[llvm-branch-commits] [llvm] [openmp] [OpenMPOpt] Ask the runtime how many of a block's threads can be workers (PR #221450)
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Sat Sep 5 07:15:34 PDT 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: Larry Meadows (lfmeadow)
<details>
<summary>Changes</summary>
The custom state machine gates a thread on `InitCB < BlockHwSize - WarpSize`,
reconstructing the worker count from the block size on the assumption that the
main thread occupies a whole warp above them. That arithmetic has to be kept in
step with whatever launch geometry the toolchain uses, and when it is not, the
workers in between are left in neither group, waiting for no parallel region
while the runtime hands them iterations.
Export the number the device runtime already computes in
`mapping::getMaxTeamThreads()`, which its own generic state machine gates on, as
`__kmpc_get_max_team_threads()`, and gate on that instead. The new entry point is
added to the virtual-use callback that keeps the state machine's callees alive,
and the existing `runtimeFnsAvailable()` check skips the rewrite against an older
device runtime.
NFC upstream, where `getMaxTeamThreads()` is `BlockSize - WarpSize`. It matters
for a toolchain whose block is not laid out that way, and it is what lets the
arithmetic live in one place.
Depends on #<!-- -->221449; the first commit is from that PR.
Tested on gfx90a: `llvm/test/Transforms/{OpenMP,Attributor}` 260 passed, 1
unsupported. No kernel in the AOMP smoke and smoke-fort suites changes execution
mode.
---
Patch is 53.25 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/221450.diff
10 Files Affected:
- (modified) llvm/include/llvm/Frontend/OpenMP/OMPKinds.def (+2)
- (modified) llvm/lib/Transforms/IPO/OpenMPOpt.cpp (+22-30)
- (modified) llvm/test/Transforms/OpenMP/callback_parallel_regions.ll (+4-8)
- (modified) llvm/test/Transforms/OpenMP/custom_state_machines.ll (+24-48)
- (modified) llvm/test/Transforms/OpenMP/spmdization.ll (+32-64)
- (modified) llvm/test/Transforms/OpenMP/spmdization_guarding.ll (+2-4)
- (modified) llvm/test/Transforms/OpenMP/spmdization_indirect.ll (+8-16)
- (modified) llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll (+6-12)
- (modified) openmp/device/include/Interface.h (+7)
- (modified) openmp/device/src/Mapping.cpp (+6-1)
``````````diff
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
index 8d94d952fb39f..d04ba1d19e87f 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
@@ -228,6 +228,7 @@ __OMP_RTL(__kmpc_omp_reg_task_with_affinity, false, Int32, IdentPtr, Int32,
__OMP_RTL(__kmpc_get_hardware_num_blocks, false, Int32, )
__OMP_RTL(__kmpc_get_hardware_num_threads_in_block, false, Int32, )
__OMP_RTL(__kmpc_get_warp_size, false, Int32, )
+__OMP_RTL(__kmpc_get_max_team_threads, false, Int32, )
__OMP_RTL(omp_get_thread_num, false, Int32, )
__OMP_RTL(omp_get_num_threads, false, Int32, )
@@ -721,6 +722,7 @@ __OMP_RTL_ATTRS(__kmpc_omp_reg_task_with_affinity, DefaultAttrs, SExt,
__OMP_RTL_ATTRS(__kmpc_get_hardware_num_blocks, GetterAttrs, ZExt, ParamAttrs())
__OMP_RTL_ATTRS(__kmpc_get_hardware_num_threads_in_block, GetterAttrs, ZExt, ParamAttrs())
__OMP_RTL_ATTRS(__kmpc_get_warp_size, GetterAttrs, ZExt, ParamAttrs())
+__OMP_RTL_ATTRS(__kmpc_get_max_team_threads, GetterAttrs, ZExt, ParamAttrs())
__OMP_RTL_ATTRS(omp_get_thread_num, GetterAttrs, SExt, ParamAttrs())
__OMP_RTL_ATTRS(omp_get_num_threads, GetterAttrs, SExt, ParamAttrs())
diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
index 5da55b76c7559..8639ae3a7d27c 100644
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -3960,8 +3960,7 @@ struct AAKernelInfoFunction : AAKernelInfo {
Attributor::VirtualUseCallbackTy CustomStateMachineUseCB =
[&](Attributor &A, const AbstractAttribute *QueryingAA) {
// Whenever we create a custom state machine we will insert calls to
- // __kmpc_get_hardware_num_threads_in_block,
- // __kmpc_get_warp_size,
+ // __kmpc_get_max_team_threads,
// __kmpc_barrier_simple_generic,
// __kmpc_kernel_parallel, and
// __kmpc_kernel_end_parallel.
