; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_30 -mattr=+ptx60 | FileCheck %s ; RUN: %if ptxas-isa-6.0 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_30 -mattr=+ptx60 | %ptxas-verify %} declare void @llvm.nvvm.bar.warp.sync(i32) declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) declare void @llvm.nvvm.barrier.cta.sync.aligned.count(i32, i32) declare void @llvm.nvvm.barrier.cta.sync.all(i32) declare void @llvm.nvvm.barrier.cta.sync.count(i32, i32) declare void @llvm.nvvm.barrier.cta.arrive.count(i32, i32) declare void @llvm.nvvm.barrier.cta.arrive.aligned.count(i32, i32) define void @barrier_warp_sync(i32 %id) { ; CHECK-LABEL: barrier_warp_sync( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<2>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_warp_sync_param_0]; ; CHECK-NEXT: bar.warp.sync %r1; ; CHECK-NEXT: bar.warp.sync 6; ; CHECK-NEXT: ret; call void @llvm.nvvm.bar.warp.sync(i32 %id) call void @llvm.nvvm.bar.warp.sync(i32 6) ret void } define void @barrier_cta_sync_aligned_all(i32 %id) { ; CHECK-LABEL: barrier_cta_sync_aligned_all( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<2>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_sync_aligned_all_param_0]; ; CHECK-NEXT: bar.sync %r1; ; CHECK-NEXT: bar.sync 3; ; CHECK-NEXT: ret; call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 %id) call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 3) ret void } define void @barrier_cta_sync_aligned(i32 %id, i32 %cnt) { ; CHECK-LABEL: barrier_cta_sync_aligned( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<3>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_sync_aligned_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_sync_aligned_param_1]; ; CHECK-NEXT: bar.sync %r1, %r2; ; CHECK-NEXT: bar.sync 3, %r2; ; CHECK-NEXT: bar.sync %r1, 64; ; CHECK-NEXT: bar.sync 4, 64; ; CHECK-NEXT: ret; call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 %id, i32 %cnt) call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 3, i32 %cnt) call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 %id, i32 64) call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 4, i32 64) ret void } define void @barrier_cta_arrive_aligned(i32 %id, i32 %cnt) { ; CHECK-LABEL: barrier_cta_arrive_aligned( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<3>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_arrive_aligned_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_arrive_aligned_param_1]; ; CHECK-NEXT: bar.arrive %r1, %r2; ; CHECK-NEXT: bar.arrive 3, %r2; ; CHECK-NEXT: bar.arrive %r1, 64; ; CHECK-NEXT: bar.arrive 4, 64; ; CHECK-NEXT: ret; call void @llvm.nvvm.barrier.cta.arrive.aligned.count(i32 %id, i32 %cnt) call void @llvm.nvvm.barrier.cta.arrive.aligned.count(i32 3, i32 %cnt) call void @llvm.nvvm.barrier.cta.arrive.aligned.count(i32 %id, i32 64) call void @llvm.nvvm.barrier.cta.arrive.aligned.count(i32 4, i32 64) ret void } define void @barrier_cta_sync_all(i32 %id) { ; CHECK-LABEL: barrier_cta_sync_all( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<2>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_sync_all_param_0]; ; CHECK-NEXT: barrier.sync %r1; ; CHECK-NEXT: barrier.sync 3; ; CHECK-NEXT: ret; call void @llvm.nvvm.barrier.cta.sync.all(i32 %id) call void @llvm.nvvm.barrier.cta.sync.all(i32 3) ret void } define void @barrier_cta_sync(i32 %id, i32 %cnt) { ; CHECK-LABEL: barrier_cta_sync( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<3>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_sync_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_sync_param_1]; ; CHECK-NEXT: barrier.sync %r1, %r2; ; CHECK-NEXT: barrier.sync 3, %r2; ; CHECK-NEXT: barrier.sync %r1, 64; ; CHECK-NEXT: barrier.sync 4, 64; ; CHECK-NEXT: ret; call void @llvm.nvvm.barrier.cta.sync.count(i32 %id, i32 %cnt) call void @llvm.nvvm.barrier.cta.sync.count(i32 3, i32 %cnt) call void @llvm.nvvm.barrier.cta.sync.count(i32 %id, i32 64) call void @llvm.nvvm.barrier.cta.sync.count(i32 4, i32 64) ret void } define void @barrier_cta_arrive(i32 %id, i32 %cnt) { ; CHECK-LABEL: barrier_cta_arrive( ; CHECK: { ; CHECK-NEXT: .reg .b32 %r<3>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_arrive_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_arrive_param_1]; ; CHECK-NEXT: barrier.arrive %r1, %r2; ; CHECK-NEXT: barrier.arrive 3, %r2; ; CHECK-NEXT: barrier.arrive %r1, 64; ; CHECK-NEXT: barrier.arrive 4, 64; ; CHECK-NEXT: ret; call void @llvm.nvvm.barrier.cta.arrive.count(i32 %id, i32 %cnt) call void @llvm.nvvm.barrier.cta.arrive.count(i32 3, i32 %cnt) call void @llvm.nvvm.barrier.cta.arrive.count(i32 %id, i32 64) call void @llvm.nvvm.barrier.cta.arrive.count(i32 4, i32 64) ret void } define void @barrier_cta_red_popc_all(i32 %id, i1 %pred) { ; CHECK-LABEL: barrier_cta_red_popc_all( ; CHECK: { ; CHECK-NEXT: .reg .pred %p<2>; ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<6>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b8 %rs1, [barrier_cta_red_popc_all_param_1]; ; CHECK-NEXT: and.b16 %rs2, %rs1, 1; ; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0; ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_popc_all_param_0]; ; CHECK-NEXT: bar.red.popc.u32 %r2, %r1, %p1; ; CHECK-NEXT: bar.red.popc.u32 %r3, 3, %p1; ; CHECK-NEXT: barrier.red.popc.u32 %r4, %r1, %p1; ; CHECK-NEXT: barrier.red.popc.u32 %r5, 3, %p1; ; CHECK-NEXT: ret; %v1 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.all(i32 %id, i1 %pred) %v2 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.all(i32 3, i1 %pred) %v3 = call i32 @llvm.nvvm.barrier.cta.red.popc.all(i32 %id, i1 %pred) %v4 = call i32 @llvm.nvvm.barrier.cta.red.popc.all(i32 3, i1 %pred) ret void } define void @barrier_cta_red_popc_count(i32 %id, i32 %cnt, i1 %pred) { ; CHECK-LABEL: barrier_cta_red_popc_count( ; CHECK: { ; CHECK-NEXT: .reg .pred %p<2>; ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<11>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b8 %rs1, [barrier_cta_red_popc_count_param_2]; ; CHECK-NEXT: and.b16 %rs2, %rs1, 1; ; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0; ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_popc_count_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_red_popc_count_param_1]; ; CHECK-NEXT: bar.red.popc.u32 %r3, %r1, %r2, %p1; ; CHECK-NEXT: bar.red.popc.u32 %r4, 3, %r2, %p1; ; CHECK-NEXT: barrier.red.popc.u32 %r5, %r1, %r2, %p1; ; CHECK-NEXT: barrier.red.popc.u32 %r6, 3, %r2, %p1; ; CHECK-NEXT: bar.red.popc.u32 %r7, %r1, 64, %p1; ; CHECK-NEXT: bar.red.popc.u32 %r8, 3, 64, %p1; ; CHECK-NEXT: barrier.red.popc.u32 %r9, %r1, 64, %p1; ; CHECK-NEXT: barrier.red.popc.u32 %r10, 3, 64, %p1; ; CHECK-NEXT: ret; %v1 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 %id, i32 %cnt, i1 %pred) %v2 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 3, i32 %cnt, i1 %pred) %v3 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 %id, i32 %cnt, i1 %pred) %v4 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 3, i32 %cnt, i1 %pred) %v5 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 %id, i32 64, i1 %pred) %v6 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 3, i32 64, i1 %pred) %v7 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 %id, i32 64, i1 %pred) %v8 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 3, i32 64, i1 %pred) ret void } define void @barrier_cta_red_and_all(i32 %id, i1 %pred) { ; CHECK-LABEL: barrier_cta_red_and_all( ; CHECK: { ; CHECK-NEXT: .reg .pred %p<6>; ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<2>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b8 %rs1, [barrier_cta_red_and_all_param_1]; ; CHECK-NEXT: and.b16 %rs2, %rs1, 1; ; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0; ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_and_all_param_0]; ; CHECK-NEXT: bar.red.and.pred %p2, %r1, %p1; ; CHECK-NEXT: bar.red.and.pred %p3, 3, %p1; ; CHECK-NEXT: barrier.red.and.pred %p4, %r1, %p1; ; CHECK-NEXT: barrier.red.and.pred %p5, 3, %p1; ; CHECK-NEXT: ret; %v1 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.all(i32 %id, i1 %pred) %v2 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.all(i32 3, i1 %pred) %v3 = call i1 @llvm.nvvm.barrier.cta.red.and.all(i32 %id, i1 %pred) %v4 = call i1 @llvm.nvvm.barrier.cta.red.and.all(i32 3, i1 %pred) ret void } define void @barrier_cta_red_and_count(i32 %id, i32 %cnt, i1 %pred) { ; CHECK-LABEL: