; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s \
; RUN: | FileCheck -check-prefix=RV32I %s
; RUN: llc -mtriple=riscv32 -target-abi ilp32 -mattr=+zfh -verify-machineinstrs \
; RUN: < %s | FileCheck -check-prefix=RV32IZFH %s
; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s \
; RUN: | FileCheck -check-prefix=RV64I %s
; RUN: llc -mtriple=riscv64 -target-abi lp64 -mattr=+zfh -verify-machineinstrs \
; RUN: < %s | FileCheck -check-prefix=RV64IZFH %s
; RUN: llc -mtriple=riscv32 -target-abi ilp32 -mattr=+zhinx -verify-machineinstrs \
; RUN: < %s | FileCheck -check-prefix=RV32IZHINX %s
; RUN: llc -mtriple=riscv64 -target-abi lp64 -mattr=+zhinx -verify-machineinstrs \
; RUN: < %s | FileCheck -check-prefix=RV64IZHINX %s
; RUN: llc -mtriple=riscv32 -target-abi ilp32 -mattr=+zfhmin -verify-machineinstrs \
; RUN: < %s | FileCheck -check-prefix=RV32IZFHMIN %s
; RUN: llc -mtriple=riscv64 -target-abi lp64 -verify-machineinstrs < %s \
; RUN: | FileCheck -check-prefix=RV64I %s
; RUN: llc -mtriple=riscv64 -target-abi lp64 -mattr=+zfhmin -verify-machineinstrs \
; RUN: < %s | FileCheck -check-prefix=RV64IZFHMIN %s
; RUN: llc -mtriple=riscv32 -target-abi ilp32 -mattr=+zhinxmin -verify-machineinstrs \
; RUN: < %s | FileCheck --check-prefixes=RVIZHINXMIN,RV32IZHINXMIN %s
; RUN: llc -mtriple=riscv64 -target-abi lp64 -mattr=+zhinxmin -verify-machineinstrs \
; RUN: < %s | FileCheck --check-prefixes=RVIZHINXMIN,RV64IZHINXMIN %s
; This file tests cases where simple floating point operations can be
; profitably handled though bit manipulation if a soft-float ABI is being used
; (e.g. fneg implemented by XORing the sign bit). This is typically handled in
; DAGCombiner::visitBITCAST, but this target-independent code may not trigger
; in cases where we perform custom legalisation (e.g. RV64F).
define half @fneg(half %a) nounwind {
; RV32I-LABEL: fneg:
; RV32I: # %bb.0:
; RV32I-NEXT: lui a1, 1048568
; RV32I-NEXT: xor a0, a0, a1
; RV32I-NEXT: ret
;
; RV32IZFH-LABEL: fneg:
; RV32IZFH: # %bb.0:
; RV32IZFH-NEXT: lui a1, 1048568
; RV32IZFH-NEXT: xor a0, a0, a1
; RV32IZFH-NEXT: ret
;
; RV64I-LABEL: fneg:
; RV64I: # %bb.0:
; RV64I-NEXT: lui a1, 1048568
; RV64I-NEXT: xor a0, a0, a1
; RV64I-NEXT: ret
;
; RV64IZFH-LABEL: fneg:
; RV64IZFH: # %bb.0:
; RV64IZFH-NEXT: lui a1, 1048568
; RV64IZFH-NEXT: xor a0, a0, a1
; RV64IZFH-NEXT: ret
;
; RV32IZHINX-LABEL: fneg:
; RV32IZHINX: # %bb.0:
; RV32IZHINX-NEXT: lui a1, 1048568
