1426 lines
47 KiB
LLVM
1426 lines
47 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=riscv32 -mattr=+zfh \
|
|
; RUN: -verify-machineinstrs -target-abi ilp32f | \
|
|
; RUN: FileCheck -check-prefixes=CHECKIZFH,RV32IZFH,RV32IFZFH %s
|
|
; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=riscv64 -mattr=+zfh \
|
|
; RUN: -verify-machineinstrs -target-abi lp64f | \
|
|
; RUN: FileCheck -check-prefixes=CHECKIZFH,RV64IZFH,RV64IFZFH %s
|
|
; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=riscv32 -mattr=+d \
|
|
; RUN: -mattr=+zfh -verify-machineinstrs -target-abi ilp32d | \
|
|
; RUN: FileCheck -check-prefixes=CHECKIZFH,RV32IZFH,RV32IDZFH %s
|
|
; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=riscv64 -mattr=+d \
|
|
; RUN: -mattr=+zfh -verify-machineinstrs -target-abi lp64d | \
|
|
; RUN: FileCheck -check-prefixes=CHECKIZFH,RV64IZFH,RV64IDZFH %s
|
|
; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=riscv32 \
|
|
; RUN: -verify-machineinstrs | \
|
|
; RUN: FileCheck -check-prefix=RV32I %s
|
|
; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=riscv64 \
|
|
; RUN: -verify-machineinstrs | \
|
|
; RUN: FileCheck -check-prefix=RV64I %s
|
|
|
|
declare half @llvm.sqrt.f16(half)
|
|
|
|
define half @sqrt_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: sqrt_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fsqrt.h fa0, fa0
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: sqrt_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call sqrtf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: sqrt_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call sqrtf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.sqrt.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.powi.f16.i32(half, i32)
|
|
|
|
define half @powi_f16(half %a, i32 %b) nounwind {
|
|
; RV32IZFH-LABEL: powi_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call __powisf2@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: powi_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: sext.w a0, a0
|
|
; RV64IZFH-NEXT: call __powisf2@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: powi_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: mv s0, a1
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, s0
|
|
; RV32I-NEXT: call __powisf2@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: powi_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 0(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: mv s0, a1
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: sext.w a1, s0
|
|
; RV64I-NEXT: call __powisf2@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 0(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.powi.f16.i32(half %a, i32 %b)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.sin.f16(half)
|
|
|
|
define half @sin_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: sin_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call sinf@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: sin_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call sinf@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: sin_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call sinf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: sin_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call sinf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.sin.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.cos.f16(half)
|
|
|
|
define half @cos_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: cos_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call cosf@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: cos_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call cosf@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: cos_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call cosf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: cos_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call cosf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.cos.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
; The sin+cos combination results in an FSINCOS SelectionDAG node.
|
|
define half @sincos_f16(half %a) nounwind {
|
|
; RV32IFZFH-LABEL: sincos_f16:
|
|
; RV32IFZFH: # %bb.0:
|
|
; RV32IFZFH-NEXT: addi sp, sp, -16
|
|
; RV32IFZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IFZFH-NEXT: fsw fs0, 8(sp) # 4-byte Folded Spill