barrier_cta_red_and_count( ; CHECK: { ; CHECK-NEXT: .reg .pred %p<10>; ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<3>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b8 %rs1, [barrier_cta_red_and_count_param_2]; ; CHECK-NEXT: and.b16 %rs2, %rs1, 1; ; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0; ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_and_count_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_red_and_count_param_1]; ; CHECK-NEXT: bar.red.and.pred %p2, %r1, %r2, %p1; ; CHECK-NEXT: bar.red.and.pred %p3, 3, %r2, %p1; ; CHECK-NEXT: barrier.red.and.pred %p4, %r1, %r2, %p1; ; CHECK-NEXT: barrier.red.and.pred %p5, 3, %r2, %p1; ; CHECK-NEXT: bar.red.and.pred %p6, %r1, 64, %p1; ; CHECK-NEXT: bar.red.and.pred %p7, 3, 32, %p1; ; CHECK-NEXT: barrier.red.and.pred %p8, %r1, 64, %p1; ; CHECK-NEXT: barrier.red.and.pred %p9, 3, 64, %p1; ; CHECK-NEXT: ret; %v1 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 %id, i32 %cnt, i1 %pred) %v2 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 3, i32 %cnt, i1 %pred) %v3 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 %id, i32 %cnt, i1 %pred) %v4 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 3, i32 %cnt, i1 %pred) %v5 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 %id, i32 64, i1 %pred) %v6 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 3, i32 32, i1 %pred) %v7 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 %id, i32 64, i1 %pred) %v8 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 3, i32 64, i1 %pred) ret void } define void @barrier_cta_red_or_all(i32 %id, i1 %pred) { ; CHECK-LABEL: barrier_cta_red_or_all( ; CHECK: { ; CHECK-NEXT: .reg .pred %p<6>; ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<2>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b8 %rs1, [barrier_cta_red_or_all_param_1]; ; CHECK-NEXT: and.b16 %rs2, %rs1, 1; ; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0; ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_or_all_param_0]; ; CHECK-NEXT: bar.red.or.pred %p2, %r1, %p1; ; CHECK-NEXT: bar.red.or.pred %p3, 3, %p1; ; CHECK-NEXT: barrier.red.or.pred %p4, %r1, %p1; ; CHECK-NEXT: barrier.red.or.pred %p5, 3, %p1; ; CHECK-NEXT: ret; %v1 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.all(i32 %id, i1 %pred) %v2 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.all(i32 3, i1 %pred) %v3 = call i1 @llvm.nvvm.barrier.cta.red.or.all(i32 %id, i1 %pred) %v4 = call i1 @llvm.nvvm.barrier.cta.red.or.all(i32 3, i1 %pred) ret void } define void @barrier_cta_red_or_count(i32 %id, i32 %cnt, i1 %pred) { ; CHECK-LABEL: barrier_cta_red_or_count( ; CHECK: { ; CHECK-NEXT: .reg .pred %p<10>; ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<3>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: ; CHECK-NEXT: ld.param.b8 %rs1, [barrier_cta_red_or_count_param_2]; ; CHECK-NEXT: and.b16 %rs2, %rs1, 1; ; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0; ; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_or_count_param_0]; ; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_red_or_count_param_1]; ; CHECK-NEXT: bar.red.or.pred %p2, %r1, %r2, %p1; ; CHECK-NEXT: bar.red.or.pred %p3, 3, %r2, %p1; ; CHECK-NEXT: barrier.red.or.pred %p4, %r1, %r2, %p1; ; CHECK-NEXT: barrier.red.or.pred %p5, 3, %r2, %p1; ; CHECK-NEXT: bar.red.or.pred %p6, %r1, 64, %p1; ; CHECK-NEXT: bar.red.or.pred %p7, 3, 32, %p1; ; CHECK-NEXT: barrier.red.or.pred %p8, %r1, 64, %p1; ; CHECK-NEXT: barrier.red.or.pred %p9, 3, 64, %p1; ; CHECK-NEXT: ret; %v1 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 %id, i32 %cnt, i1 %pred) %v2 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 3, i32 %cnt, i1 %pred) %v3 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 %id, i32 %cnt, i1 %pred) %v4 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 3, i32 %cnt, i1 %pred) %v5 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 %id, i32 64, i1 %pred) %v6 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 3, i32 32, i1 %pred) %v7 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 %id, i32 64, i1 %pred) %v8 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 3, i32 64, i1 %pred) ret void }