blob: c440caaf98abac1e918533b3b245a98b11b65516 [file] [log] [blame] [edit]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_90 -mattr=+ptx80| FileCheck --check-prefixes=CHECK-PTX64 %s
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_90 -mattr=+ptx80 --nvptx-short-ptr| FileCheck --check-prefixes=CHECK-PTX-SHARED32 %s
; RUN: %if ptxas-sm_90 && ptxas-isa-8.0 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_90 -mattr=+ptx80| %ptxas-verify -arch=sm_90 %}
; RUN: %if ptxas-sm_90 && ptxas-isa-8.0 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_90 -mattr=+ptx80 --nvptx-short-ptr| %ptxas-verify -arch=sm_90 %}
; === space_cta (addrspace 3) ===
define void @test_mbarrier_arrive_scope_cta_space_cta(ptr addrspace(3) %mbar, i32 %tx) {
; CHECK-PTX64-LABEL: test_mbarrier_arrive_scope_cta_space_cta(
; CHECK-PTX64: {
; CHECK-PTX64-NEXT: .reg .b32 %r<2>;
; CHECK-PTX64-NEXT: .reg .b64 %rd<6>;
; CHECK-PTX64-EMPTY:
; CHECK-PTX64-NEXT: // %bb.0:
; CHECK-PTX64-NEXT: ld.param.b64 %rd1, [test_mbarrier_arrive_scope_cta_space_cta_param_0];
; CHECK-PTX64-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cta_space_cta_param_1];
; CHECK-PTX64-NEXT: mbarrier.arrive.shared.b64 %rd2, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive.expect_tx.shared.b64 %rd3, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.shared.b64 %rd4, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.expect_tx.shared.b64 %rd5, [%rd1], %r1;
; CHECK-PTX64-NEXT: ret;
;
; CHECK-PTX-SHARED32-LABEL: test_mbarrier_arrive_scope_cta_space_cta(
; CHECK-PTX-SHARED32: {
; CHECK-PTX-SHARED32-NEXT: .reg .b32 %r<3>;
; CHECK-PTX-SHARED32-NEXT: .reg .b64 %rd<5>;
; CHECK-PTX-SHARED32-EMPTY:
; CHECK-PTX-SHARED32-NEXT: // %bb.0:
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cta_space_cta_param_0];
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r2, [test_mbarrier_arrive_scope_cta_space_cta_param_1];
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.shared.b64 %rd1, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.expect_tx.shared.b64 %rd2, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.shared.b64 %rd3, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.expect_tx.shared.b64 %rd4, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: ret;
%r1 = call i64 @llvm.nvvm.mbarrier.arrive.scope.cta.space.cta(ptr addrspace(3) %mbar, i32 %tx)
%r2 = call i64 @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cta.space.cta(ptr addrspace(3) %mbar, i32 %tx)
%r3 = call i64 @llvm.nvvm.mbarrier.arrive.drop.scope.cta.space.cta(ptr addrspace(3) %mbar, i32 %tx)
%r4 = call i64 @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cta.space.cta(ptr addrspace(3) %mbar, i32 %tx)
ret void
}
define void @test_mbarrier_arrive_scope_cluster_space_cta(ptr addrspace(3) %mbar, i32 %tx) {
; CHECK-PTX64-LABEL: test_mbarrier_arrive_scope_cluster_space_cta(
; CHECK-PTX64: {
; CHECK-PTX64-NEXT: .reg .b32 %r<2>;
; CHECK-PTX64-NEXT: .reg .b64 %rd<6>;
; CHECK-PTX64-EMPTY:
; CHECK-PTX64-NEXT: // %bb.0:
; CHECK-PTX64-NEXT: ld.param.b64 %rd1, [test_mbarrier_arrive_scope_cluster_space_cta_param_0];
; CHECK-PTX64-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cluster_space_cta_param_1];
; CHECK-PTX64-NEXT: mbarrier.arrive.release.cluster.shared.b64 %rd2, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive.expect_tx.release.cluster.shared.b64 %rd3, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.release.cluster.shared.b64 %rd4, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.expect_tx.release.cluster.shared.b64 %rd5, [%rd1], %r1;
; CHECK-PTX64-NEXT: ret;
;
; CHECK-PTX-SHARED32-LABEL: test_mbarrier_arrive_scope_cluster_space_cta(
; CHECK-PTX-SHARED32: {
; CHECK-PTX-SHARED32-NEXT: .reg .b32 %r<3>;
; CHECK-PTX-SHARED32-NEXT: .reg .b64 %rd<5>;
; CHECK-PTX-SHARED32-EMPTY:
; CHECK-PTX-SHARED32-NEXT: // %bb.0:
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cluster_space_cta_param_0];
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r2, [test_mbarrier_arrive_scope_cluster_space_cta_param_1];
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.release.cluster.shared.b64 %rd1, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.expect_tx.release.cluster.shared.b64 %rd2, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.release.cluster.shared.b64 %rd3, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.expect_tx.release.cluster.shared.b64 %rd4, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: ret;
%r1 = call i64 @llvm.nvvm.mbarrier.arrive.scope.cluster.space.cta(ptr addrspace(3) %mbar, i32 %tx)
%r2 = call i64 @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cluster.space.cta(ptr addrspace(3) %mbar, i32 %tx)
%r3 = call i64 @llvm.nvvm.mbarrier.arrive.drop.scope.cluster.space.cta(ptr addrspace(3) %mbar, i32 %tx)
%r4 = call i64 @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cluster.space.cta(ptr addrspace(3) %mbar, i32 %tx)
ret void
}
; === space_cluster (addrspace 7) ===
define void @test_mbarrier_arrive_scope_cta_space_cluster(ptr addrspace(7) %mbar, i32 %tx) {