@@ -3976,9 +3975,8 @@ struct AAKernelInfoFunction : AAKernelInfo {
// Not needed if we are pre-runtime merge.
if (!KernelInitCB->getCalledFunction()->isDeclaration()) {
- RegisterVirtualUse(OMPRTL___kmpc_get_hardware_num_threads_in_block,
+ RegisterVirtualUse(OMPRTL___kmpc_get_max_team_threads,
CustomStateMachineUseCB);
- RegisterVirtualUse(OMPRTL___kmpc_get_warp_size, CustomStateMachineUseCB);
RegisterVirtualUse(OMPRTL___kmpc_barrier_simple_generic,
CustomStateMachineUseCB);
RegisterVirtualUse(OMPRTL___kmpc_kernel_parallel,
@@ -4438,10 +4436,10 @@ struct AAKernelInfoFunction : AAKernelInfo {
return false;
auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
- if (!OMPInfoCache.runtimeFnsAvailable(
- {OMPRTL___kmpc_get_hardware_num_threads_in_block,
- OMPRTL___kmpc_get_warp_size, OMPRTL___kmpc_barrier_simple_generic,
- OMPRTL___kmpc_kernel_parallel, OMPRTL___kmpc_kernel_end_parallel}))
+ if (!OMPInfoCache.runtimeFnsAvailable({OMPRTL___kmpc_get_max_team_threads,
+ OMPRTL___kmpc_barrier_simple_generic,
+ OMPRTL___kmpc_kernel_parallel,
+ OMPRTL___kmpc_kernel_end_parallel}))
return false;
ConstantStruct *ExistingKernelEnvC =
@@ -4520,13 +4518,10 @@ struct AAKernelInfoFunction : AAKernelInfo {
// Create all the blocks:
//
// InitCB = __kmpc_target_init(...)
- // BlockHwSize =
- // __kmpc_get_hardware_num_threads_in_block();
- // WarpSize = __kmpc_get_warp_size();
- // BlockSize = BlockHwSize - WarpSize;
+ // MaxTeamThreads = __kmpc_get_max_team_threads();
// IsWorkerCheckBB: bool IsWorker = InitCB != -1;
// if (IsWorker) {
- // if (InitCB >= BlockSize) return;
+ // if (InitCB >= MaxTeamThreads) return;
// SMBeginBB: __kmpc_barrier_simple_generic(...);
// void *WorkFn;
// bool Active = __kmpc_kernel_parallel(&WorkFn);
@@ -4591,26 +4586,23 @@ struct AAKernelInfoFunction : AAKernelInfo {
IsWorker->setDebugLoc(DLoc);
CondBrInst::Create(IsWorker, IsWorkerCheckBB, UserCodeEntryBB, InitBB);
+ // How many of the block's threads can be worker threads is a property of
+ // the launch geometry, which the runtime knows and the block size alone
+ // does not determine: a target may host the main thread in a whole warp
+ // above the workers, or in a single thread above them. Ask the runtime
+ // rather than subtracting a warp here, or the workers in between are left
+ // in neither group, waiting for no parallel region while it is handed
+ // iterations.
