forked from OSchip/llvm-project
Fix crashes due to unaligned parameters
Due to the recent introduction of isl_id, parameters need now always to be aligned. This was not yet taken care of in the code path of vectorization and dependence analysis. llvm-svn: 138555
This commit is contained in:
Tobias Grosser
parent
b406d227d2
commit
c532f12965
|
|
@ -264,6 +264,7 @@ bool Dependences::isParallelDimension(isl_set *loopDomain,
|
|||
unsigned parallelDimension) {
|
||||
Scop *S = &getCurScop();
|
||||
isl_union_map *schedule = getCombinedScheduleForDim(S, parallelDimension);
|
||||
isl_dim *dimModel = isl_union_map_get_dim(schedule);
|
||||
|
||||
// Calculate distance vector.
|
||||
isl_union_set *scheduleSubset;
|
||||
|
|
@ -310,8 +311,7 @@ bool Dependences::isParallelDimension(isl_set *loopDomain,
|
|||
|
||||
isl_union_set *distance_waw = isl_union_map_deltas(restrictedDeps_waw);
|
||||
|
||||
isl_dim *dim = isl_dim_set_alloc(S->getCtx(), S->getNumParams(),
|
||||
parallelDimension);
|
||||
isl_dim *dim = isl_dim_set_alloc(S->getCtx(), 0, parallelDimension);
|
||||
|
||||
// [0, 0, 0, 0] - All zero
|
||||
isl_basic_set *allZeroBS = isl_basic_set_universe(isl_dim_copy(dim));
|
||||
|
|
@ -328,6 +328,7 @@ bool Dependences::isParallelDimension(isl_set *loopDomain,
|
|||
}
|
||||
|
||||
isl_set *allZero = isl_set_from_basic_set(allZeroBS);
|
||||
allZero = isl_set_align_params(allZero, isl_dim_copy(dimModel));
|
||||
|
||||
// All zero, last unknown.
|
||||
// [0, 0, 0, ?]
|
||||
|
|
@ -345,6 +346,7 @@ bool Dependences::isParallelDimension(isl_set *loopDomain,
|
|||
}
|
||||
|
||||
isl_set *lastUnknown = isl_set_from_basic_set(lastUnknownBS);
|
||||
lastUnknown = isl_set_align_params(lastUnknown, dimModel);
|
||||
|
||||
// Valid distance vectors
|
||||
isl_set *validDistances = isl_set_subtract(lastUnknown, allZero);
|
||||
|
|
|
|||
|
|
@ -1139,8 +1139,10 @@ public:
|
|||
|
||||
isl_set *elements = isl_map_range(sub);
|
||||
|
||||
if (!isl_set_is_singleton(elements))
|
||||
if (!isl_set_is_singleton(elements)) {
|
||||
isl_set_free(elements);
|
||||
return -1;
|
||||
}
|
||||
|
||||
isl_point *p = isl_set_sample_point(elements);
|
||||
|
||||
|
|
|
|||
|
|
@ -76,8 +76,7 @@ static void extendScattering(Scop &S, unsigned scatDimensions) {
|
|||
|
||||
isl_map *scattering = stmt->getScattering();
|
||||
isl_dim *dim = isl_dim_alloc(isl_map_get_ctx(scattering),
|
||||
isl_map_n_param(scattering),
|
||||
isl_map_n_out(scattering),
|
||||
0, isl_map_n_out(scattering),
|
||||
scatDimensions);
|
||||
isl_basic_map *changeScattering = isl_basic_map_universe(isl_dim_copy(dim));
|
||||
|
||||
|
|
@ -96,6 +95,8 @@ static void extendScattering(Scop &S, unsigned scatDimensions) {
|
|||
|
||||
isl_map *changeScatteringMap = isl_map_from_basic_map(changeScattering);
|
||||
|
||||
isl_dim *dimModel = isl_map_get_dim(scattering);
|
||||
changeScatteringMap = isl_map_align_params(changeScatteringMap, dimModel);
|
||||
stmt->setScattering(isl_map_apply_range(scattering, changeScatteringMap));
|
||||
isl_dim_free(dim);
|
||||
}
|
||||
|
|
@ -128,7 +129,7 @@ static void extendScattering(Scop &S, unsigned scatDimensions) {
|
|||
// S(i,j)
|
||||
//
|
||||
static isl_basic_map *getTileMap(isl_ctx *ctx, int scheduleDimensions,
|
||||
int parameterDimensions, int tileSize = 32) {
|
||||
isl_dim *dimModel, int tileSize = 32) {
|
||||
// We construct
|
||||
//
|
||||
// tileMap := [p0] -> {[s0, s1] -> [t0, t1, p0, p1, a0, a1]:
|
||||
|
|
@ -136,7 +137,7 @@ static isl_basic_map *getTileMap(isl_ctx *ctx, int scheduleDimensions,
|
|||
// s1 = a1 * 32 and s1 = p1 and t1 <= p1 < t1 + 32}
|
||||
//
|
||||
// and project out the auxilary dimensions a0 and a1.