|
|
; RV32IFZFH-NEXT: fsw fs1, 4(sp) # 4-byte Folded Spill
|
|
; RV32IFZFH-NEXT: fcvt.s.h fs0, fa0
|
|
; RV32IFZFH-NEXT: fmv.s fa0, fs0
|
|
; RV32IFZFH-NEXT: call sinf@plt
|
|
; RV32IFZFH-NEXT: fcvt.h.s fs1, fa0
|
|
; RV32IFZFH-NEXT: fmv.s fa0, fs0
|
|
; RV32IFZFH-NEXT: call cosf@plt
|
|
; RV32IFZFH-NEXT: fcvt.h.s ft0, fa0
|
|
; RV32IFZFH-NEXT: fadd.h fa0, fs1, ft0
|
|
; RV32IFZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IFZFH-NEXT: flw fs0, 8(sp) # 4-byte Folded Reload
|
|
; RV32IFZFH-NEXT: flw fs1, 4(sp) # 4-byte Folded Reload
|
|
; RV32IFZFH-NEXT: addi sp, sp, 16
|
|
; RV32IFZFH-NEXT: ret
|
|
;
|
|
; RV64IFZFH-LABEL: sincos_f16:
|
|
; RV64IFZFH: # %bb.0:
|
|
; RV64IFZFH-NEXT: addi sp, sp, -16
|
|
; RV64IFZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IFZFH-NEXT: fsw fs0, 4(sp) # 4-byte Folded Spill
|
|
; RV64IFZFH-NEXT: fsw fs1, 0(sp) # 4-byte Folded Spill
|
|
; RV64IFZFH-NEXT: fcvt.s.h fs0, fa0
|
|
; RV64IFZFH-NEXT: fmv.s fa0, fs0
|
|
; RV64IFZFH-NEXT: call sinf@plt
|
|
; RV64IFZFH-NEXT: fcvt.h.s fs1, fa0
|
|
; RV64IFZFH-NEXT: fmv.s fa0, fs0
|
|
; RV64IFZFH-NEXT: call cosf@plt
|
|
; RV64IFZFH-NEXT: fcvt.h.s ft0, fa0
|
|
; RV64IFZFH-NEXT: fadd.h fa0, fs1, ft0
|
|
; RV64IFZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IFZFH-NEXT: flw fs0, 4(sp) # 4-byte Folded Reload
|
|
; RV64IFZFH-NEXT: flw fs1, 0(sp) # 4-byte Folded Reload
|
|
; RV64IFZFH-NEXT: addi sp, sp, 16
|
|
; RV64IFZFH-NEXT: ret
|
|
;
|
|
; RV32IDZFH-LABEL: sincos_f16:
|
|
; RV32IDZFH: # %bb.0:
|
|
; RV32IDZFH-NEXT: addi sp, sp, -32
|
|
; RV32IDZFH-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
|
|
; RV32IDZFH-NEXT: fsd fs0, 16(sp) # 8-byte Folded Spill
|
|
; RV32IDZFH-NEXT: fsd fs1, 8(sp) # 8-byte Folded Spill
|
|
; RV32IDZFH-NEXT: fcvt.s.h fs0, fa0
|
|
; RV32IDZFH-NEXT: fmv.s fa0, fs0
|
|
; RV32IDZFH-NEXT: call sinf@plt
|
|
; RV32IDZFH-NEXT: fcvt.h.s fs1, fa0
|
|
; RV32IDZFH-NEXT: fmv.s fa0, fs0
|
|
; RV32IDZFH-NEXT: call cosf@plt
|
|
; RV32IDZFH-NEXT: fcvt.h.s ft0, fa0
|
|
; RV32IDZFH-NEXT: fadd.h fa0, fs1, ft0
|
|
; RV32IDZFH-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
|
|
; RV32IDZFH-NEXT: fld fs0, 16(sp) # 8-byte Folded Reload
|
|
; RV32IDZFH-NEXT: fld fs1, 8(sp) # 8-byte Folded Reload
|
|
; RV32IDZFH-NEXT: addi sp, sp, 32
|
|
; RV32IDZFH-NEXT: ret
|
|
;
|
|
; RV64IDZFH-LABEL: sincos_f16:
|
|
; RV64IDZFH: # %bb.0:
|
|
; RV64IDZFH-NEXT: addi sp, sp, -32
|
|
; RV64IDZFH-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
|
|
; RV64IDZFH-NEXT: fsd fs0, 16(sp) # 8-byte Folded Spill
|
|
; RV64IDZFH-NEXT: fsd fs1, 8(sp) # 8-byte Folded Spill
|
|
; RV64IDZFH-NEXT: fcvt.s.h fs0, fa0
|
|
; RV64IDZFH-NEXT: fmv.s fa0, fs0
|
|
; RV64IDZFH-NEXT: call sinf@plt
|
|
; RV64IDZFH-NEXT: fcvt.h.s fs1, fa0
|
|
; RV64IDZFH-NEXT: fmv.s fa0, fs0
|
|
; RV64IDZFH-NEXT: call cosf@plt
|
|
; RV64IDZFH-NEXT: fcvt.h.s ft0, fa0
|
|
; RV64IDZFH-NEXT: fadd.h fa0, fs1, ft0
|
|
; RV64IDZFH-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
|
|
; RV64IDZFH-NEXT: fld fs0, 16(sp) # 8-byte Folded Reload
|
|
; RV64IDZFH-NEXT: fld fs1, 8(sp) # 8-byte Folded Reload
|
|
; RV64IDZFH-NEXT: addi sp, sp, 32
|
|
; RV64IDZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: sincos_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: lui a1, 16
|
|
; RV32I-NEXT: addi s2, a1, -1
|
|
; RV32I-NEXT: and a0, a0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s0, a0
|
|
; RV32I-NEXT: call sinf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: mv a0, s0
|
|
; RV32I-NEXT: call cosf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: mv s0, a0
|
|
; RV32I-NEXT: and a0, s1, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: and a0, s0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, a0
|
|
; RV32I-NEXT: mv a0, s1
|
|
; RV32I-NEXT: call __addsf3@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: sincos_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -32
|
|
; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: lui a1, 16
|
|
; RV64I-NEXT: addiw s2, a1, -1
|
|
; RV64I-NEXT: and a0, a0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s0, a0
|
|
; RV64I-NEXT: call sinf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: mv a0, s0
|
|
; RV64I-NEXT: call cosf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: mv s0, a0
|
|
