CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
233 lines
9.3 KiB
LLVM
233 lines
9.3 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_90 | FileCheck %s
|
|
; RUN: %if ptxas-sm_90 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_90 | %ptxas-verify -arch=sm_90 %}
|
|
|
|
%struct.64 = type <{ i64 }>
|
|
declare i64 @callee(ptr %p);
|
|
declare i64 @callee_variadic(ptr %p, ...);
|
|
|
|
define %struct.64 @test_return_type_mismatch(ptr %p) {
|
|
; CHECK-LABEL: test_return_type_mismatch(
|
|
; CHECK: {
|
|
; CHECK-NEXT: .reg .b64 %rd<32>;
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: // %bb.0:
|
|
; CHECK-NEXT: ld.param.b64 %rd1, [test_return_type_mismatch_param_0];
|
|
; CHECK-NEXT: { // callseq 0, 0
|
|
; CHECK-NEXT: .param .b64 param0;
|
|
; CHECK-NEXT: .param .align 1 .b8 retval0[8];
|
|
; CHECK-NEXT: st.param.b64 [param0], %rd1;
|
|
; CHECK-NEXT: prototype_0 : .callprototype (.param .align 1 .b8 _[8]) _ (.param .b64 _);
|
|
; CHECK-NEXT: mov.b64 %rd2, callee;
|
|
; CHECK-NEXT: call (retval0), %rd2, (param0), prototype_0;
|
|
; CHECK-NEXT: ld.param.b8 %rd3, [retval0+7];
|
|
; CHECK-NEXT: ld.param.b8 %rd4, [retval0+6];
|
|
; CHECK-NEXT: ld.param.b8 %rd5, [retval0+5];
|
|
; CHECK-NEXT: ld.param.b8 %rd6, [retval0+4];
|
|
; CHECK-NEXT: ld.param.b8 %rd7, [retval0+3];
|
|
; CHECK-NEXT: ld.param.b8 %rd8, [retval0+2];
|
|
; CHECK-NEXT: ld.param.b8 %rd9, [retval0+1];
|
|
; CHECK-NEXT: ld.param.b8 %rd10, [retval0];
|
|
; CHECK-NEXT: } // callseq 0
|
|
; CHECK-NEXT: shl.b64 %rd11, %rd9, 8;
|
|
; CHECK-NEXT: or.b64 %rd12, %rd11, %rd10;
|
|
; CHECK-NEXT: shl.b64 %rd13, %rd8, 16;
|
|
; CHECK-NEXT: shl.b64 %rd14, %rd7, 24;
|
|
; CHECK-NEXT: or.b64 %rd15, %rd14, %rd13;
|
|
; CHECK-NEXT: or.b64 %rd16, %rd15, %rd12;
|
|
; CHECK-NEXT: shl.b64 %rd17, %rd5, 8;
|
|
; CHECK-NEXT: or.b64 %rd18, %rd17, %rd6;
|
|
; CHECK-NEXT: shl.b64 %rd19, %rd4, 16;
|
|
; CHECK-NEXT: shl.b64 %rd20, %rd3, 24;
|
|
; CHECK-NEXT: or.b64 %rd21, %rd20, %rd19;
|
|
; CHECK-NEXT: or.b64 %rd22, %rd21, %rd18;
|
|
; CHECK-NEXT: shl.b64 %rd23, %rd22, 32;
|
|
; CHECK-NEXT: or.b64 %rd24, %rd23, %rd16;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0], %rd10;
|
|
; CHECK-NEXT: shr.u64 %rd25, %rd24, 56;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+7], %rd25;
|
|
; CHECK-NEXT: shr.u64 %rd26, %rd24, 48;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+6], %rd26;
|
|
; CHECK-NEXT: shr.u64 %rd27, %rd24, 40;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+5], %rd27;
|
|
