summaryrefslogtreecommitdiffstats
path: root/clang
diff options
context:
space:
mode:
authorAlexey Bataev <a.bataev@hotmail.com>2015-06-17 06:21:39 +0000
committerAlexey Bataev <a.bataev@hotmail.com>2015-06-17 06:21:39 +0000
commit89e7e8eb0e18316647ccf1953154ffa29f97f545 (patch)
treebbeab00b38fc00db40c15d4d31426f89893c49aa /clang
parent8321006d44c351e4cd4d1383de7e9d11764da56e (diff)
downloadbcm5719-llvm-89e7e8eb0e18316647ccf1953154ffa29f97f545.tar.gz
bcm5719-llvm-89e7e8eb0e18316647ccf1953154ffa29f97f545.zip
[OPENMP] Supported reduction clause in omp simd construct.
The following code is generated for reduction clause within 'omp simd' loop construct: #pragma omp simd reduction(op:var) for (...) <body> alloca priv_var priv_var = <initial reduction value>; <loop_start>: <body> // references to original 'var' are replaced by 'priv_var' <loop_end>: var op= priv_var; llvm-svn: 239881
Diffstat (limited to 'clang')
-rw-r--r--clang/lib/CodeGen/CGOpenMPRuntime.cpp15
-rw-r--r--clang/lib/CodeGen/CGOpenMPRuntime.h2
-rw-r--r--clang/lib/CodeGen/CGStmtOpenMP.cpp6
-rw-r--r--clang/test/OpenMP/simd_codegen.cpp33
4 files changed, 50 insertions, 6 deletions
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 1238accf42d..269799dffc1 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2242,7 +2242,7 @@ void CGOpenMPRuntime::emitReduction(CodeGenFunction &CGF, SourceLocation Loc,
ArrayRef<const Expr *> LHSExprs,
ArrayRef<const Expr *> RHSExprs,
ArrayRef<const Expr *> ReductionOps,
- bool WithNowait) {
+ bool WithNowait, bool SimpleReduction) {
// Next code should be emitted for reduction:
//
// static kmp_critical_name lock = { 0 };
@@ -2272,9 +2272,22 @@ void CGOpenMPRuntime::emitReduction(CodeGenFunction &CGF, SourceLocation Loc,
// break;
// default:;
// }
+ //
+ // if SimpleReduction is true, only the next code is generated:
+ // ...
+ // <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+ // ...
auto &C = CGM.getContext();
+ if (SimpleReduction) {
+ CodeGenFunction::RunCleanupsScope Scope(CGF);
+ for (auto *E : ReductionOps) {
+ CGF.EmitIgnoredExpr(E);
+ }
+ return;
+ }
+
// 1. Build a list of reduction variables.
// void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
llvm::APInt ArraySize(/*unsigned int numBits=*/32, RHSExprs.size());
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.h b/clang/lib/CodeGen/CGOpenMPRuntime.h
index f5aa4a51df9..4db3db4ae9d 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.h
@@ -632,7 +632,7 @@ public:
ArrayRef<const Expr *> LHSExprs,
ArrayRef<const Expr *> RHSExprs,
ArrayRef<const Expr *> ReductionOps,
- bool WithNowait);
+ bool WithNowait, bool SimpleReduction);
/// \brief Emit code for 'taskwait' directive.
virtual void emitTaskwaitCall(CodeGenFunction &CGF, SourceLocation Loc);
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index 907fe93efc4..ed960fa858d 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -443,7 +443,9 @@ void CodeGenFunction::EmitOMPReductionClauseFinal(
CGM.getOpenMPRuntime().emitReduction(
*this, D.getLocEnd(), LHSExprs, RHSExprs, ReductionOps,
D.getSingleClause(OMPC_nowait) ||
- isOpenMPParallelDirective(D.getDirectiveKind()));
+ isOpenMPParallelDirective(D.getDirectiveKind()) ||
+ D.getDirectiveKind() == OMPD_simd,
+ D.getDirectiveKind() == OMPD_simd);
}
}
@@ -807,6 +809,7 @@ void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
EmitPrivateLoopCounters(CGF, LoopScope, S.counters());
EmitPrivateLinearVars(CGF, S, LoopScope);
CGF.EmitOMPPrivateClause(S, LoopScope);
+ CGF.EmitOMPReductionClauseInit(S, LoopScope);
HasLastprivateClause = CGF.EmitOMPLastprivateClauseInit(S, LoopScope);
(void)LoopScope.Privatize();
CGF.EmitOMPInnerLoop(S, LoopScope.requiresCleanups(),
@@ -820,6 +823,7 @@ void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
if (HasLastprivateClause) {
CGF.EmitOMPLastprivateClauseFinal(S);
}
+ CGF.EmitOMPReductionClauseFinal(S);
}
CGF.EmitOMPSimdFinal(S);
// Emit: if (PreCond) - end.
diff --git a/clang/test/OpenMP/simd_codegen.cpp b/clang/test/OpenMP/simd_codegen.cpp
index 0a5b38a0ce3..4bd06a7fdf2 100644
--- a/clang/test/OpenMP/simd_codegen.cpp
+++ b/clang/test/OpenMP/simd_codegen.cpp
@@ -185,9 +185,6 @@ void simple(float *a, float *b, float *c, float *d) {
// CHECK: store i32 -1, i32* [[A:%.+]],
A = -1;
#pragma omp simd lastprivate(A)
-// Clause 'lastprivate' implementation is not completed yet.
-// Test checks that one iteration is separated in presence of lastprivate.
-//
// CHECK: store i64 0, i64* [[OMP_IV7:%[^,]+]]
// CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
// CHECK: [[SIMD_LOOP7_COND]]
@@ -212,6 +209,36 @@ void simple(float *a, float *b, float *c, float *d) {
// CHECK: [[SIMPLE_LOOP7_END]]
// CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]],
// CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* [[A]],
+ int R;
+ // CHECK: store i32 -1, i32* [[R:%.+]],
+ R = -1;
+// CHECK: store i64 0, i64* [[OMP_IV8:%[^,]+]]
+// CHECK: store i32 1, i32* [[R_PRIV:%.+]],
+ #pragma omp simd reduction(*:R)
+// CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
+// CHECK: [[SIMD_LOOP8_COND]]
+// CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID:[0-9]+]]
+// CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
+// CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
+ for (long long i = -10; i < 10; i += 3) {
+// CHECK: [[SIMPLE_LOOP8_BODY]]
+// Start of body: calculate i from IV:
+// CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
+// CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
+// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+ R *= i;
+// CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
+// CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+ }
+// CHECK: [[SIMPLE_LOOP8_END]]
+// CHECK-NEXT: [[R_VAL:%.+]] = load i32, i32* [[R]],
+// CHECK-NEXT: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]],
+// CHECK-NEXT: [[RED:%.+]] = mul nsw i32 [[R_VAL]], [[R_PRIV_VAL]]
+// CHECK-NEXT: store i32 [[RED]], i32* [[R]],
// CHECK-NEXT: ret void
}
OpenPOWER on IntegriCloud