Skip to content

Commit e28936f

Browse files
jdoerfertjdoerfert
committed
[OpenMP][Opt] Annotate known runtime functions and deduplicate more
This adds ~27 more runtime calls to the OpenMPKinds.def file, all with attributes. We deduplicate 16 of those automatically in function = thread scope. And we annotate all of them automatically during the OpenMPOpt discovery step. A test with all omp_XXXX runtime calls to track annotation coverage is included. Reviewed By: JonChesterfield Differential Revision: https://reviews.llvm.org/D69984
1 parent 8df173f commit e28936f

File tree

2 files changed

+104
-2
lines changed

2 files changed

+104
-2
lines changed

‎llvm/include/llvm/Frontend/OpenMP/OMPKinds.def

Lines changed: 73 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -176,6 +176,34 @@ __OMP_RTL(__kmpc_serialized_parallel, false, Void, IdentPtr, Int32)
176176
__OMP_RTL(__kmpc_end_serialized_parallel, false, Void, IdentPtr, Int32)
177177

178178
__OMP_RTL(omp_get_thread_num, false, Int32, )
179+
__OMP_RTL(omp_get_num_threads, false, Int32, )
180+
__OMP_RTL(omp_get_max_threads, false, Int32, )
181+
__OMP_RTL(omp_in_parallel, false, Int32, )
182+
__OMP_RTL(omp_get_dynamic, false, Int32, )
183+
__OMP_RTL(omp_get_cancellation, false, Int32, )
184+
__OMP_RTL(omp_get_nested, false, Int32, )
185+
__OMP_RTL(omp_get_schedule, false, Void, Int32Ptr, Int32Ptr)
186+
__OMP_RTL(omp_get_thread_limit, false, Int32, )
187+
__OMP_RTL(omp_get_supported_active_levels, false, Int32, )
188+
__OMP_RTL(omp_get_max_active_levels, false, Int32, )
189+
__OMP_RTL(omp_get_level, false, Int32, )
190+
__OMP_RTL(omp_get_ancestor_thread_num, false, Int32, )
191+
__OMP_RTL(omp_get_team_size, false, Int32, )
192+
__OMP_RTL(omp_get_active_level, false, Int32, )
193+
__OMP_RTL(omp_in_final, false, Int32, )
194+
__OMP_RTL(omp_get_proc_bind, false, Int32, )
195+
__OMP_RTL(omp_get_num_places, false, Int32, )
196+
__OMP_RTL(omp_get_num_procs, false, Int32, )
197+
__OMP_RTL(omp_get_place_proc_ids, false, Void, Int32, Int32Ptr)
198+
__OMP_RTL(omp_get_place_num, false, Int32, )
199+
__OMP_RTL(omp_get_partition_num_places, false, Int32, )
200+
__OMP_RTL(omp_get_partition_place_nums, false, Int32, )
201+
202+
__OMP_RTL(omp_set_num_threads, false, Void, Int32)
203+
__OMP_RTL(omp_set_dynamic, false, Void, Int32)
204+
__OMP_RTL(omp_set_nested, false, Void, Int32)
205+
__OMP_RTL(omp_set_schedule, false, Void, Int32, Int32)
206+
__OMP_RTL(omp_set_max_active_levels, false, Void, Int32)
179207

180208
__OMP_RTL(__last, false, Void, )
181209

@@ -197,6 +225,16 @@ __OMP_ATTRS_SET(GetterAttrs,
197225
? AttributeSet(EnumAttr(NoUnwind), EnumAttr(ReadOnly),
198226
EnumAttr(NoSync), EnumAttr(NoFree))
199227
: AttributeSet(EnumAttr(NoUnwind)))
228+
__OMP_ATTRS_SET(GetterArgWriteAttrs,
229+
OptimisticAttributes
230+
? AttributeSet(EnumAttr(NoUnwind), EnumAttr(NoSync),
231+
EnumAttr(NoFree))
232+
: AttributeSet(EnumAttr(NoUnwind)))
233+
__OMP_ATTRS_SET(SetterAttrs,
234+
OptimisticAttributes
235+
? AttributeSet(EnumAttr(NoUnwind), EnumAttr(WriteOnly),
236+
EnumAttr(NoSync), EnumAttr(NoFree))
237+
: AttributeSet(EnumAttr(NoUnwind)))
200238