; RV64I-NEXT: and a0, s1, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: and a0, s0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a1, a0
|
|
; RV64I-NEXT: mv a0, s1
|
|
; RV64I-NEXT: call __addsf3@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 32
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.sin.f16(half %a)
|
|
%2 = call half @llvm.cos.f16(half %a)
|
|
%3 = fadd half %1, %2
|
|
ret half %3
|
|
}
|
|
|
|
declare half @llvm.pow.f16(half, half)
|
|
|
|
define half @pow_f16(half %a, half %b) nounwind {
|
|
; RV32IZFH-LABEL: pow_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: fcvt.s.h fa1, fa1
|
|
; RV32IZFH-NEXT: call powf@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: pow_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: fcvt.s.h fa1, fa1
|
|
; RV64IZFH-NEXT: call powf@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: pow_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: mv s0, a1
|
|
; RV32I-NEXT: lui a1, 16
|
|
; RV32I-NEXT: addi s2, a1, -1
|
|
; RV32I-NEXT: and a0, a0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: and a0, s0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, a0
|
|
; RV32I-NEXT: mv a0, s1
|
|
; RV32I-NEXT: call powf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: pow_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -32
|
|
; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: mv s0, a1
|
|
; RV64I-NEXT: lui a1, 16
|
|
; RV64I-NEXT: addiw s2, a1, -1
|
|
; RV64I-NEXT: and a0, a0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: and a0, s0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a1, a0
|
|
; RV64I-NEXT: mv a0, s1
|
|
; RV64I-NEXT: call powf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 32
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.pow.f16(half %a, half %b)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.exp.f16(half)
|
|
|
|
define half @exp_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: exp_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call expf@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: exp_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call expf@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: exp_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call expf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: exp_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call expf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.exp.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.exp2.f16(half)
|
|
|
|
define half @exp2_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: exp2_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call exp2f@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: exp2_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call exp2f@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: exp2_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call exp2f@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: exp2_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call exp2f@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.exp2.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.log.f16(half)
|
|
|
|
define half @log_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: log_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call logf@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: log_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call logf@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: log_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call logf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: log_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call logf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.log.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.log10.f16(half)
|
|
|
|
define half @log10_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: log10_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call log10f@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: log10_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call log10f@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: log10_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call log10f@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: log10_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call log10f@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.log10.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.log2.f16(half)
