0f1b5f115a
ptxas is a proprietary compiler from Nvidia that can compile PTX to machine code (SASS). It has a lot of diagnostics to catch errors in PTX, which can be used to verify PTX output from llc. Set -DPXTAS_EXECUTABLE=/path/to/ptxas CMake option to enable it. If this option is not set, then ptxas is substituted to true which effectively disables all ptxas RUN lines. LLVM_PTXAS_EXECUTABLE environment variable takes precedence over the CMake option, and allows to override ptxas executable that is used for LIT without complete re-configuration. Differential Revision: https://reviews.llvm.org/D121727
76 lines
2.4 KiB
LLVM
76 lines
2.4 KiB
LLVM
; RUN: llc < %s -march=nvptx | FileCheck %s
|
|
; RUN: %if ptxas %{ llc < %s -march=nvptx | %ptxas-verify %}
|
|
|
|
; CHECK: .b8 half_array[8] = {1, 2, 3, 4, 5, 6, 7, 8};
|
|
@"half_array" = addrspace(1) constant [4 x half]
|
|
[half 0xH0201, half 0xH0403, half 0xH0605, half 0xH0807]
|
|
|
|
define void @test_load_store(half addrspace(1)* %in, half addrspace(1)* %out) {
|
|
; CHECK-LABEL: @test_load_store
|
|
; CHECK: ld.global.b16 [[TMP:%h[0-9]+]], [{{%r[0-9]+}}]
|
|
; CHECK: st.global.b16 [{{%r[0-9]+}}], [[TMP]]
|
|
%val = load half, half addrspace(1)* %in
|
|
store half %val, half addrspace(1) * %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_bitcast_from_half(half addrspace(1)* %in, i16 addrspace(1)* %out) {
|
|
; CHECK-LABEL: @test_bitcast_from_half
|
|
; CHECK: ld.global.b16 [[TMP:%h[0-9]+]], [{{%r[0-9]+}}]
|
|
; CHECK: st.global.b16 [{{%r[0-9]+}}], [[TMP]]
|
|
%val = load half, half addrspace(1) * %in
|
|
%val_int = bitcast half %val to i16
|
|
store i16 %val_int, i16 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_bitcast_to_half(half addrspace(1)* %out, i16 addrspace(1)* %in) {
|
|
; CHECK-LABEL: @test_bitcast_to_half
|
|
; CHECK: ld.global.u16 [[TMP:%rs[0-9]+]], [{{%r[0-9]+}}]
|
|
; CHECK: st.global.u16 [{{%r[0-9]+}}], [[TMP]]
|
|
%val = load i16, i16 addrspace(1)* %in
|
|
%val_fp = bitcast i16 %val to half
|
|
store half %val_fp, half addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_extend32(half addrspace(1)* %in, float addrspace(1)* %out) {
|
|
; CHECK-LABEL: @test_extend32
|
|
; CHECK: cvt.f32.f16
|
|
|
|
%val16 = load half, half addrspace(1)* %in
|
|
%val32 = fpext half %val16 to float
|
|
store float %val32, float addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_extend64(half addrspace(1)* %in, double addrspace(1)* %out) {
|
|
; CHECK-LABEL: @test_extend64
|
|
; CHECK: cvt.f64.f16
|
|
|
|
%val16 = load half, half addrspace(1)* %in
|
|
%val64 = fpext half %val16 to double
|
|
store double %val64, double addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_trunc32(float addrspace(1)* %in, half addrspace(1)* %out) {
|
|
; CHECK-LABEL: test_trunc32
|
|
; CHECK: cvt.rn.f16.f32
|
|
|
|
%val32 = load float, float addrspace(1)* %in
|
|
%val16 = fptrunc float %val32 to half
|
|
store half %val16, half addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_trunc64(double addrspace(1)* %in, half addrspace(1)* %out) {
|
|
; CHECK-LABEL: @test_trunc64
|
|
; CHECK: cvt.rn.f16.f64
|
|
|
|
%val32 = load double, double addrspace(1)* %in
|
|
%val16 = fptrunc double %val32 to half
|
|
store half %val16, half addrspace(1)* %out
|
|
ret void
|
|
}
|