; CHECK-NEXT: shr.u64 %rd28, %rd24, 32;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+4], %rd28;
|
|
; CHECK-NEXT: shr.u64 %rd29, %rd24, 24;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+3], %rd29;
|
|
; CHECK-NEXT: shr.u64 %rd30, %rd24, 16;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+2], %rd30;
|
|
; CHECK-NEXT: shr.u64 %rd31, %rd24, 8;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+1], %rd31;
|
|
; CHECK-NEXT: ret;
|
|
%ret = call %struct.64 @callee(ptr %p)
|
|
ret %struct.64 %ret
|
|
}
|
|
|
|
define i64 @test_param_type_mismatch(ptr %p) {
|
|
; CHECK-LABEL: test_param_type_mismatch(
|
|
; CHECK: {
|
|
; CHECK-NEXT: .reg .b64 %rd<3>;
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: // %bb.0:
|
|
; CHECK-NEXT: { // callseq 1, 0
|
|
; CHECK-NEXT: .param .b64 param0;
|
|
; CHECK-NEXT: .param .b64 retval0;
|
|
; CHECK-NEXT: prototype_1 : .callprototype (.param .b64 _) _ (.param .b64 _);
|
|
; CHECK-NEXT: st.param.b64 [param0], 7;
|
|
; CHECK-NEXT: mov.b64 %rd1, callee;
|
|
; CHECK-NEXT: call (retval0), %rd1, (param0), prototype_1;
|
|
; CHECK-NEXT: ld.param.b64 %rd2, [retval0];
|
|
; CHECK-NEXT: } // callseq 1
|
|
; CHECK-NEXT: st.param.b64 [func_retval0], %rd2;
|
|
; CHECK-NEXT: ret;
|
|
%ret = call i64 @callee(i64 7)
|
|
ret i64 %ret
|
|
}
|
|
|
|
define i64 @test_param_count_mismatch(ptr %p) {
|
|
; CHECK-LABEL: test_param_count_mismatch(
|
|
; CHECK: {
|
|
; CHECK-NEXT: .reg .b64 %rd<4>;
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: // %bb.0:
|
|
; CHECK-NEXT: ld.param.b64 %rd1, [test_param_count_mismatch_param_0];
|
|
; CHECK-NEXT: { // callseq 2, 0
|
|
; CHECK-NEXT: .param .b64 param0;
|
|
; CHECK-NEXT: .param .b64 param1;
|
|
; CHECK-NEXT: .param .b64 retval0;
|
|
; CHECK-NEXT: st.param.b64 [param0], %rd1;
|
|
; CHECK-NEXT: prototype_2 : .callprototype (.param .b64 _) _ (.param .b64 _, .param .b64 _);
|
|
; CHECK-NEXT: st.param.b64 [param1], 7;
|
|
; CHECK-NEXT: mov.b64 %rd2, callee;
|
|
; CHECK-NEXT: call (retval0), %rd2, (param0, param1), prototype_2;
|
|
; CHECK-NEXT: ld.param.b64 %rd3, [retval0];
|
|
; CHECK-NEXT: } // callseq 2
|
|
; CHECK-NEXT: st.param.b64 [func_retval0], %rd3;
|
|
; CHECK-NEXT: ret;
|
|
%ret = call i64 @callee(ptr %p, i64 7)
|
|
ret i64 %ret
|
|
}
|
|
|
|
define %struct.64 @test_return_type_mismatch_variadic(ptr %p) {
|
|
; CHECK-LABEL: test_return_type_mismatch_variadic(
|
|
; CHECK: {
|
|
; CHECK-NEXT: .local .align 1 .b8 __local_depot3[1];
|
|
; CHECK-NEXT: .reg .b64 %SP;
|
|
; CHECK-NEXT: .reg .b64 %SPL;
|
|
; CHECK-NEXT: .reg .b64 %rd<33>;
