPY $s2
# CHECK-NEXT: [[B:%.*]]:fpr32 = COPY $s1
# CHECK-NEXT: [[A:%.*]]:fpr32 = COPY $s0
# CHECK-NEXT: [[MUL:%.*]]:fpr32 = contract FMULSrr [[B]], [[A]]
# CHECK-NEXT: fpr32 = contract FADDSrr killed [[MUL]], [[C]]
# CHECK-NEXT: :fpr32 = FMADDSrrr [[B]], [[A]], [[C]]
---
name: scalar_fmadd_contract
@ -81,7 +79,7 @@ body: |
...
# Do not create FMADD, because we don't have the contract flag on moth instructions.
# Do not create FMADD, because we don't have the contract flag on the FADD.
# CHECK-LABEL: name: scalar_fmadd_contract_op0
# CHECK: [[C:%.*]]:fpr32 = COPY $s2
@ -121,14 +119,13 @@ body: |
...
# Do not create FMADD, because we don't have the contract flag on moth instructions.
# Do create FMADD, because we have the contract flag on the FADD.
#
# CHECK-LABEL: name: scalar_fmadd_contract_op1
# CHECK: [[C:%.*]]:fpr32 = COPY $s2
# CHECK-NEXT: [[B:%.*]]:fpr32 = COPY $s1
# CHECK-NEXT: [[A:%.*]]:fpr32 = COPY $s0
# CHECK-NEXT: [[MUL:%.*]]:fpr32 = FMULSrr [[B]], [[A]]
# CHECK-NEXT: fpr32 = contract FADDSrr killed [[MUL]], [[C]]
# CHECK-NEXT: :fpr32 = FMADDSrrr [[B]], [[A]], [[C]]
---
name: scalar_fmadd_contract_op1
@ -203,14 +200,13 @@ body: |
...
# Can create FMADD, because both the fmul and fadd have all fast-math flags.
# Can create FMLA, because both the fmul and fadd have all fast-math flags.
#
# CHECK-LABEL: name: vector_fmadd_fast
# CHECK: [[C:%.*]]:fpr128 = COPY $q2
# CHECK-NEXT: [[B:%.*]]:fpr128 = COPY $q1
# CHECK-NEXT: [[A:%.*]]:fpr128 = COPY $q0
# CHECK-NEXT: [[MUL:%.*]]:fpr128 = nnan ninf nsz arcp contract afn reassoc FMULv2f64 [[B]], [[A]]
# CHECK-NEXT: fpr128 = nnan ninf nsz arcp contract afn reassoc FADDv2f64 killed [[MUL]], [[C]]
# CHECK-NEXT: fpr128 = FMLAv2f64 [[C]], [[B]], [[A]]
---
name: vector_fmadd_fast
alignment: 4
@ -243,14 +239,13 @@ body: |
...
# Can create FMADD, because both the fmul and fadd have the contract fast-math flag.
# Can create FMLA, because both the fmul and fadd have the contract fast-math flag.
#
# CHECK-LABEL: name: vector_fmadd_contract
# CHECK: [[C:%.*]]:fpr128 = COPY $q2
# CHECK-NEXT: [[B:%.*]]:fpr128 = COPY $q1
# CHECK-NEXT: [[A:%.*]]:fpr128 = COPY $q0
# CHECK-NEXT: [[MUL:%.*]]:fpr128 = contract FMULv2f64 [[B]], [[A]]
# CHECK-NEXT: fpr128 = contract FADDv2f64 killed [[MUL]], [[C]]
# CHECK-NEXT: fpr128 = FMLAv2f64 [[C]], [[B]], [[A]]
---
name: vector_fmadd_contract
alignment: 4
@ -283,7 +278,7 @@ body: |
...
# Do not create FMADD, because we don't have the contract flag on moth instructions.
# Do not create FMLA, because we don't have the contract flag on the FADD.
#
# CHECK-LABEL: name: vector_fmadd_contract_op0
# CHECK: [[C:%.*]]:fpr128 = COPY $q2
@ -323,14 +318,13 @@ body: |
...
# Do not create FMADD, because we don't have the contract flag on moth instructions.
# Do create FMLA, because we have the contract flag on the FADD.
#
# CHECK-LABEL: name: vector_fmadd_contract_op1
# CHECK: [[C:%.*]]:fpr128 = COPY $q2
# CHECK-NEXT: [[B:%.*]]:fpr128 = COPY $q1
# CHECK-NEXT: [[A:%.*]]:fpr128 = COPY $q0
# CHECK-NEXT: [[MUL:%.*]]:fpr128 = FMULv2f64 [[B]], [[A]]
# CHECK-NEXT: fpr128 = contract FADDv2f64 killed [[MUL]], [[C]]
# CHECK-NEXT: fpr128 = FMLAv2f64 [[C]], [[B]], [[A]]
---
name: vector_fmadd_contract_op1
@ -364,7 +358,7 @@ body: |
...
# Do not create FMADD, as nsz flag does not allow it.
# Do not create FMLA, as nsz flag does not allow it.
#
# CHECK-LABEL: name: vector_fmadd_nsz
# CHECK: [[C:%.*]]:fpr128 = COPY $q2

1
test/CodeGen/AArch64/neon-fma-FMF.ll
View File

@ -1,4 +1,5 @@
; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s
; RUN: llc < %s -O3 -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s
define <2 x float> @fma_1(<2 x float> %A, <2 x float> %B, <2 x float> %C) {
; CHECK-LABEL: fma_1: