summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--polly/lib/Support/SCEVAffinator.cpp16
-rw-r--r--polly/test/ScopInfo/NonAffine/non_affine_but_srem.ll8
-rw-r--r--polly/test/ScopInfo/const_srem_sdiv.ll103
3 files changed, 115 insertions, 12 deletions
diff --git a/polly/lib/Support/SCEVAffinator.cpp b/polly/lib/Support/SCEVAffinator.cpp
index 73e3a00ed2c..347c12f60fd 100644
--- a/polly/lib/Support/SCEVAffinator.cpp
+++ b/polly/lib/Support/SCEVAffinator.cpp
@@ -548,17 +548,17 @@ __isl_give PWACtx SCEVAffinator::visitSDivInstruction(Instruction *SDiv) {
__isl_give PWACtx SCEVAffinator::visitSRemInstruction(Instruction *SRem) {
assert(SRem->getOpcode() == Instruction::SRem && "Assumed SRem instruction!");
- auto *Divisor = dyn_cast<ConstantInt>(SRem->getOperand(1));
- assert(Divisor && "SRem is no parameter but has a non-constant RHS.");
- auto *DivisorVal = isl_valFromAPInt(Ctx, Divisor->getValue(),
- /* isSigned */ true);
+ auto *Scope = getScope();
+ auto *Divisor = SRem->getOperand(1);
+ auto *DivisorSCEV = SE.getSCEVAtScope(Divisor, Scope);
+ auto DivisorPWAC = visit(DivisorSCEV);
+ assert(isa<ConstantInt>(Divisor) &&
+ "SRem is no parameter but has a non-constant RHS.");
auto *Dividend = SRem->getOperand(0);
- auto *DividendSCEV = SE.getSCEVAtScope(Dividend, getScope());
+ auto *DividendSCEV = SE.getSCEVAtScope(Dividend, Scope);
auto DividendPWAC = visit(DividendSCEV);
-
- DividendPWAC.first =
- isl_pw_aff_mod_val(DividendPWAC.first, isl_val_abs(DivisorVal));
+ combine(DividendPWAC, DivisorPWAC, isl_pw_aff_tdiv_r);
return DividendPWAC;
}
diff --git a/polly/test/ScopInfo/NonAffine/non_affine_but_srem.ll b/polly/test/ScopInfo/NonAffine/non_affine_but_srem.ll
index f9d0c5ae4c9..24f7953e225 100644
--- a/polly/test/ScopInfo/NonAffine/non_affine_but_srem.ll
+++ b/polly/test/ScopInfo/NonAffine/non_affine_but_srem.ll
@@ -19,9 +19,9 @@
; CHECK-NEXT: Schedule :=
; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> [i0] };
; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]
-; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and 0 <= o0 <= 41 };
+; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and -41 <= o0 <= 41 and ((n < 0 and o0 <= 0) or (n >= 0 and o0 >= 0)) }
; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
-; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and 0 <= o0 <= 41 };
+; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and -41 <= o0 <= 41 and ((n < 0 and o0 <= 0) or (n >= 0 and o0 >= 0)) }
; CHECK-NEXT: }
;
; CHECK: Statements {
@@ -31,9 +31,9 @@
; CHECK-NEXT: Schedule :=
; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> [i0] };
; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]
-; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and 0 <= o0 <= 41 };
+; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and -41 <= o0 <= 41 and ((n > 0 and o0 >= 0) or (n <= 0 and o0 <= 0)) }
; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
-; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and 0 <= o0 <= 41 };
+; CHECK-NEXT: [n] -> { Stmt_bb2[i0] -> MemRef_A[o0] : 42*floor((-n + o0)/42) = -n + o0 and -41 <= o0 <= 41 and ((n > 0 and o0 >= 0) or (n <= 0 and o0 <= 0)) }
; CHECK-NEXT: }
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
diff --git a/polly/test/ScopInfo/const_srem_sdiv.ll b/polly/test/ScopInfo/const_srem_sdiv.ll
new file mode 100644
index 00000000000..bf093afe1b4
--- /dev/null
+++ b/polly/test/ScopInfo/const_srem_sdiv.ll
@@ -0,0 +1,103 @@
+; RUN: opt %loadPolly -polly-scops -analyze < %s | FileCheck %s
+;
+; See http://research.microsoft.com/pubs/151917/divmodnote-letter.pdf
+;
+; void f(long *A) {
