52 lines
2.2 KiB
LLVM
52 lines
2.2 KiB
LLVM
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
||
|
|
; RUN: llc < %s -global-isel=0 -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefixes=GFX10,GFX10-SDAG
|
||
|
|
; RUN: llc < %s -global-isel=1 -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefixes=GFX10,GFX10-GISEL
|
||
|
|
|
||
|
|
declare double @llvm.amdgcn.flat.atomic.fmin.f64.p1.f64(ptr %ptr, double %data)
|
||
|
|
declare double @llvm.amdgcn.flat.atomic.fmax.f64.p1.f64(ptr %ptr, double %data)
|
||
|
|
|
||
|
|
define amdgpu_cs void @flat_atomic_fmin_f64_noret(ptr %ptr, double %data) {
|
||
|
|
; GFX10-LABEL: flat_atomic_fmin_f64_noret:
|
||
|
|
; GFX10: ; %bb.0:
|
||
|
|
; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3]
|
||
|
|
; GFX10-NEXT: s_endpgm
|
||
|
|
%ret = call double @llvm.amdgcn.flat.atomic.fmin.f64.p1.f64(ptr %ptr, double %data)
|
||
|
|
ret void
|
||
|
|
}
|
||
|
|
|
||
|
|
define amdgpu_cs void @flat_atomic_fmax_f64_noret(ptr %ptr, double %data) {
|
||
|
|
; GFX10-LABEL: flat_atomic_fmax_f64_noret:
|
||
|
|
; GFX10: ; %bb.0:
|
||
|
|
; GFX10-NEXT: flat_atomic_fmax_x2 v[0:1], v[2:3]
|
||
|
|
; GFX10-NEXT: s_endpgm
|
||
|
|
%ret = call double @llvm.amdgcn.flat.atomic.fmax.f64.p1.f64(ptr %ptr, double %data)
|
||
|
|
ret void
|
||
|
|
}
|
||
|
|
|
||
|
|
define amdgpu_cs void @flat_atomic_fmin_f64_rtn(ptr %ptr, double %data, ptr %out) {
|
||
|
|
; GFX10-LABEL: flat_atomic_fmin_f64_rtn:
|
||
|
|
; GFX10: ; %bb.0:
|
||
|
|
; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc
|
||
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
||
|
|
; GFX10-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
|
||
|
|
; GFX10-NEXT: s_endpgm
|
||
|
|
%ret = call double @llvm.amdgcn.flat.atomic.fmin.f64.p1.f64(ptr %ptr, double %data)
|
||
|
|
store double %ret, ptr %out
|
||
|
|
ret void
|
||
|
|
}
|
||
|
|
|
||
|
|
define amdgpu_cs void @flat_atomic_fmax_f64_rtn(ptr %ptr, double %data, ptr %out) {
|
||
|
|
; GFX10-LABEL: flat_atomic_fmax_f64_rtn:
|
||
|
|
; GFX10: ; %bb.0:
|
||
|
|
; GFX10-NEXT: flat_atomic_fmax_x2 v[0:1], v[0:1], v[2:3] glc
|
||
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
||
|
|
; GFX10-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
|
||
|
|
; GFX10-NEXT: s_endpgm
|
||
|
|
%ret = call double @llvm.amdgcn.flat.atomic.fmax.f64.p1.f64(ptr %ptr, double %data)
|
||
|
|
store double %ret, ptr %out
|
||
|
|
ret void
|
||
|
|
}
|
||
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
||
|
|
; GFX10-GISEL: {{.*}}
|
||
|
|
; GFX10-SDAG: {{.*}}
|