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: // %bb.0:
|
|
; CHECK-NEXT: mov.b64 %SPL, __local_depot3;
|
|
; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
|
|
; CHECK-NEXT: ld.param.b64 %rd1, [test_return_type_mismatch_variadic_param_0];
|
|
; CHECK-NEXT: { // callseq 3, 0
|
|
; CHECK-NEXT: .param .b64 param0;
|
|
; CHECK-NEXT: .param .b64 param1;
|
|
; CHECK-NEXT: .param .align 1 .b8 retval0[8];
|
|
; CHECK-NEXT: add.u64 %rd2, %SP, 0;
|
|
; CHECK-NEXT: st.param.b64 [param1], %rd2;
|
|
; CHECK-NEXT: st.param.b64 [param0], %rd1;
|
|
; CHECK-NEXT: prototype_3 : .callprototype (.param .align 1 .b8 _[8]) _ (.param .b64 _, .param .b64 _);
|
|
; CHECK-NEXT: mov.b64 %rd3, callee_variadic;
|
|
; CHECK-NEXT: call (retval0), %rd3, (param0, param1), prototype_3;
|
|
; CHECK-NEXT: ld.param.b8 %rd4, [retval0+7];
|
|
; CHECK-NEXT: ld.param.b8 %rd5, [retval0+6];
|
|
; CHECK-NEXT: ld.param.b8 %rd6, [retval0+5];
|
|
; CHECK-NEXT: ld.param.b8 %rd7, [retval0+4];
|
|
; CHECK-NEXT: ld.param.b8 %rd8, [retval0+3];
|
|
; CHECK-NEXT: ld.param.b8 %rd9, [retval0+2];
|
|
; CHECK-NEXT: ld.param.b8 %rd10, [retval0+1];
|
|
; CHECK-NEXT: ld.param.b8 %rd11, [retval0];
|
|
; CHECK-NEXT: } // callseq 3
|
|
; CHECK-NEXT: shl.b64 %rd12, %rd10, 8;
|
|
; CHECK-NEXT: or.b64 %rd13, %rd12, %rd11;
|
|
; CHECK-NEXT: shl.b64 %rd14, %rd9, 16;
|
|
; CHECK-NEXT: shl.b64 %rd15, %rd8, 24;
|
|
; CHECK-NEXT: or.b64 %rd16, %rd15, %rd14;
|
|
; CHECK-NEXT: or.b64 %rd17, %rd16, %rd13;
|
|
; CHECK-NEXT: shl.b64 %rd18, %rd6, 8;
|
|
; CHECK-NEXT: or.b64 %rd19, %rd18, %rd7;
|
|
; CHECK-NEXT: shl.b64 %rd20, %rd5, 16;
|
|
; CHECK-NEXT: shl.b64 %rd21, %rd4, 24;
|
|
; CHECK-NEXT: or.b64 %rd22, %rd21, %rd20;
|
|
; CHECK-NEXT: or.b64 %rd23, %rd22, %rd19;
|
|
; CHECK-NEXT: shl.b64 %rd24, %rd23, 32;
|
|
; CHECK-NEXT: or.b64 %rd25, %rd24, %rd17;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0], %rd11;
|
|
; CHECK-NEXT: shr.u64 %rd26, %rd25, 56;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+7], %rd26;
|
|
; CHECK-NEXT: shr.u64 %rd27, %rd25, 48;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+6], %rd27;
|
|
; CHECK-NEXT: shr.u64 %rd28, %rd25, 40;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+5], %rd28;
|
|
; CHECK-NEXT: shr.u64 %rd29, %rd25, 32;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+4], %rd29;
|
|
; CHECK-NEXT: shr.u64 %rd30, %rd25, 24;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+3], %rd30;
|
|
; CHECK-NEXT: shr.u64 %rd31, %rd25, 16;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+2], %rd31;
|
|
; CHECK-NEXT: shr.u64 %rd32, %rd25, 8;