; CHECK-PTX64-LABEL: test_mbarrier_arrive_scope_cta_space_cluster(
; CHECK-PTX64: {
; CHECK-PTX64-NEXT: .reg .b32 %r<2>;
; CHECK-PTX64-NEXT: .reg .b64 %rd<2>;
; CHECK-PTX64-EMPTY:
; CHECK-PTX64-NEXT: // %bb.0:
; CHECK-PTX64-NEXT: ld.param.b64 %rd1, [test_mbarrier_arrive_scope_cta_space_cluster_param_0];
; CHECK-PTX64-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cta_space_cluster_param_1];
; CHECK-PTX64-NEXT: mbarrier.arrive.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive.expect_tx.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.expect_tx.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: ret;
;
; CHECK-PTX-SHARED32-LABEL: test_mbarrier_arrive_scope_cta_space_cluster(
; CHECK-PTX-SHARED32: {
; CHECK-PTX-SHARED32-NEXT: .reg .b32 %r<3>;
; CHECK-PTX-SHARED32-EMPTY:
; CHECK-PTX-SHARED32-NEXT: // %bb.0:
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cta_space_cluster_param_0];
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r2, [test_mbarrier_arrive_scope_cta_space_cluster_param_1];
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.expect_tx.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.expect_tx.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: ret;
call void @llvm.nvvm.mbarrier.arrive.scope.cta.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
call void @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cta.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
call void @llvm.nvvm.mbarrier.arrive.drop.scope.cta.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
call void @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cta.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
ret void
}
define void @test_mbarrier_arrive_scope_cluster_space_cluster(ptr addrspace(7) %mbar, i32 %tx) {
; CHECK-PTX64-LABEL: test_mbarrier_arrive_scope_cluster_space_cluster(
; CHECK-PTX64: {
; CHECK-PTX64-NEXT: .reg .b32 %r<2>;
; CHECK-PTX64-NEXT: .reg .b64 %rd<2>;
; CHECK-PTX64-EMPTY:
; CHECK-PTX64-NEXT: // %bb.0:
; CHECK-PTX64-NEXT: ld.param.b64 %rd1, [test_mbarrier_arrive_scope_cluster_space_cluster_param_0];
; CHECK-PTX64-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cluster_space_cluster_param_1];
; CHECK-PTX64-NEXT: mbarrier.arrive.release.cluster.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive.expect_tx.release.cluster.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.release.cluster.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: mbarrier.arrive_drop.expect_tx.release.cluster.shared::cluster.b64 _, [%rd1], %r1;
; CHECK-PTX64-NEXT: ret;
;
; CHECK-PTX-SHARED32-LABEL: test_mbarrier_arrive_scope_cluster_space_cluster(
; CHECK-PTX-SHARED32: {
; CHECK-PTX-SHARED32-NEXT: .reg .b32 %r<3>;
; CHECK-PTX-SHARED32-EMPTY:
; CHECK-PTX-SHARED32-NEXT: // %bb.0:
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r1, [test_mbarrier_arrive_scope_cluster_space_cluster_param_0];
; CHECK-PTX-SHARED32-NEXT: ld.param.b32 %r2, [test_mbarrier_arrive_scope_cluster_space_cluster_param_1];
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.release.cluster.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive.expect_tx.release.cluster.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.release.cluster.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: mbarrier.arrive_drop.expect_tx.release.cluster.shared::cluster.b64 _, [%r1], %r2;
; CHECK-PTX-SHARED32-NEXT: ret;
call void @llvm.nvvm.mbarrier.arrive.scope.cluster.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
call void @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cluster.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
call void @llvm.nvvm.mbarrier.arrive.drop.scope.cluster.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
call void @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cluster.space.cluster(ptr addrspace(7) %mbar, i32 %tx)
ret void
}
declare i64 @llvm.nvvm.mbarrier.arrive.scope.cta.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cta.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.drop.scope.cta.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cta.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.scope.cluster.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cluster.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.drop.scope.cluster.space.cta(ptr addrspace(3), i32)
declare i64 @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cluster.space.cta(ptr addrspace(3), i32)
declare void @llvm.nvvm.mbarrier.arrive.scope.cta.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cta.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.drop.scope.cta.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cta.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.scope.cluster.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.expect.tx.scope.cluster.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.drop.scope.cluster.space.cluster(ptr addrspace(7), i32)
declare void @llvm.nvvm.mbarrier.arrive.drop.expect.tx.scope.cluster.space.cluster(ptr addrspace(7), i32)