This change updates architecture support checks for tcgen05 intrinsics (except tcgen05.mma.*). The newer checks will support family-specific architecture variants as well. After this change, the arch checks will be accurate and match with PTX ISA. Intrinsics affected: - tcgen05.ld/st - tcgen05.alloc/dealloc/relinquish - tcgen05.cp - tcgen05.fence/wait - tcgen05.commit - tcgen05.shift
218 lines
11 KiB
LLVM
218 lines
11 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc < %s -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 | FileCheck --check-prefixes=CHECK_PTX64 %s
|
|
; RUN: llc < %s -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 --nvptx-short-ptr | FileCheck --check-prefixes=CHECK_PTX64_SHARED32 %s
|
|
; RUN: llc < %s -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 | FileCheck --check-prefixes=CHECK_PTX64 %s
|
|
; RUN: llc < %s -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 | FileCheck --check-prefixes=CHECK_PTX64 %s
|
|
; RUN: llc < %s -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 | FileCheck --check-prefixes=CHECK_PTX64 %s
|
|
; RUN: %if ptxas-sm_100a && ptxas-isa-8.6 %{ llc < %s -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 | %ptxas-verify -arch=sm_100a %}
|
|
; RUN: %if ptxas-sm_100a && ptxas-isa-8.6 %{ llc < %s -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 --nvptx-short-ptr | %ptxas-verify -arch=sm_100a %}
|
|
; RUN: %if ptxas-sm_103a && ptxas-isa-8.8 %{ llc < %s -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 | %ptxas-verify -arch=sm_103a %}
|
|
; RUN: %if ptxas-sm_100f && ptxas-isa-8.8 %{ llc < %s -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 | %ptxas-verify -arch=sm_100f %}
|
|
; RUN: %if ptxas-sm_110f && ptxas-isa-9.0 %{ llc < %s -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 | %ptxas-verify -arch=sm_110f %}
|
|
|
|
declare void @llvm.nvvm.tcgen05.commit.cg1(ptr %bar_addr)
|
|
declare void @llvm.nvvm.tcgen05.commit.cg2(ptr %bar_addr)
|
|
declare void @llvm.nvvm.tcgen05.commit.shared.cg1(ptr addrspace(3) %bar_addr)
|
|
declare void @llvm.nvvm.tcgen05.commit.shared.cg2(ptr addrspace(3) %bar_addr)
|
|
|
|
define void @test_tcgen05_commit_cg1(ptr %bar_addr) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_cg1(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_cg1_param_0];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%rd1];
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_cg1(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_cg1_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%rd1];
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.cg1(ptr %bar_addr)
|
|
|
|
ret void
|
|
}
|
|
|
|
define void @test_tcgen05_commit_cg2(ptr %bar_addr) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_cg2(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_cg2_param_0];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.b64 [%rd1];
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_cg2(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_cg2_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.b64 [%rd1];
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.cg2(ptr %bar_addr)
|
|
|
|
ret void
|
|
}
|
|
|
|
define void @test_tcgen05_commit_shared_cg1(ptr addrspace(3) %bar_addr) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_shared_cg1(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_shared_cg1_param_0];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%rd1];
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_shared_cg1(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b32 %r<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b32 %r1, [test_tcgen05_commit_shared_cg1_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r1];
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.shared.cg1(ptr addrspace(3) %bar_addr)
|
|
|
|
ret void
|
|
}
|
|
|
|
define void @test_tcgen05_commit_shared_cg2(ptr addrspace(3) %bar_addr) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_shared_cg2(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_shared_cg2_param_0];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.b64 [%rd1];
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_shared_cg2(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b32 %r<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b32 %r1, [test_tcgen05_commit_shared_cg2_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.b64 [%r1];