|
|
|
|
define half @log2_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: log2_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call log2f@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: log2_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call log2f@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: log2_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call log2f@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: log2_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call log2f@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.log2.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.fma.f16(half, half, half)
|
|
|
|
define half @fma_f16(half %a, half %b, half %c) nounwind {
|
|
; CHECKIZFH-LABEL: fma_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fmadd.h fa0, fa0, fa1, fa2
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: fma_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -32
|
|
; RV32I-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: mv s0, a2
|
|
; RV32I-NEXT: mv s1, a1
|
|
; RV32I-NEXT: lui a1, 16
|
|
; RV32I-NEXT: addi s3, a1, -1
|
|
; RV32I-NEXT: and a0, a0, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s2, a0
|
|
; RV32I-NEXT: and a0, s1, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: and a0, s0, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a2, a0
|
|
; RV32I-NEXT: mv a0, s2
|
|
; RV32I-NEXT: mv a1, s1
|
|
; RV32I-NEXT: call fmaf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 32
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: fma_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -48
|
|
; RV64I-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: mv s0, a2
|
|
; RV64I-NEXT: mv s1, a1
|
|
; RV64I-NEXT: lui a1, 16
|
|
; RV64I-NEXT: addiw s3, a1, -1
|
|
; RV64I-NEXT: and a0, a0, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s2, a0
|
|
; RV64I-NEXT: and a0, s1, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: and a0, s0, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a2, a0
|
|
; RV64I-NEXT: mv a0, s2
|
|
; RV64I-NEXT: mv a1, s1
|
|
; RV64I-NEXT: call fmaf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 48
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.fma.f16(half %a, half %b, half %c)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.fmuladd.f16(half, half, half)
|
|
|
|
define half @fmuladd_f16(half %a, half %b, half %c) nounwind {
|
|
; CHECKIZFH-LABEL: fmuladd_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fmadd.h fa0, fa0, fa1, fa2
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: fmuladd_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -32
|
|
; RV32I-NEXT: sw ra, 28(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 24(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s1, 20(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s2, 16(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s3, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: mv s0, a2
|
|
; RV32I-NEXT: mv s1, a1
|
|
; RV32I-NEXT: lui a1, 16
|
|
; RV32I-NEXT: addi s3, a1, -1
|
|
; RV32I-NEXT: and a0, a0, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s2, a0
|
|
; RV32I-NEXT: and a0, s1, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, a0
|
|
; RV32I-NEXT: mv a0, s2
|
|
; RV32I-NEXT: call __mulsf3@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: and a0, s0, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s0, a0
|
|
; RV32I-NEXT: and a0, s1, s3
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, s0
|
|
; RV32I-NEXT: call __addsf3@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 28(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 24(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s1, 20(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s2, 16(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s3, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 32
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: fmuladd_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -48
|
|
; RV64I-NEXT: sd ra, 40(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 32(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s1, 24(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s2, 16(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s3, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: mv s0, a2