Module &M = *Kernel->getParent();
- FunctionCallee BlockHwSizeFn =
+ FunctionCallee MaxTeamThreadsFn =
OMPInfoCache.OMPBuilder.getOrCreateRuntimeFunction(
- M, OMPRTL___kmpc_get_hardware_num_threads_in_block);
- FunctionCallee WarpSizeFn =
- OMPInfoCache.OMPBuilder.getOrCreateRuntimeFunction(
- M, OMPRTL___kmpc_get_warp_size);
- CallInst *BlockHwSize =
- CallInst::Create(BlockHwSizeFn, "block.hw_size", IsWorkerCheckBB);
- OMPInfoCache.setCallingConvention(BlockHwSizeFn, BlockHwSize);
- BlockHwSize->setDebugLoc(DLoc);
- CallInst *WarpSize =
- CallInst::Create(WarpSizeFn, "warp.size", IsWorkerCheckBB);
- OMPInfoCache.setCallingConvention(WarpSizeFn, WarpSize);
- WarpSize->setDebugLoc(DLoc);
- Instruction *BlockSize = BinaryOperator::CreateSub(
- BlockHwSize, WarpSize, "block.size", IsWorkerCheckBB);
- BlockSize->setDebugLoc(DLoc);
+ M, OMPRTL___kmpc_get_max_team_threads);
+ CallInst *MaxTeamThreads =
+ CallInst::Create(MaxTeamThreadsFn, "max_team_threads", IsWorkerCheckBB);
+ OMPInfoCache.setCallingConvention(MaxTeamThreadsFn, MaxTeamThreads);
+ MaxTeamThreads->setDebugLoc(DLoc);
Instruction *IsMainOrWorker = ICmpInst::Create(
- ICmpInst::ICmp, llvm::CmpInst::ICMP_SLT, KernelInitCB, BlockSize,
+ ICmpInst::ICmp, llvm::CmpInst::ICMP_SLT, KernelInitCB, MaxTeamThreads,
"thread.is_main_or_worker", IsWorkerCheckBB);
IsMainOrWorker->setDebugLoc(DLoc);
CondBrInst::Create(IsMainOrWorker, StateMachineBeginBB,
diff --git a/llvm/test/Transforms/OpenMP/callback_parallel_regions.ll b/llvm/test/Transforms/OpenMP/callback_parallel_regions.ll
index 9031c05350b5e..06dea403c30d3 100644
--- a/llvm/test/Transforms/OpenMP/callback_parallel_regions.ll
+++ b/llvm/test/Transforms/OpenMP/callback_parallel_regions.ll
@@ -39,10 +39,8 @@ define weak amdgpu_kernel void @visible_body(ptr %dyn, ptr %payload) #0 {
; CHECK-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; CHECK-NEXT: br i1 [[THREAD_IS_WORKER]], label %[[IS_WORKER_CHECK:.*]], label %[[THREAD_USER_CODE_CHECK:.*]]
; CHECK: [[IS_WORKER_CHECK]]:
-; CHECK-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; CHECK-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; CHECK-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; CHECK-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; CHECK-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; CHECK-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; CHECK-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label %[[WORKER_STATE_MACHINE_BEGIN:.*]], label %[[WORKER_STATE_MACHINE_FINISHED:.*]]
; CHECK: [[WORKER_STATE_MACHINE_BEGIN]]:
; CHECK-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -101,10 +99,8 @@ define weak amdgpu_kernel void @opaque_body(ptr %dyn, ptr %payload) #0 {
; CHECK-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; CHECK-NEXT: br i1 [[THREAD_IS_WORKER]], label %[[IS_WORKER_CHECK:.*]], label %[[THREAD_USER_CODE_CHECK:.*]]
; CHECK: [[IS_WORKER_CHECK]]:
-; CHECK-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; CHECK-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; CHECK-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; CHECK-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; CHECK-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; CHECK-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; CHECK-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label %[[WORKER_STATE_MACHINE_BEGIN:.*]], label %[[WORKER_STATE_MACHINE_FINISHED:.*]]
; CHECK: [[WORKER_STATE_MACHINE_BEGIN]]:
; CHECK-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
diff --git a/llvm/test/Transforms/OpenMP/custom_state_machines.ll b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
index a03a4b7f3fa67..ac35d7a4a9765 100644
--- a/llvm/test/Transforms/OpenMP/custom_state_machines.ll
+++ b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
@@ -1152,10 +1152,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; AMDGPU-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; AMDGPU-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; AMDGPU: is_worker_check:
-; AMDGPU-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; AMDGPU-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; AMDGPU-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; AMDGPU-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; AMDGPU-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; AMDGPU: worker_state_machine.begin:
; AMDGPU-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -1273,10 +1271,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; AMDGPU-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; AMDGPU-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; AMDGPU: is_worker_check:
-; AMDGPU-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; AMDGPU-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; AMDGPU-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; AMDGPU-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; AMDGPU-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; AMDGPU: worker_state_machine.begin:
; AMDGPU-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -1420,10 +1416,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; AMDGPU-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; AMDGPU-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; AMDGPU: is_worker_check:
-; AMDGPU-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; AMDGPU-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; AMDGPU-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; AMDGPU-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; AMDGPU-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; AMDGPU: worker_state_machine.begin:
; AMDGPU-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -1542,10 +1536,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; AMDGPU-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; AMDGPU-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; AMDGPU: is_worker_check:
-; AMDGPU-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; AMDGPU-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; AMDGPU-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; AMDGPU-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; AMDGPU-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; AMDGPU: worker_state_machine.begin:
; AMDGPU-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -1662,10 +1654,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; AMDGPU-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; AMDGPU-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; AMDGPU: is_worker_check:
-; AMDGPU-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; AMDGPU-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; AMDGPU-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; AMDGPU-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; AMDGPU-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; AMDGPU: worker_state_machine.begin:
; AMDGPU-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -1854,10 +1844,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; AMDGPU-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; AMDGPU-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; AMDGPU: is_worker_check:
-; AMDGPU-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; AMDGPU-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; AMDGPU-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; AMDGPU-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; AMDGPU-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; AMDGPU-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; AMDGPU: worker_state_machine.begin:
; AMDGPU-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -2078,10 +2066,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; NVPTX-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; NVPTX-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; NVPTX: is_worker_check:
-; NVPTX-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; NVPTX-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; NVPTX-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; NVPTX-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; NVPTX-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; NVPTX: worker_state_machine.begin:
; NVPTX-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -2198,10 +2184,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; NVPTX-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; NVPTX-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; NVPTX: is_worker_check:
-; NVPTX-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; NVPTX-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; NVPTX-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; NVPTX-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; NVPTX-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; NVPTX: worker_state_machine.begin:
; NVPTX-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -2344,10 +2328,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; NVPTX-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; NVPTX-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; NVPTX: is_worker_check:
-; NVPTX-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; NVPTX-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; NVPTX-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; NVPTX-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; NVPTX-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; NVPTX: worker_state_machine.begin:
; NVPTX-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -2465,10 +2447,8 @@ attributes #9 = { convergent nounwind willreturn memory(read) }
; NVPTX-NEXT: [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
; NVPTX-NEXT: br i1 [[THREAD_IS_WORKER]], label [[IS_WORKER_CHECK:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
; NVPTX: is_worker_check:
-; NVPTX-NEXT: [[BLOCK_HW_SIZE:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
-; NVPTX-NEXT: [[WARP_SIZE:%.*]] = call i32 @__kmpc_get_warp_size()
-; NVPTX-NEXT: [[BLOCK_SIZE:%.*]] = sub i32 [[BLOCK_HW_SIZE]], [[WARP_SIZE]]
-; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[BLOCK_SIZE]]
+; NVPTX-NEXT: [[MAX_TEAM_THREADS:%.*]] = call i32 @__kmpc_get_max_team_threads()
+; NVPTX-NEXT: [[THREAD_IS_MAIN_OR_WORKER:%.*]] = icmp slt i32 [[TMP0]], [[MAX_TEAM_THREADS]]
; NVPTX-NEXT: br i1 [[THREAD_IS_MAIN_OR_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]]
; NVPTX: worker_state_machine.begin:
; NVPTX-NEXT: call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
@@ -2584...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/221450
More information about the llvm-branch-commits
mailing list