summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--polly/lib/Analysis/ScopDetection.cpp17
-rw-r--r--polly/lib/Analysis/TempScopInfo.cpp30
-rw-r--r--polly/test/Isl/Ast/OpenMP/nested_loop_both_parallel_parametric.ll21
-rw-r--r--polly/test/ScopInfo/constant_start_integer.ll4
4 files changed, 41 insertions, 31 deletions
diff --git a/polly/lib/Analysis/ScopDetection.cpp b/polly/lib/Analysis/ScopDetection.cpp
index 5f03e7a5b45..9ce0fc767c7 100644
--- a/polly/lib/Analysis/ScopDetection.cpp
+++ b/polly/lib/Analysis/ScopDetection.cpp
@@ -364,12 +364,19 @@ bool ScopDetection::isValidMemoryAccess(Instruction &Inst,
return invalid<ReportVariantBasePtr>(Context, /*Assert=*/false, BaseValue);
AccessFunction = SE->getMinusSCEV(AccessFunction, BasePointer);
- const SCEVAddRecExpr *AF = dyn_cast<SCEVAddRecExpr>(AccessFunction);
if (AllowNonAffine) {
// Do not check whether AccessFunction is affine.
- } else if (PollyDelinearize && AF) {
- // Try to delinearize AccessFunction.
+ } else if (!isAffineExpr(&Context.CurRegion, AccessFunction, *SE,
+ BaseValue)) {
+ const SCEVAddRecExpr *AF = dyn_cast<SCEVAddRecExpr>(AccessFunction);
+ if (!PollyDelinearize || !AF)
+ return invalid<ReportNonAffineAccess>(Context, /*Assert=*/true,
+ AccessFunction);
+
+ // Try to delinearize AccessFunction only when the expression is known to
+ // not be affine: as all affine functions can be represented without
+ // problems in Polly, we do not have to delinearize them.
SmallVector<const SCEV *, 4> Subscripts, Sizes;
AF->delinearize(*SE, Subscripts, Sizes);
int size = Subscripts.size();
@@ -378,10 +385,6 @@ bool ScopDetection::isValidMemoryAccess(Instruction &Inst,
if (!isAffineExpr(&Context.CurRegion, Subscripts[i], *SE, BaseValue))
return invalid<ReportNonAffineAccess>(Context, /*Assert=*/true,
AccessFunction);
- } else if (!isAffineExpr(&Context.CurRegion, AccessFunction, *SE,
- BaseValue)) {
- return invalid<ReportNonAffineAccess>(Context, /*Assert=*/true,
- AccessFunction);
}
// FIXME: Alias Analysis thinks IntToPtrInst aliases with alloca instructions
diff --git a/polly/lib/Analysis/TempScopInfo.cpp b/polly/lib/Analysis/TempScopInfo.cpp
index 34a4f82e507..7bfdeb12a88 100644
--- a/polly/lib/Analysis/TempScopInfo.cpp
+++ b/polly/lib/Analysis/TempScopInfo.cpp
@@ -167,30 +167,32 @@ IRAccess TempScopInfo::buildIRAccess(Instruction *Inst, Loop *L, Region *R) {
AccessFunction = SE->getMinusSCEV(AccessFunction, BasePointer);
SmallVector<const SCEV *, 4> Subscripts, Sizes;
- bool IsAffine = true;
+ bool IsAffine = isAffineExpr(R, AccessFunction, *SE, BasePointer->getValue());
const SCEVAddRecExpr *AF = dyn_cast<SCEVAddRecExpr>(AccessFunction);
- if (PollyDelinearize && AF) {
+ if (!IsAffine && PollyDelinearize && AF) {
const SCEV *Remainder = AF->delinearize(*SE, Subscripts, Sizes);
int NSubs = Subscripts.size();
- // Normalize the last dimension: integrate the size of the "scalar
- // dimension" and the remainder of the delinearization.
- Subscripts[NSubs - 1] =
- SE->getMulExpr(Subscripts[NSubs - 1], Sizes[NSubs - 1]);
- Subscripts[NSubs - 1] = SE->getAddExpr(Subscripts[NSubs - 1], Remainder);
-
- for (int i = 0; i < NSubs; ++i)
- if (!isAffineExpr(R, Subscripts[i], *SE, BasePointer->getValue())) {
- IsAffine = false;
- break;
- }
+ if (NSubs > 0) {
+ // Normalize the last dimension: integrate the size of the "scalar
+ // dimension" and the remainder of the delinearization.