|
|
; RV64I-NEXT: mv s1, a1
|
|
; RV64I-NEXT: lui a1, 16
|
|
; RV64I-NEXT: addiw s3, a1, -1
|
|
; RV64I-NEXT: and a0, a0, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s2, a0
|
|
; RV64I-NEXT: and a0, s1, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a1, a0
|
|
; RV64I-NEXT: mv a0, s2
|
|
; RV64I-NEXT: call __mulsf3@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: and a0, s0, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s0, a0
|
|
; RV64I-NEXT: and a0, s1, s3
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a1, s0
|
|
; RV64I-NEXT: call __addsf3@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 40(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 32(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s1, 24(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s2, 16(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s3, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 48
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.fmuladd.f16(half %a, half %b, half %c)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.fabs.f16(half)
|
|
|
|
define half @fabs_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: fabs_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fabs.h fa0, fa0
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: fabs_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: slli a0, a0, 17
|
|
; RV32I-NEXT: srli a0, a0, 17
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: fabs_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: slli a0, a0, 49
|
|
; RV64I-NEXT: srli a0, a0, 49
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.fabs.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.minnum.f16(half, half)
|
|
|
|
define half @minnum_f16(half %a, half %b) nounwind {
|
|
; CHECKIZFH-LABEL: minnum_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fmin.h fa0, fa0, fa1
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: minnum_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: mv s0, a1
|
|
; RV32I-NEXT: lui a1, 16
|
|
; RV32I-NEXT: addi s2, a1, -1
|
|
; RV32I-NEXT: and a0, a0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: and a0, s0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, a0
|
|
; RV32I-NEXT: mv a0, s1
|
|
; RV32I-NEXT: call fminf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: minnum_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -32
|
|
; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: mv s0, a1
|
|
; RV64I-NEXT: lui a1, 16
|
|
; RV64I-NEXT: addiw s2, a1, -1
|
|
; RV64I-NEXT: and a0, a0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: and a0, s0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a1, a0
|
|
; RV64I-NEXT: mv a0, s1
|
|
; RV64I-NEXT: call fminf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 32
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.minnum.f16(half %a, half %b)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.maxnum.f16(half, half)
|
|
|
|
define half @maxnum_f16(half %a, half %b) nounwind {
|
|
; CHECKIZFH-LABEL: maxnum_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fmax.h fa0, fa0, fa1
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: maxnum_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s1, 4(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: sw s2, 0(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: mv s0, a1
|
|
; RV32I-NEXT: lui a1, 16
|
|
; RV32I-NEXT: addi s2, a1, -1
|
|
; RV32I-NEXT: and a0, a0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv s1, a0
|
|
; RV32I-NEXT: and a0, s0, s2
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: mv a1, a0
|
|
; RV32I-NEXT: mv a0, s1
|
|
; RV32I-NEXT: call fmaxf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s1, 4(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: lw s2, 0(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: maxnum_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -32
|
|
; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: mv s0, a1
|
|
; RV64I-NEXT: lui a1, 16
|
|
; RV64I-NEXT: addiw s2, a1, -1
|
|
; RV64I-NEXT: and a0, a0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv s1, a0
|
|
; RV64I-NEXT: and a0, s0, s2
