Skip to content

Commit 2641ed7

Browse files
jdoerfertjdoerfert
authored
[OpenMP][FIX] Check for requirements early (#104836)
If we can't transform the region to SPMD, we should not wait till the end to decide that. Other AAs might assume SPMD, and we did set the constant initializer to indicate SPMD, but we did not change the code properly.
1 parent 61f8ab3 commit 2641ed7

File tree

2 files changed

+14
-15
lines changed

2 files changed

+14
-15
lines changed

‎llvm/lib/Transforms/IPO/OpenMPOpt.cpp

Lines changed: 6 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -3760,6 +3760,11 @@ struct AAKernelInfoFunction : AAKernelInfo {
37603760
A.registerGlobalVariableSimplificationCallback(
37613761
*KernelEnvGV, KernelConfigurationSimplifyCB);
37623762

3763+
// We cannot change to SPMD mode if the runtime functions aren't availible.
3764+
bool CanChangeToSPMD = OMPInfoCache.runtimeFnsAvailable(
3765+
{OMPRTL___kmpc_get_hardware_thread_id_in_block,
3766+
OMPRTL___kmpc_barrier_simple_spmd});
3767+
37633768
// Check if we know we are in SPMD-mode already.
37643769
ConstantInt *ExecModeC =
37653770
KernelInfo::getExecModeFromKernelEnvironment(KernelEnvC);
@@ -3768,7 +3773,7 @@ struct AAKernelInfoFunction : AAKernelInfo {
37683773
ExecModeC->getSExtValue() | OMP_TGT_EXEC_MODE_GENERIC_SPMD);
37693774
if (ExecModeC->getSExtValue() & OMP_TGT_EXEC_MODE_SPMD)
37703775
SPMDCompatibilityTracker.indicateOptimisticFixpoint();
3771-
else if (DisableOpenMPOptSPMDization)
3776+
else if (DisableOpenMPOptSPMDization || !CanChangeToSPMD)
37723777
// This is a generic region but SPMDization is disabled so stop
37733778
// tracking.
37743779
SPMDCompatibilityTracker.indicatePessimisticFixpoint();
@@ -4215,12 +4220,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
42154220
bool changeToSPMDMode(Attributor &A, ChangeStatus &Changed) {
42164221
auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
42174222

4218-
// We cannot change to SPMD mode if the runtime functions aren't availible.
4219-
if (!OMPInfoCache.runtimeFnsAvailable(
4220-
{OMPRTL___kmpc_get_hardware_thread_id_in_block,
4221-
OMPRTL___kmpc_barrier_simple_spmd}))
4222-
return false;
4223-
42244223
if (!SPMDCompatibilityTracker.isAssumed()) {
42254224
for (Instruction *NonCompatibleI : SPMDCompatibilityTracker) {
42264225
if (!NonCompatibleI)

‎llvm/test/Transforms/OpenMP/spmdization.ll

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -180,10 +180,10 @@
180180
;.
181181
; AMDGPU-DISABLED2: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
182182
; AMDGPU-DISABLED2: @[[GLOB1:[0-9]+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, ptr @[[GLOB0]] }, align 8
183-
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_l5_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
184-
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_stack_var_l20_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
185-
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_l35_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
186-
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_guarded_l50_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
183+
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_l5_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
184+
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_stack_var_l20_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
185+
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_l35_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
186+
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_guarded_l50_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
187187
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_do_not_spmdize_target_l65_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 0, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
188188
; AMDGPU-DISABLED2: @__omp_offloading_fd02_2044372e_do_not_spmdize_task_l74_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
189189
; AMDGPU-DISABLED2: @x_shared = internal addrspace(3) global [4 x i8] poison, align 4
@@ -212,10 +212,10 @@
212212
;.
213213
; NVPTX-DISABLED2: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
214214
; NVPTX-DISABLED2: @[[GLOB1:[0-9]+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, ptr @[[GLOB0]] }, align 8
215-
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_l5_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
216-
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_stack_var_l20_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
217-
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_l35_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
218-
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_guarded_l50_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
215+
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_l5_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
216+
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_stack_var_l20_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
217+
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_l35_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
218+
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_sequential_loop_to_shared_var_guarded_l50_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
219219
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_do_not_spmdize_target_l65_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 0, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
220220
; NVPTX-DISABLED2: @__omp_offloading_fd02_2044372e_do_not_spmdize_task_l74_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 1, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null }
221221
; NVPTX-DISABLED2: @x_shared = internal addrspace(3) global [4 x i8] poison, align 4

0 commit comments

Comments
 (0)