|
|
; CHECK-NEXT: st.param.b8 [func_retval0+1], %rd32;
|
|
; CHECK-NEXT: ret;
|
|
%ret = call %struct.64 (ptr, ...) @callee_variadic(ptr %p)
|
|
ret %struct.64 %ret
|
|
}
|
|
|
|
define i64 @test_param_type_mismatch_variadic(ptr %p) {
|
|
; CHECK-LABEL: test_param_type_mismatch_variadic(
|
|
; CHECK: {
|
|
; CHECK-NEXT: .local .align 8 .b8 __local_depot4[8];
|
|
; CHECK-NEXT: .reg .b64 %SP;
|
|
; CHECK-NEXT: .reg .b64 %SPL;
|
|
; CHECK-NEXT: .reg .b64 %rd<4>;
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: // %bb.0:
|
|
; CHECK-NEXT: mov.b64 %SPL, __local_depot4;
|
|
; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
|
|
; CHECK-NEXT: ld.param.b64 %rd1, [test_param_type_mismatch_variadic_param_0];
|
|
; CHECK-NEXT: st.b64 [%SP], 7;
|
|
; CHECK-NEXT: { // callseq 4, 0
|
|
; CHECK-NEXT: .param .b64 param0;
|
|
; CHECK-NEXT: .param .b64 param1;
|
|
; CHECK-NEXT: .param .b64 retval0;
|
|
; CHECK-NEXT: add.u64 %rd2, %SP, 0;
|
|
; CHECK-NEXT: st.param.b64 [param1], %rd2;
|
|
; CHECK-NEXT: st.param.b64 [param0], %rd1;
|
|
; CHECK-NEXT: call.uni (retval0), callee_variadic, (param0, param1);
|
|
; CHECK-NEXT: ld.param.b64 %rd3, [retval0];
|
|
; CHECK-NEXT: } // callseq 4
|
|
; CHECK-NEXT: st.param.b64 [func_retval0], %rd3;
|
|
; CHECK-NEXT: ret;
|
|
%ret = call i64 (ptr, ...) @callee_variadic(ptr %p, i64 7)
|
|
ret i64 %ret
|
|
}
|
|
|
|
define i64 @test_param_count_mismatch_variadic(ptr %p) {
|
|
; CHECK-LABEL: test_param_count_mismatch_variadic(
|
|
; CHECK: {
|
|
; CHECK-NEXT: .local .align 8 .b8 __local_depot5[8];
|
|
; CHECK-NEXT: .reg .b64 %SP;
|
|
; CHECK-NEXT: .reg .b64 %SPL;
|
|
; CHECK-NEXT: .reg .b64 %rd<4>;
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: // %bb.0:
|
|
; CHECK-NEXT: mov.b64 %SPL, __local_depot5;
|
|
; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
|
|
; CHECK-NEXT: ld.param.b64 %rd1, [test_param_count_mismatch_variadic_param_0];
|
|
; CHECK-NEXT: st.b64 [%SP], 7;
|
|
; CHECK-NEXT: { // callseq 5, 0
|
|
; CHECK-NEXT: .param .b64 param0;
|
|
; CHECK-NEXT: .param .b64 param1;
|
|
; CHECK-NEXT: .param .b64 retval0;
|
|
; CHECK-NEXT: add.u64 %rd2, %SP, 0;
|
|
; CHECK-NEXT: st.param.b64 [param1], %rd2;
|
|
; CHECK-NEXT: st.param.b64 [param0], %rd1;
|
|
; CHECK-NEXT: call.uni (retval0), callee_variadic, (param0, param1);
|
|
; CHECK-NEXT: ld.param.b64 %rd3, [retval0];
|
|
; CHECK-NEXT: } // callseq 5
|
|
; CHECK-NEXT: st.param.b64 [func_retval0], %rd3;
|
|
; CHECK-NEXT: ret;
|
|
%ret = call i64 (ptr, ...) @callee_variadic(ptr %p, i64 7)
|
|
ret i64 %ret
|
|
}
|