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: mv a1, a0
|
|
; RV64I-NEXT: mv a0, s1
|
|
; RV64I-NEXT: call fmaxf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 32
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.maxnum.f16(half %a, half %b)
|
|
ret half %1
|
|
}
|
|
|
|
; TODO: FMINNAN and FMAXNAN aren't handled in
|
|
; SelectionDAGLegalize::ExpandNode.
|
|
|
|
; declare half @llvm.minimum.f16(half, half)
|
|
|
|
; define half @fminimum_f16(half %a, half %b) nounwind {
|
|
; %1 = call half @llvm.minimum.f16(half %a, half %b)
|
|
; ret half %1
|
|
; }
|
|
|
|
; declare half @llvm.maximum.f16(half, half)
|
|
|
|
; define half @fmaximum_f16(half %a, half %b) nounwind {
|
|
; %1 = call half @llvm.maximum.f16(half %a, half %b)
|
|
; ret half %1
|
|
; }
|
|
|
|
declare half @llvm.copysign.f16(half, half)
|
|
|
|
define half @copysign_f16(half %a, half %b) nounwind {
|
|
; CHECKIZFH-LABEL: copysign_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, fa0, fa1
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: copysign_f16:
|
|
; RV32I: # %bb.0:
|
|
; 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
|
|
;
|
|
; RV64I-LABEL: copysign_f16:
|
|
; RV64I: # %bb.0:
|
|
; 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
|
|
%1 = call half @llvm.copysign.f16(half %a, half %b)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.floor.f16(half)
|
|
|
|
define half @floor_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: floor_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: lui a0, %hi(.LCPI17_0)
|
|
; CHECKIZFH-NEXT: flh ft0, %lo(.LCPI17_0)(a0)
|
|
; CHECKIZFH-NEXT: fabs.h ft1, fa0
|
|
; CHECKIZFH-NEXT: flt.h a0, ft1, ft0
|
|
; CHECKIZFH-NEXT: beqz a0, .LBB17_2
|
|
; CHECKIZFH-NEXT: # %bb.1:
|
|
; CHECKIZFH-NEXT: fcvt.w.h a0, fa0, rdn
|
|
; CHECKIZFH-NEXT: fcvt.h.w ft0, a0, rdn
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, ft0, fa0
|
|
; CHECKIZFH-NEXT: .LBB17_2:
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: floor_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call floorf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: floor_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call floorf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.floor.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.ceil.f16(half)
|
|
|
|
define half @ceil_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: ceil_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: lui a0, %hi(.LCPI18_0)
|
|
; CHECKIZFH-NEXT: flh ft0, %lo(.LCPI18_0)(a0)
|
|
; CHECKIZFH-NEXT: fabs.h ft1, fa0
|
|
; CHECKIZFH-NEXT: flt.h a0, ft1, ft0
|
|
; CHECKIZFH-NEXT: beqz a0, .LBB18_2
|
|
; CHECKIZFH-NEXT: # %bb.1:
|
|
; CHECKIZFH-NEXT: fcvt.w.h a0, fa0, rup
|
|
; CHECKIZFH-NEXT: fcvt.h.w ft0, a0, rup
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, ft0, fa0
|
|
; CHECKIZFH-NEXT: .LBB18_2:
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: ceil_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call ceilf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: ceil_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call ceilf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.ceil.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.trunc.f16(half)
|
|
|
|
define half @trunc_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: trunc_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: lui a0, %hi(.LCPI19_0)
|
|
; CHECKIZFH-NEXT: flh ft0, %lo(.LCPI19_0)(a0)
|
|
; CHECKIZFH-NEXT: fabs.h ft1, fa0
|
|
; CHECKIZFH-NEXT: flt.h a0, ft1, ft0
|
|
; CHECKIZFH-NEXT: beqz a0, .LBB19_2
|
|
; CHECKIZFH-NEXT: # %bb.1:
|
|
; CHECKIZFH-NEXT: fcvt.w.h a0, fa0, rtz
|
|
; CHECKIZFH-NEXT: fcvt.h.w ft0, a0, rtz
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, ft0, fa0
|
|
; CHECKIZFH-NEXT: .LBB19_2:
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: trunc_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call truncf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: trunc_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call truncf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.trunc.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.rint.f16(half)
|
|
|
|
define half @rint_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: rint_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: lui a0, %hi(.LCPI20_0)
|
|