201239
#undef __OMP_ATTRS_SET
202240
#undef OMP_ATTRS_SET
@@ -213,6 +251,41 @@ __OMP_RTL_ATTRS(__kmpc_fork_call, AttributeSet(EnumAttr(NoUnwind)),
213251

214252
__OMP_RTL_ATTRS(__kmpc_global_thread_num, GetterAttrs, AttributeSet(), {})
215253
__OMP_RTL_ATTRS(omp_get_thread_num, GetterAttrs, AttributeSet(), {})
254+
__OMP_RTL_ATTRS(omp_get_num_threads, GetterAttrs, AttributeSet(), {})
255+
__OMP_RTL_ATTRS(omp_get_max_threads, GetterAttrs, AttributeSet(), {})
256+
__OMP_RTL_ATTRS(omp_in_parallel, GetterAttrs, AttributeSet(), {})
257+
__OMP_RTL_ATTRS(omp_get_dynamic, GetterAttrs, AttributeSet(), {})
258+
__OMP_RTL_ATTRS(omp_get_cancellation, GetterAttrs, AttributeSet(), {})
259+
__OMP_RTL_ATTRS(omp_get_nested, GetterAttrs, AttributeSet(), {})
260+
__OMP_RTL_ATTRS(omp_get_schedule, GetterArgWriteAttrs, AttributeSet(),
261+
ArrayRef<AttributeSet>(
262+
{AttributeSet(EnumAttr(NoCapture), EnumAttr(WriteOnly)),
263+
AttributeSet(EnumAttr(NoCapture), EnumAttr(WriteOnly))}))
264+
__OMP_RTL_ATTRS(omp_get_thread_limit, GetterAttrs, AttributeSet(), {})
265+
__OMP_RTL_ATTRS(omp_get_supported_active_levels, GetterAttrs, AttributeSet(),
266+
{})
267+
__OMP_RTL_ATTRS(omp_get_max_active_levels, GetterAttrs, AttributeSet(), {})
268+
__OMP_RTL_ATTRS(omp_get_level, GetterAttrs, AttributeSet(), {})
269+
__OMP_RTL_ATTRS(omp_get_ancestor_thread_num, GetterAttrs, AttributeSet(), {})
270+
__OMP_RTL_ATTRS(omp_get_team_size, GetterAttrs, AttributeSet(), {})
271+
__OMP_RTL_ATTRS(omp_get_active_level, GetterAttrs, AttributeSet(), {})
272+
__OMP_RTL_ATTRS(omp_in_final, GetterAttrs, AttributeSet(), {})
273+
__OMP_RTL_ATTRS(omp_get_proc_bind, GetterAttrs, AttributeSet(), {})
274+
__OMP_RTL_ATTRS(omp_get_num_places, GetterAttrs, AttributeSet(), {})
275+
__OMP_RTL_ATTRS(omp_get_num_procs, GetterAttrs, AttributeSet(), {})
276+
__OMP_RTL_ATTRS(omp_get_place_proc_ids, GetterArgWriteAttrs, AttributeSet(),
277+
ArrayRef<AttributeSet>({AttributeSet(),
278+
AttributeSet(EnumAttr(NoCapture),
279+
EnumAttr(WriteOnly))}))
280+
__OMP_RTL_ATTRS(omp_get_place_num, GetterAttrs, AttributeSet(), {})
281+
__OMP_RTL_ATTRS(omp_get_partition_num_places, GetterAttrs, AttributeSet(), {})
282+
__OMP_RTL_ATTRS(omp_get_partition_place_nums, GetterAttrs, AttributeSet(), {})
283+
284+
__OMP_RTL_ATTRS(omp_set_num_threads, SetterAttrs, AttributeSet(), {})
285+
__OMP_RTL_ATTRS(omp_set_dynamic, SetterAttrs, AttributeSet(), {})
286+
__OMP_RTL_ATTRS(omp_set_nested, SetterAttrs, AttributeSet(), {})
287+
__OMP_RTL_ATTRS(omp_set_schedule, SetterAttrs, AttributeSet(), {})
288+
__OMP_RTL_ATTRS(omp_set_max_active_levels, SetterAttrs, AttributeSet(), {})
216289