+; for (long i = 0; i < 10; i++) {
+; A[8 / 3] = A[8 % 3];
+; A[8 / -3] = A[8 % -3];
+; A[-8 / 3] = A[-8 % 3];
+; A[-8 / -3] = A[-8 % -3];
+; A[1 / 2] = A[1 % 2];
+; A[1 / -2] = A[1 % -2];
+; A[-1 / 2] = A[-1 % 2];
+; A[-1 / -2] = A[-1 % -2];
+; }
+; }
+;
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[-2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[-2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[1] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[1] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[-1] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_R[-1] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[-2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[-2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[2] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[0] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[0] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[0] };
+; CHECK: { Stmt_for_body[i0] -> MemRef_D[0] };
+;
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @f(i64* %D, i64* %R) {
+entry:
+ br label %for.cond
+
+for.cond: ; preds = %for.inc, %entry
+ %i.0 = phi i64 [ 0, %entry ], [ %inc, %for.inc ]
+ %exitcond = icmp ne i64 %i.0, 10
+ br i1 %exitcond, label %for.body, label %for.end
+
+for.body: ; preds = %for.cond
+ %rem = srem i64 8, 3
+ %arrayidx = getelementptr inbounds i64, i64* %R, i64 %rem
+ %tmp = load i64, i64* %arrayidx, align 8
+ %div = sdiv i64 8, 3
+ %arrayidx1 = getelementptr inbounds i64, i64* %D, i64 %div
+ store i64 %tmp, i64* %arrayidx1, align 8
+ %rem2 = srem i64 8, -3
+ %arrayidx3 = getelementptr inbounds i64, i64* %R, i64 %rem2
+ %tmp1 = load i64, i64* %arrayidx3, align 8
+ %div5 = sdiv i64 8, -3
+ %arrayidx6 = getelementptr inbounds i64, i64* %D, i64 %div5
+ store i64 %tmp1, i64* %arrayidx6, align 8
+ %rem8 = srem i64 -8, 3
+ %arrayidx9 = getelementptr inbounds i64, i64* %R, i64 %rem8
+ %tmp2 = load i64, i64* %arrayidx9, align 8
+ %div11 = sdiv i64 -8, 3
+ %arrayidx12 = getelementptr inbounds i64, i64* %D, i64 %div11
+ store i64 %tmp2, i64* %arrayidx12, align 8
+ %rem15 = srem i64 -8, -3
+ %arrayidx16 = getelementptr inbounds i64, i64* %R, i64 %rem15
+ %tmp3 = load i64, i64* %arrayidx16, align 8
+ %div19 = sdiv i64 -8, -3
+ %arrayidx20 = getelementptr inbounds i64, i64* %D, i64 %div19
+ store i64 %tmp3, i64* %arrayidx20, align 8
+ %rem29 = srem i64 1, 2
+ %arrayidx30 = getelementptr inbounds i64, i64* %R, i64 %rem29
+ %tmp5 = load i64, i64* %arrayidx30, align 8
+ %div31 = sdiv i64 1, 2
+ %arrayidx32 = getelementptr inbounds i64, i64* %D, i64 %div31
+ store i64 %tmp5, i64* %arrayidx32, align 8
+ %rem34 = srem i64 1, -2
+ %arrayidx35 = getelementptr inbounds i64, i64* %R, i64 %rem34
+ %tmp6 = load i64, i64* %arrayidx35, align 8
+ %div37 = sdiv i64 1, -2
+ %arrayidx38 = getelementptr inbounds i64, i64* %D, i64 %div37
+ store i64 %tmp6, i64* %arrayidx38, align 8
+ %rem40 = srem i64 -1, 2
+ %arrayidx41 = getelementptr inbounds i64, i64* %R, i64 %rem40
+ %tmp7 = load i64, i64* %arrayidx41, align 8
+ %div43 = sdiv i64 -1, 2
+ %arrayidx44 = getelementptr inbounds i64, i64* %D, i64 %div43
+ store i64 %tmp7, i64* %arrayidx44, align 8
+ %rem47 = srem i64 -1, -2
+ %arrayidx48 = getelementptr inbounds i64, i64* %R, i64 %rem47
+ %tmp8 = load i64, i64* %arrayidx48, align 8
+ %div51 = sdiv i64 -1, -2
+ %arrayidx52 = getelementptr inbounds i64, i64* %D, i64 %div51
+ store i64 %tmp8, i64* %arrayidx52, align 8
+ br label %for.inc
+
+for.inc: ; preds = %for.body
+ %inc = add nuw nsw i64 %i.0, 1
+ br label %for.cond
+
+for.end: ; preds = %for.cond
+ ret void
+}
OpenPOWER on IntegriCloud