; CHECKIZFH-NEXT: flh ft0, %lo(.LCPI20_0)(a0)
|
|
; CHECKIZFH-NEXT: fabs.h ft1, fa0
|
|
; CHECKIZFH-NEXT: flt.h a0, ft1, ft0
|
|
; CHECKIZFH-NEXT: beqz a0, .LBB20_2
|
|
; CHECKIZFH-NEXT: # %bb.1:
|
|
; CHECKIZFH-NEXT: fcvt.w.h a0, fa0
|
|
; CHECKIZFH-NEXT: fcvt.h.w ft0, a0
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, ft0, fa0
|
|
; CHECKIZFH-NEXT: .LBB20_2:
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: rint_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call rintf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: rint_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call rintf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.rint.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.nearbyint.f16(half)
|
|
|
|
define half @nearbyint_f16(half %a) nounwind {
|
|
; RV32IZFH-LABEL: nearbyint_f16:
|
|
; RV32IZFH: # %bb.0:
|
|
; RV32IZFH-NEXT: addi sp, sp, -16
|
|
; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV32IZFH-NEXT: call nearbyintf@plt
|
|
; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32IZFH-NEXT: addi sp, sp, 16
|
|
; RV32IZFH-NEXT: ret
|
|
;
|
|
; RV64IZFH-LABEL: nearbyint_f16:
|
|
; RV64IZFH: # %bb.0:
|
|
; RV64IZFH-NEXT: addi sp, sp, -16
|
|
; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
|
|
; RV64IZFH-NEXT: call nearbyintf@plt
|
|
; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
|
|
; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64IZFH-NEXT: addi sp, sp, 16
|
|
; RV64IZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: nearbyint_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call nearbyintf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: nearbyint_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call nearbyintf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.nearbyint.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.round.f16(half)
|
|
|
|
define half @round_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: round_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: lui a0, %hi(.LCPI22_0)
|
|
; CHECKIZFH-NEXT: flh ft0, %lo(.LCPI22_0)(a0)
|
|
; CHECKIZFH-NEXT: fabs.h ft1, fa0
|
|
; CHECKIZFH-NEXT: flt.h a0, ft1, ft0
|
|
; CHECKIZFH-NEXT: beqz a0, .LBB22_2
|
|
; CHECKIZFH-NEXT: # %bb.1:
|
|
; CHECKIZFH-NEXT: fcvt.w.h a0, fa0, rmm
|
|
; CHECKIZFH-NEXT: fcvt.h.w ft0, a0, rmm
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, ft0, fa0
|
|
; CHECKIZFH-NEXT: .LBB22_2:
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: round_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call roundf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: round_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call roundf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.round.f16(half %a)
|
|
ret half %1
|
|
}
|
|
|
|
declare half @llvm.roundeven.f16(half)
|
|
|
|
define half @roundeven_f16(half %a) nounwind {
|
|
; CHECKIZFH-LABEL: roundeven_f16:
|
|
; CHECKIZFH: # %bb.0:
|
|
; CHECKIZFH-NEXT: lui a0, %hi(.LCPI23_0)
|
|
; CHECKIZFH-NEXT: flh ft0, %lo(.LCPI23_0)(a0)
|
|
; CHECKIZFH-NEXT: fabs.h ft1, fa0
|
|
; CHECKIZFH-NEXT: flt.h a0, ft1, ft0
|
|
; CHECKIZFH-NEXT: beqz a0, .LBB23_2
|
|
; CHECKIZFH-NEXT: # %bb.1:
|
|
; CHECKIZFH-NEXT: fcvt.w.h a0, fa0, rne
|
|
; CHECKIZFH-NEXT: fcvt.h.w ft0, a0, rne
|
|
; CHECKIZFH-NEXT: fsgnj.h fa0, ft0, fa0
|
|
; CHECKIZFH-NEXT: .LBB23_2:
|
|
; CHECKIZFH-NEXT: ret
|
|
;
|
|
; RV32I-LABEL: roundeven_f16:
|
|
; RV32I: # %bb.0:
|
|
; RV32I-NEXT: addi sp, sp, -16
|
|
; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
|
|
; RV32I-NEXT: slli a0, a0, 16
|
|
; RV32I-NEXT: srli a0, a0, 16
|
|
; RV32I-NEXT: call __extendhfsf2@plt
|
|
; RV32I-NEXT: call roundevenf@plt
|
|
; RV32I-NEXT: call __truncsfhf2@plt
|
|
; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
|
|
; RV32I-NEXT: addi sp, sp, 16
|
|
; RV32I-NEXT: ret
|
|
;
|
|
; RV64I-LABEL: roundeven_f16:
|
|
; RV64I: # %bb.0:
|
|
; RV64I-NEXT: addi sp, sp, -16
|
|
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
|
|
; RV64I-NEXT: slli a0, a0, 48
|
|
; RV64I-NEXT: srli a0, a0, 48
|
|
; RV64I-NEXT: call __extendhfsf2@plt
|
|
; RV64I-NEXT: call roundevenf@plt
|
|
; RV64I-NEXT: call __truncsfhf2@plt
|
|
; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
|
|
; RV64I-NEXT: addi sp, sp, 16
|
|
; RV64I-NEXT: ret
|
|
%1 = call half @llvm.roundeven.f16(half %a)
|
|
ret half %1
|
|
}
|