; RV32IZHINX-NEXT: xor a0, a0, a1
; RV32IZHINX-NEXT: ret
;
; RV64IZHINX-LABEL: fneg:
; RV64IZHINX: # %bb.0:
; RV64IZHINX-NEXT: lui a1, 1048568
; RV64IZHINX-NEXT: xor a0, a0, a1
; RV64IZHINX-NEXT: ret
;
; RV32IZFHMIN-LABEL: fneg:
; RV32IZFHMIN: # %bb.0:
; RV32IZFHMIN-NEXT: lui a1, 1048568
; RV32IZFHMIN-NEXT: xor a0, a0, a1
; RV32IZFHMIN-NEXT: ret
;
; RV64IZFHMIN-LABEL: fneg:
; RV64IZFHMIN: # %bb.0:
; RV64IZFHMIN-NEXT: lui a1, 1048568
; RV64IZFHMIN-NEXT: xor a0, a0, a1
; RV64IZFHMIN-NEXT: ret
;
; RVIZHINXMIN-LABEL: fneg:
; RVIZHINXMIN: # %bb.0:
; RVIZHINXMIN-NEXT: lui a1, 1048568
; RVIZHINXMIN-NEXT: xor a0, a0, a1
; RVIZHINXMIN-NEXT: ret
%1 = fneg half %a
ret half %1
}
declare half @llvm.fabs.f16(half)
define half @fabs(half %a) nounwind {
; RV32I-LABEL: fabs:
; RV32I: # %bb.0:
; RV32I-NEXT: slli a0, a0, 17
; RV32I-NEXT: srli a0, a0, 17
; RV32I-NEXT: ret
;
; RV32IZFH-LABEL: fabs:
; RV32IZFH: # %bb.0:
; RV32IZFH-NEXT: slli a0, a0, 17
; RV32IZFH-NEXT: srli a0, a0, 17
; RV32IZFH-NEXT: ret
;
; RV64I-LABEL: fabs:
; RV64I: # %bb.0:
; RV64I-NEXT: slli a0, a0, 49
; RV64I-NEXT: srli a0, a0, 49
; RV64I-NEXT: ret
;
; RV64IZFH-LABEL: fabs:
; RV64IZFH: # %bb.0:
; RV64IZFH-NEXT: slli a0, a0, 49
; RV64IZFH-NEXT: srli a0, a0, 49
; RV64IZFH-NEXT: ret
;
; RV32IZHINX-LABEL: fabs:
; RV32IZHINX: # %bb.0:
; RV32IZHINX-NEXT: slli a0, a0, 17
; RV32IZHINX-NEXT: srli a0, a0, 17
; RV32IZHINX-NEXT: ret
;
; RV64IZHINX-LABEL: fabs:
; RV64IZHINX: # %bb.0:
; RV64IZHINX-NEXT: slli a0, a0, 49
; RV64IZHINX-NEXT: srli a0, a0, 49
; RV64IZHINX-NEXT: ret
;
; RV32IZFHMIN-LABEL: fabs:
; RV32IZFHMIN: # %bb.0:
; RV32IZFHMIN-NEXT: slli a0, a0, 17
; RV32IZFHMIN-NEXT: srli a0, a0, 17
; RV32IZFHMIN-NEXT: ret
;
; RV64IZFHMIN-LABEL: fabs:
; RV64IZFHMIN: # %bb.0:
; RV64IZFHMIN-NEXT: slli a0, a0, 49
; RV64IZFHMIN-NEXT: srli a0, a0, 49
; RV64IZFHMIN-NEXT: ret
;
; RV32IZHINXMIN-LABEL: fabs:
; RV32IZHINXMIN: # %bb.0:
; RV32IZHINXMIN-NEXT: slli a0, a0, 17
; RV32IZHINXMIN-NEXT: srli a0, a0, 17
; RV32IZHINXMIN-NEXT: ret
;
; RV64IZHINXMIN-LABEL: fabs:
; RV64IZHINXMIN: # %bb.0:
; RV64IZHINXMIN-NEXT: slli a0, a0, 49
; RV64IZHINXMIN-NEXT: srli a0, a0, 49
; RV64IZHINXMIN-NEXT: ret
%1 = call half @llvm.fabs.f16(half %a)
ret half %1
}
declare half @llvm.copysign.f16(half, half)
; DAGTypeLegalizer::SoftenFloatRes_FCOPYSIGN will convert to bitwise
; operations if half precision floating point isn't supported. A combine could
; be written to do the same even when f16 is legal.
