2208c97c1b
Similar to 806761a7629df268c8aed49657aeccffa6bca449 -mtriple= specifies the full target triple while -march= merely sets the architecture part of the default target triple, leaving a target triple which may not make sense. Therefore, -march= is error-prone and not recommended for tests without a target triple. The issue has been benign as we recognize $unknown-apple-darwin as ELF instead of rejecting it outrightly.
29 lines
914 B
LLVM
29 lines
914 B
LLVM
; RUN: llc -mtriple=hexagon < %s | FileCheck %s
|
|
; CHECK: r{{[0-9]*}} += sfmpy(r{{[0-9]*}},r{{[0-9]*}},p{{[0-3]}}):scale
|
|
|
|
target triple = "hexagon"
|
|
|
|
@g0 = private unnamed_addr constant [65 x i8] c"%f : Q6_R_sfmpyacc_RRp_scale(FLT_MIN,FLT_MIN,FLT_MIN,CHAR_MIN)\0A\00", align 1
|
|
|
|
; Function Attrs: nounwind
|
|
declare i32 @f0(ptr, ...) #0
|
|
|
|
; Function Attrs: nounwind
|
|
define i32 @f1() #0 {
|
|
b0:
|
|
%v0 = alloca i32, align 4
|
|
%v1 = alloca i32, align 4
|
|
store i32 0, ptr %v0
|
|
store i32 0, ptr %v1, align 4
|
|
%v2 = call float @llvm.hexagon.F2.sffma.sc(float 0x3810000000000000, float 0x3810000000000000, float 0x3810000000000000, i32 0)
|
|
%v3 = fpext float %v2 to double
|
|
%v4 = call i32 (ptr, ...) @f0(ptr @g0, double %v3) #0
|
|
ret i32 0
|
|
}
|
|
|
|
; Function Attrs: readnone
|
|
declare float @llvm.hexagon.F2.sffma.sc(float, float, float, i32) #1
|
|
|
|
attributes #0 = { nounwind "target-cpu"="hexagonv55" }
|
|
attributes #1 = { readnone }
|