217290
#undef __OMP_RTL_ATTRS
218291
#undef OMP_RTL_ATTRS

‎llvm/lib/Transforms/IPO/OpenMPOpt.cpp

Lines changed: 31 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -19,6 +19,7 @@
1919
#include "llvm/Analysis/CallGraph.h"
2020
#include "llvm/Analysis/CallGraphSCCPass.h"
2121
#include "llvm/Frontend/OpenMP/OMPConstants.h"
22+
#include "llvm/Frontend/OpenMP/OMPIRBuilder.h"
2223
#include "llvm/IR/CallSite.h"
2324
#include "llvm/InitializePasses.h"
2425
#include "llvm/Support/CommandLine.h"
@@ -52,9 +53,10 @@ struct OpenMPOpt {
5253
SmallPtrSetImpl<Function *> &ModuleSlice,
5354
CallGraphUpdater &CGUpdater)
5455
: M(*(*SCC.begin())->getParent()), SCC(SCC), ModuleSlice(ModuleSlice),
55-
CGUpdater(CGUpdater) {
56+
OMPBuilder(M), CGUpdater(CGUpdater) {
5657
initializeTypes(M);
5758
initializeRuntimeFunctions();
59+
OMPBuilder.initialize();
5860
}
5961

6062
/// Generic information that describes a runtime function
@@ -118,12 +120,36 @@ struct OpenMPOpt {
118120
bool deduplicateRuntimeCalls() {
119121
bool Changed = false;
120122

123+
RuntimeFunction DeduplicableRuntimeCallIDs[] = {
124+
OMPRTL_omp_get_num_threads,
125+
OMPRTL_omp_in_parallel,
126+
OMPRTL_omp_get_cancellation,
127+
OMPRTL_omp_get_thread_limit,
128+
OMPRTL_omp_get_supported_active_levels,
129+
OMPRTL_omp_get_level,
130+
OMPRTL_omp_get_ancestor_thread_num,
131+
OMPRTL_omp_get_team_size,
132+
OMPRTL_omp_get_active_level,
133+
OMPRTL_omp_in_final,
134+
OMPRTL_omp_get_proc_bind,
135+
OMPRTL_omp_get_num_places,
136+
OMPRTL_omp_get_num_procs,
137+
OMPRTL_omp_get_place_num,
138+
OMPRTL_omp_get_partition_num_places,
139+
OMPRTL_omp_get_partition_place_nums};
140+
141+
// Global-tid is handled separatly.
121142
SmallSetVector<Value *, 16> GTIdArgs;
122143
collectGlobalThreadIdArguments(GTIdArgs);
123144
LLVM_DEBUG(dbgs() << TAG << "Found " << GTIdArgs.size()
124145
<< " global thread ID arguments\n");
125146

126147
for (Function *F : SCC) {
148+
for (auto DeduplicableRuntimeCallID : DeduplicableRuntimeCallIDs)
149+
deduplicateRuntimeCalls(*F, RFIs[DeduplicableRuntimeCallID]);
150+
151+
// __kmpc_global_thread_num is special as we can replace it with an
152+
// argument in enough cases to make it worth trying.
127153
Value *GTIdArg = nullptr;
128154
for (Argument &Arg : F->args())
129155
if (GTIdArgs.count(&Arg)) {
@@ -132,7 +158,6 @@ struct OpenMPOpt {
132158
}
133159
Changed |= deduplicateRuntimeCalls(
134160
*F, RFIs[OMPRTL___kmpc_global_thread_num], GTIdArg);
135-
Changed |= deduplicateRuntimeCalls(*F, RFIs[OMPRTL_omp_get_thread_num]);
136161
}
137162

138163
return Changed;
@@ -259,6 +284,7 @@ struct OpenMPOpt {
259284
unsigned NumUses = 0;
260285
if (!RFI.Declaration)
261286
return NumUses;
287+
OMPBuilder.addAttributes(RFI.Kind, *RFI.Declaration);
262288

263289
NumOpenMPRuntimeFunctionsIdentified += 1;
264290
NumOpenMPRuntimeFunctionUsesIdentified += RFI.Declaration->getNumUses();
@@ -312,6 +338,9 @@ struct OpenMPOpt {
312338
/// The slice of the module we are allowed to look at.
313339
SmallPtrSetImpl<Function *> &ModuleSlice;
314340

341+
/// An OpenMP-IR-Builder instance
342+
OpenMPIRBuilder OMPBuilder;
343+
315344
/// Callback to update the call graph, the first argument is a removed call,
316345
/// the second an optional replacement call.
317346
CallGraphUpdater &CGUpdater;

0 commit comments

Comments
 (0)