llvm-6502/test/CodeGen/NVPTX/arithmetic-fp-sm20.ll
Justin Holewinski 11ae250ec9 [NVPTX] Improve handling of FP fusion
We now consider the FPOpFusion flag when determining whether
to fuse ops.  We also explicitly emit add.rn when fusion is
disabled to prevent ptxas from fusing the operations on its
own.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213287 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-17 18:10:09 +00:00

73 lines
1.7 KiB
LLVM

; RUN: llc < %s -march=nvptx -mcpu=sm_20 -fp-contract=fast | FileCheck %s
; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -fp-contract=fast | FileCheck %s
;; These tests should run for all targets
;;===-- Basic instruction selection tests ---------------------------------===;;
;;; f64
define double @fadd_f64(double %a, double %b) {
; CHECK: add.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}
; CHECK: ret
%ret = fadd double %a, %b
ret double %ret
}
define double @fsub_f64(double %a, double %b) {
; CHECK: sub.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}
; CHECK: ret
%ret = fsub double %a, %b
ret double %ret
}
define double @fmul_f64(double %a, double %b) {
; CHECK: mul.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}
; CHECK: ret
%ret = fmul double %a, %b
ret double %ret
}
define double @fdiv_f64(double %a, double %b) {
; CHECK: div.rn.f64 %fd{{[0-9]+}}, %fd{{[0-9]+}}, %fd{{[0-9]+}}
; CHECK: ret
%ret = fdiv double %a, %b
ret double %ret
}
;; PTX does not have a floating-point rem instruction
;;; f32
define float @fadd_f32(float %a, float %b) {
; CHECK: add.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
; CHECK: ret
%ret = fadd float %a, %b
ret float %ret
}
define float @fsub_f32(float %a, float %b) {
; CHECK: sub.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
; CHECK: ret
%ret = fsub float %a, %b
ret float %ret
}
define float @fmul_f32(float %a, float %b) {
; CHECK: mul.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
; CHECK: ret
%ret = fmul float %a, %b
ret float %ret
}
define float @fdiv_f32(float %a, float %b) {
; CHECK: div.rn.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
; CHECK: ret
%ret = fdiv float %a, %b
ret float %ret
}
;; PTX does not have a floating-point rem instruction