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.shared.cg2(ptr addrspace(3) %bar_addr)
|
|
|
|
ret void
|
|
}
|
|
|
|
declare void @llvm.nvvm.tcgen05.commit.mc.cg1(ptr %bar_addr, i16 %cta_mask)
|
|
declare void @llvm.nvvm.tcgen05.commit.mc.cg2(ptr %bar_addr, i16 %cta_mask)
|
|
declare void @llvm.nvvm.tcgen05.commit.mc.shared.cg1(ptr addrspace(3) %bar_addr, i16 %cta_mask)
|
|
declare void @llvm.nvvm.tcgen05.commit.mc.shared.cg2(ptr addrspace(3) %bar_addr, i16 %cta_mask)
|
|
|
|
define void @test_tcgen05_commit_mc_cg1(ptr %bar_addr, i16 %cta_mask) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_mc_cg1(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_mc_cg1_param_0];
|
|
; CHECK_PTX64-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_cg1_param_1];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%rd1], %rs1;
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_mc_cg1(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_mc_cg1_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_cg1_param_1];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%rd1], %rs1;
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.mc.cg1(ptr %bar_addr, i16 %cta_mask)
|
|
ret void
|
|
}
|
|
|
|
define void @test_tcgen05_commit_mc_cg2(ptr %bar_addr, i16 %cta_mask) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_mc_cg2(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_mc_cg2_param_0];
|
|
; CHECK_PTX64-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_cg2_param_1];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%rd1], %rs1;
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_mc_cg2(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_mc_cg2_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_cg2_param_1];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%rd1], %rs1;
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.mc.cg2(ptr %bar_addr, i16 %cta_mask)
|
|
ret void
|
|
}
|
|
|
|
define void @test_tcgen05_commit_mc_shared_cg1(ptr addrspace(3) %bar_addr, i16 %cta_mask) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_mc_shared_cg1(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_mc_shared_cg1_param_0];
|
|
; CHECK_PTX64-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_shared_cg1_param_1];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%rd1], %rs1;
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_mc_shared_cg1(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b32 %r<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b32 %r1, [test_tcgen05_commit_mc_shared_cg1_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_shared_cg1_param_1];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%r1], %rs1;
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.mc.shared.cg1(ptr addrspace(3) %bar_addr, i16 %cta_mask)
|
|
ret void
|
|
}
|
|
|
|
define void @test_tcgen05_commit_mc_shared_cg2(ptr addrspace(3) %bar_addr, i16 %cta_mask) {
|
|
; CHECK_PTX64-LABEL: test_tcgen05_commit_mc_shared_cg2(
|
|
; CHECK_PTX64: {
|
|
; CHECK_PTX64-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64-NEXT: .reg .b64 %rd<2>;
|
|
; CHECK_PTX64-EMPTY:
|
|
; CHECK_PTX64-NEXT: // %bb.0:
|
|
; CHECK_PTX64-NEXT: ld.param.b64 %rd1, [test_tcgen05_commit_mc_shared_cg2_param_0];
|
|
; CHECK_PTX64-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_shared_cg2_param_1];
|
|
; CHECK_PTX64-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%rd1], %rs1;
|
|
; CHECK_PTX64-NEXT: ret;
|
|
;
|
|
; CHECK_PTX64_SHARED32-LABEL: test_tcgen05_commit_mc_shared_cg2(
|
|
; CHECK_PTX64_SHARED32: {
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b16 %rs<2>;
|
|
; CHECK_PTX64_SHARED32-NEXT: .reg .b32 %r<2>;
|
|
; CHECK_PTX64_SHARED32-EMPTY:
|
|
; CHECK_PTX64_SHARED32-NEXT: // %bb.0:
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b32 %r1, [test_tcgen05_commit_mc_shared_cg2_param_0];
|
|
; CHECK_PTX64_SHARED32-NEXT: ld.param.b16 %rs1, [test_tcgen05_commit_mc_shared_cg2_param_1];
|
|
; CHECK_PTX64_SHARED32-NEXT: tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [%r1], %rs1;
|
|
; CHECK_PTX64_SHARED32-NEXT: ret;
|
|
call void @llvm.nvvm.tcgen05.commit.mc.shared.cg2(ptr addrspace(3) %bar_addr, i16 %cta_mask)
|
|
ret void
|
|
}
|