+ Subscripts[NSubs - 1] =
+ SE->getMulExpr(Subscripts[NSubs - 1], Sizes[NSubs - 1]);
+ Subscripts[NSubs - 1] = SE->getAddExpr(Subscripts[NSubs - 1], Remainder);
+
+ IsAffine = true;
+ for (int i = 0; i < NSubs; ++i)
+ if (!isAffineExpr(R, Subscripts[i], *SE, BasePointer->getValue())) {
+ IsAffine = false;
+ break;
+ }
+ }
}
if (Subscripts.size() == 0) {
Subscripts.push_back(AccessFunction);
Sizes.push_back(SE->getConstant(ZeroOffset->getType(), Size));
- IsAffine = isAffineExpr(R, AccessFunction, *SE, BasePointer->getValue());
}
return IRAccess(Type, BasePointer->getValue(), AccessFunction, Size, IsAffine,
diff --git a/polly/test/Isl/Ast/OpenMP/nested_loop_both_parallel_parametric.ll b/polly/test/Isl/Ast/OpenMP/nested_loop_both_parallel_parametric.ll
index 2c11fc5d203..39b9108de1f 100644
--- a/polly/test/Isl/Ast/OpenMP/nested_loop_both_parallel_parametric.ll
+++ b/polly/test/Isl/Ast/OpenMP/nested_loop_both_parallel_parametric.ll
@@ -1,11 +1,12 @@
; RUN: opt %loadPolly -polly-ast -polly-ast-detect-parallel -analyze -polly-delinearize < %s | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
target triple = "x86_64-pc-linux-gnu"
-
-; for (i = 0; i < n; i++)
-; for (j = 0; j < n; j++)
-; A[i][j] = 1;
-
+; int A[1024][1024];
+; void bar(int n) {
+; for (i = 0; i < n; i++)
+; for (j = 0; j < n; j++)
+; A[i][j] = 1;
+; }
@A = common global [1024 x [1024 x i32]] zeroinitializer
define void @bar(i64 %n) {
start:
@@ -40,17 +41,21 @@ ret:
ret void
}
-; At the first look both loops seem parallel, however due to the delinearization
-; we get the following dependences:
+; At the first look both loops seem parallel, however due to the linearization
+; of memory access functions, we get the following dependences:
; [n] -> { loop_body[i0, i1] -> loop_body[1024 + i0, -1 + i1]:
; 0 <= i0 < n - 1024 and 1 <= i1 < n}
; They cause the outer loop to be non-parallel. We can only prove their
; absence, if we know that n < 1024. This information is currently not available
; to polly. However, we should be able to obtain it due to the out of bounds
; memory accesses, that would happen if n >= 1024.
+
+; Note that we do not delinearize this access function because it is considered
+; to already be affine: {{0,+,4}<%loop.i>,+,4096}<%loop.j>.
+
;
-; CHECK: #pragma omp parallel for
; CHECK: for (int c1 = 0; c1 < n; c1 += 1)
; CHECK: #pragma simd
+; CHECK: #pragma omp parallel for
; CHECK: for (int c3 = 0; c3 < n; c3 += 1)
; CHECK: Stmt_loop_body(c1, c3);
diff --git a/polly/test/ScopInfo/constant_start_integer.ll b/polly/test/ScopInfo/constant_start_integer.ll
index 5e1c901dc2e..f28b9d83557 100644
--- a/polly/test/ScopInfo/constant_start_integer.ll
+++ b/polly/test/ScopInfo/constant_start_integer.ll
@@ -16,9 +16,9 @@ target triple = "x86_64-unknown-linux-gnu"
; CHECK-NOT: p1
; CHECK: ReadAccess
-; CHECK: [p_0] -> { Stmt_for_body3[i0] -> MemRef_input[p_0, 1 + i0] };
+; CHECK: [p_0] -> { Stmt_for_body3[i0] -> MemRef_input[o0] : 4o0 = 4 + p_0 + 4i0 };
; CHECK: MustWriteAccess
-; CHECK: [p_0] -> { Stmt_for_body3[i0] -> MemRef_input[p_0, i0] };
+; CHECK: [p_0] -> { Stmt_for_body3[i0] -> MemRef_input[o0] : 4o0 = p_0 + 4i0 };
define void @foo(float* nocapture %input) {
entry:
OpenPOWER on IntegriCloud