a587ccd87d
fixes https://github.com/llvm/llvm-project/issues/98389 As the issue describes, promoting `llvm.fma.f16` to `llvm.fma.f32` does not work, because there is not enough precision to handle the repeated rounding. `f64` does have sufficient space. So this PR explicitly promotes the 16-bit fma to a 64-bit fma. I could not find examples of a libcall being used for fma, but that's something that could be looked in separately to work around code size issues.
79 lines
2.3 KiB
LLVM
79 lines
2.3 KiB
LLVM
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z14 | FileCheck %s
|
|
|
|
declare double @llvm.fma.f64(double %f1, double %f2, double %f3)
|
|
declare float @llvm.fma.f32(float %f1, float %f2, float %f3)
|
|
declare half @llvm.fma.f16(half %f1, half %f2, half %f3)
|
|
|
|
define double @f1(double %f1, double %f2, double %acc) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK: wfnmadb %f0, %f0, %f2, %f4
|
|
; CHECK: br %r14
|
|
%res = call double @llvm.fma.f64 (double %f1, double %f2, double %acc)
|
|
%negres = fneg double %res
|
|
ret double %negres
|
|
}
|
|
|
|
define double @f2(double %f1, double %f2, double %acc) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK: wfnmsdb %f0, %f0, %f2, %f4
|
|
; CHECK: br %r14
|
|
%negacc = fneg double %acc
|
|
%res = call double @llvm.fma.f64 (double %f1, double %f2, double %negacc)
|
|
%negres = fneg double %res
|
|
ret double %negres
|
|
}
|
|
|
|
define half @f3_half(half %f1, half %f2, half %acc) {
|
|
; CHECK-LABEL: f3_half:
|
|
; CHECK: brasl %r14, __extendhfdf2@PLT
|
|
; CHECK: brasl %r14, __extendhfdf2@PLT
|
|
; CHECK: brasl %r14, __extendhfdf2@PLT
|
|
; CHECK: wfmadb %f0, %f0, %f8, %f10
|
|
; CHECK: brasl %r14, __truncdfhf2@PLT
|
|
; CHECK-NOT: brasl
|
|
; CHECK: lcdfr %f0, %f0
|
|
; CHECK-NEXT: lmg
|
|
; CHECK-NEXT: br %r14
|
|
%res = call half @llvm.fma.f16 (half %f1, half %f2, half %acc)
|
|
%negres = fneg half %res
|
|
ret half %negres
|
|
}
|
|
|
|
define float @f3(float %f1, float %f2, float %acc) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK: wfnmasb %f0, %f0, %f2, %f4
|
|
; CHECK: br %r14
|
|
%res = call float @llvm.fma.f32 (float %f1, float %f2, float %acc)
|
|
%negres = fneg float %res
|
|
ret float %negres
|
|
}
|
|
|
|
define half @f4_half(half %f1, half %f2, half %acc) {
|
|
; CHECK-LABEL: f4_half:
|
|
; CHECK-NOT: brasl
|
|
; CHECK: lcdfr %f0, %f4
|
|
; CHECK: brasl %r14, __extendhfdf2@PLT
|
|
; CHECK: brasl %r14, __extendhfdf2@PLT
|
|
; CHECK: brasl %r14, __extendhfdf2@PLT
|
|
; CHECK: wfmadb %f0, %f0, %f8, %f10
|
|
; CHECK: brasl %r14, __truncdfhf2@PLT
|
|
; CHECK-NOT: brasl
|
|
; CHECK: lcdfr %f0, %f0
|
|
; CHECK-NEXT: lmg
|
|
; CHECK-NEXT: br %r14
|
|
%negacc = fneg half %acc
|
|
%res = call half @llvm.fma.f16 (half %f1, half %f2, half %negacc)
|
|
%negres = fneg half %res
|
|
ret half %negres
|
|
}
|
|
|
|
define float @f4(float %f1, float %f2, float %acc) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK: wfnmssb %f0, %f0, %f2, %f4
|
|
; CHECK: br %r14
|
|
%negacc = fneg float %acc
|
|
%res = call float @llvm.fma.f32 (float %f1, float %f2, float %negacc)
|
|
%negres = fneg float %res
|
|
ret float %negres
|
|
}
|