define half @fcopysign_fneg(half %a, half %b) nounwind {
; RV32I-LABEL: fcopysign_fneg:
; RV32I: # %bb.0:
; RV32I-NEXT: not a1, a1
; RV32I-NEXT: lui a2, 1048568
; RV32I-NEXT: and a1, a1, a2
; RV32I-NEXT: slli a0, a0, 17
; RV32I-NEXT: srli a0, a0, 17
; RV32I-NEXT: or a0, a0, a1
; RV32I-NEXT: ret
;
; RV32IZFH-LABEL: fcopysign_fneg:
; RV32IZFH: # %bb.0:
; RV32IZFH-NEXT: fmv.h.x fa5, a1
; RV32IZFH-NEXT: fmv.h.x fa4, a0
; RV32IZFH-NEXT: fsgnjn.h fa5, fa4, fa5
; RV32IZFH-NEXT: fmv.x.h a0, fa5
; RV32IZFH-NEXT: ret
;
; RV64I-LABEL: fcopysign_fneg:
; RV64I: # %bb.0:
; RV64I-NEXT: not a1, a1
; RV64I-NEXT: lui a2, 1048568
; RV64I-NEXT: and a1, a1, a2
; RV64I-NEXT: slli a0, a0, 49
; RV64I-NEXT: srli a0, a0, 49
; RV64I-NEXT: or a0, a0, a1
; RV64I-NEXT: ret
;
; RV64IZFH-LABEL: fcopysign_fneg:
; RV64IZFH: # %bb.0:
; RV64IZFH-NEXT: fmv.h.x fa5, a1
; RV64IZFH-NEXT: fmv.h.x fa4, a0
; RV64IZFH-NEXT: fsgnjn.h fa5, fa4, fa5
; RV64IZFH-NEXT: fmv.x.h a0, fa5
; RV64IZFH-NEXT: ret
;
; RV32IZHINX-LABEL: fcopysign_fneg:
; RV32IZHINX: # %bb.0:
; RV32IZHINX-NEXT: fsgnjn.h a0, a0, a1
; RV32IZHINX-NEXT: ret
;
; RV64IZHINX-LABEL: fcopysign_fneg:
; RV64IZHINX: # %bb.0:
; RV64IZHINX-NEXT: fsgnjn.h a0, a0, a1
; RV64IZHINX-NEXT: ret
;
; RV32IZFHMIN-LABEL: fcopysign_fneg:
; RV32IZFHMIN: # %bb.0:
; RV32IZFHMIN-NEXT: addi sp, sp, -16
; RV32IZFHMIN-NEXT: fmv.h.x fa5, a1
; RV32IZFHMIN-NEXT: fsh fa5, 4(sp)
; RV32IZFHMIN-NEXT: lbu a1, 5(sp)
; RV32IZFHMIN-NEXT: xori a1, a1, 128
; RV32IZFHMIN-NEXT: sb a1, 5(sp)
; RV32IZFHMIN-NEXT: flh fa5, 4(sp)
; RV32IZFHMIN-NEXT: fmv.h.x fa4, a0
; RV32IZFHMIN-NEXT: fsh fa4, 8(sp)
; RV32IZFHMIN-NEXT: fsh fa5, 12(sp)
; RV32IZFHMIN-NEXT: lbu a0, 9(sp)
; RV32IZFHMIN-NEXT: lbu a1, 13(sp)
; RV32IZFHMIN-NEXT: andi a0, a0, 127
; RV32IZFHMIN-NEXT: andi a1, a1, 128
; RV32IZFHMIN-NEXT: or a0, a0, a1
; RV32IZFHMIN-NEXT: sb a0, 9(sp)
; RV32IZFHMIN-NEXT: flh fa5, 8(sp)
; RV32IZFHMIN-NEXT: fmv.x.h a0, fa5
; RV32IZFHMIN-NEXT: addi sp, sp, 16
; RV32IZFHMIN-NEXT: ret
;
; RV64IZFHMIN-LABEL: fcopysign_fneg:
; RV64IZFHMIN: # %bb.0:
; RV64IZFHMIN-NEXT: addi sp, sp, -32
; RV64IZFHMIN-NEXT: fmv.h.x fa5, a1
; RV64IZFHMIN-NEXT: fsh fa5, 8(sp)
; RV64IZFHMIN-NEXT: lbu a1, 9(sp)
; RV64IZFHMIN-NEXT: xori a1, a1, 128
; RV64IZFHMIN-NEXT: sb a1, 9(sp)
; RV64IZFHMIN-NEXT: flh fa5, 8(sp)
; RV64IZFHMIN-NEXT: fmv.h.x fa4, a0
; RV64IZFHMIN-NEXT: fsh fa4, 16(sp)
; RV64IZFHMIN-NEXT: fsh fa5, 24(sp)
; RV64IZFHMIN-NEXT: lbu a0, 17(sp)
; RV64IZFHMIN-NEXT: lbu a1, 25(sp)
; RV64IZFHMIN-NEXT: andi a0, a0, 127
; RV64IZFHMIN-NEXT: andi a1, a1, 128
; RV64IZFHMIN-NEXT: or a0, a0, a1
; RV64IZFHMIN-NEXT: sb a0, 17(sp)
; RV64IZFHMIN-NEXT: flh fa5, 16(sp)
; RV64IZFHMIN-NEXT: fmv.x.h a0, fa5
; RV64IZFHMIN-NEXT: addi sp, sp, 32
; RV64IZFHMIN-NEXT: ret
;
; RV32IZHINXMIN-LABEL: fcopysign_fneg:
; RV32IZHINXMIN: # %bb.0:
; RV32IZHINXMIN-NEXT: addi sp, sp, -16
; RV32IZHINXMIN-NEXT: sh a1, 4(sp)
; RV32IZHINXMIN-NEXT: lbu a1, 5(sp)
; RV32IZHINXMIN-NEXT: xori a1, a1, 128
; RV32IZHINXMIN-NEXT: sb a1, 5(sp)
; RV32IZHINXMIN-NEXT: lh a1, 4(sp)
; RV32IZHINXMIN-NEXT: sh a0, 8(sp)
; RV32IZHINXMIN-NEXT: sh a1, 12(sp)
; RV32IZHINXMIN-NEXT: lbu a0, 9(sp)
; RV32IZHINXMIN-NEXT: lbu a1, 13(sp)
; RV32IZHINXMIN-NEXT: andi a0, a0, 127
; RV32IZHINXMIN-NEXT: andi a1, a1, 128
; RV32IZHINXMIN-NEXT: or a0, a0, a1
; RV32IZHINXMIN-NEXT: sb a0, 9(sp)
; RV32IZHINXMIN-NEXT: lh a0, 8(sp)
; RV32IZHINXMIN-NEXT: addi sp, sp, 16
; RV32IZHINXMIN-NEXT: ret
;
; RV64IZHINXMIN-LABEL: fcopysign_fneg:
; RV64IZHINXMIN: # %bb.0:
; RV64IZHINXMIN-NEXT: addi sp, sp, -32
; RV64IZHINXMIN-NEXT: sh a1, 8(sp)
; RV64IZHINXMIN-NEXT: lbu a1, 9(sp)
; RV64IZHINXMIN-NEXT: xori a1, a1, 128
; RV64IZHINXMIN-NEXT: sb a1, 9(sp)
; RV64IZHINXMIN-NEXT: lh a1, 8(sp)
; RV64IZHINXMIN-NEXT: sh a0, 16(sp)
; RV64IZHINXMIN-NEXT: sh a1, 24(sp)
; RV64IZHINXMIN-NEXT: lbu a0, 17(sp)
; RV64IZHINXMIN-NEXT: lbu a1, 25(sp)
; RV64IZHINXMIN-NEXT: andi a0, a0, 127
; RV64IZHINXMIN-NEXT: andi a1, a1, 128
; RV64IZHINXMIN-NEXT: or a0, a0, a1
; RV64IZHINXMIN-NEXT: sb a0, 17(sp)
; RV64IZHINXMIN-NEXT: lh a0, 16(sp)
; RV64IZHINXMIN-NEXT: addi sp, sp, 32
; RV64IZHINXMIN-NEXT: ret
%1 = fneg half %b
%2 = call half @llvm.copysign.f16(half %a, half %1)
ret half %2
}
Codebase Browser
llvm