|
||||
isl_dim *dim = isl_dim_alloc(ctx, parameterDimensions, scheduleDimensions,
|
||||
isl_dim *dim = isl_dim_alloc(ctx, 0, scheduleDimensions,
|
||||
scheduleDimensions * 3);
|
||||
isl_basic_map *tileMap = isl_basic_map_universe(isl_dim_copy(dim));
|
||||
|
||||
|
|
@ -188,7 +189,7 @@ static isl_basic_map *getTileMap(isl_ctx *ctx, int scheduleDimensions,
|
|||
|
||||
isl_union_map *getTiledPartialSchedule(isl_band *band) {
|
||||
isl_union_map *partialSchedule;
|
||||
int scheduleDimensions, parameterDimensions;
|
||||
int scheduleDimensions;
|
||||
isl_ctx *ctx;
|
||||
isl_dim *dim;
|
||||
isl_basic_map *tileMap;
|
||||
|
|
@ -198,15 +199,12 @@ isl_union_map *getTiledPartialSchedule(isl_band *band) {
|
|||
ctx = isl_union_map_get_ctx(partialSchedule);
|
||||
dim = isl_union_map_get_dim(partialSchedule);
|
||||
scheduleDimensions = isl_band_n_member(band);
|
||||
parameterDimensions = isl_dim_size(dim, isl_dim_param);
|
||||
|
||||
tileMap = getTileMap(ctx, scheduleDimensions, parameterDimensions);
|
||||
tileMap = getTileMap(ctx, scheduleDimensions, dim);
|
||||
tileUnionMap = isl_union_map_from_map(isl_map_from_basic_map(tileMap));
|
||||
|
||||
tileUnionMap = isl_union_map_align_params(tileUnionMap, dim);
|
||||
partialSchedule = isl_union_map_apply_range(partialSchedule, tileUnionMap);
|
||||
|
||||
isl_dim_free(dim);
|
||||
|
||||
return partialSchedule;
|
||||
}
|
||||
|
||||
|
|
@ -288,8 +286,6 @@ static isl_union_map *tileBandList(isl_band_list *blist) {
|
|||
partialSchedule = getTiledPartialSchedule(band);
|
||||
int scheduleDimensions = isl_band_n_member(band);
|
||||
isl_dim *dim = isl_union_map_get_dim(partialSchedule);
|
||||
int parameterDimensions = isl_dim_size(dim, isl_dim_param);
|
||||
isl_dim_free(dim);
|
||||
|
||||
|
||||
if (isl_band_has_children(band)) {
|
||||
|
|
@ -307,9 +303,10 @@ static isl_union_map *tileBandList(isl_band_list *blist) {
|
|||
for (int i = scheduleDimensions - 1 ; i >= 0 ; i--) {
|
||||
if (isl_band_member_is_zero_distance(band, i)) {
|
||||
tileMap = getPrevectorMap(ctx, scheduleDimensions + i,
|
||||
scheduleDimensions * 2,
|
||||
parameterDimensions);
|
||||
scheduleDimensions * 2, 0);
|
||||
tileUnionMap = isl_union_map_from_map(tileMap);
|
||||
tileUnionMap = isl_union_map_align_params(tileUnionMap,
|
||||
isl_dim_copy(dim));
|
||||
partialSchedule = isl_union_map_apply_range(partialSchedule,
|
||||
tileUnionMap);
|
||||
break;
|
||||
|
|
@ -323,6 +320,7 @@ static isl_union_map *tileBandList(isl_band_list *blist) {
|
|||
finalSchedule = partialSchedule;
|
||||
|
||||
isl_band_free(band);
|
||||
isl_dim_free(dim);
|
||||
}
|
||||
|
||||
return finalSchedule;
|
||||
|
|
|
|||
|
|
@ -1,6 +1,7 @@
|
|||
set(POLLY_TEST_DIRECTORIES
|
||||
"ScopInfo"
|
||||
"AffineIterator"
|
||||
"ScopInfo"
|
||||
"ScheduleOptimizer"
|
||||
"CodeGen"
|
||||
"OpenMP"
|
||||
"polybench"
|
||||
|
|
|
|||
|
|
@ -0,0 +1,50 @@
|
|||
; RUN: opt %loadPolly -polly-optimize-isl -polly-cloog -analyze %s -S | FileCheck %s
|
||||
; RUN: opt %loadPolly -polly-optimize-isl -polly-cloog -analyze -enable-schedule-prevector %s -S | FileCheck %s -check-prefix=VECTOR
|
||||
|
||||
|
||||
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"
|
||||
|
||||
define void @f(i32* nocapture %A, i32 %N, i32 %C) nounwind {
|
||||
bb:
|
||||
%tmp1 = icmp sgt i32 %N, 0
|
||||
br i1 %tmp1, label %.lr.ph, label %._crit_edge
|
||||
|
||||
.lr.ph: ; preds = %bb
|
||||
%tmp = zext i32 %N to i64
|
||||
br label %bb2
|
||||
|
||||
bb2: ; preds = %bb2, %.lr.ph
|
||||
%indvar = phi i64 [ 0, %.lr.ph ], [ %indvar.next, %bb2 ]
|
||||
%scevgep = getelementptr i32* %A, i64 %indvar
|
||||
%tmp3 = load i32* %scevgep, align 4
|
||||
%tmp4 = add nsw i32 %tmp3, %C
|
||||
store i32 %tmp4, i32* %scevgep, align 4
|
||||
%indvar.next = add i64 %indvar, 1
|
||||
%exitcond = icmp eq i64 %indvar.next, %tmp
|
||||
br i1 %exitcond, label %._crit_edge, label %bb2
|
||||
|
||||
._crit_edge: ; preds = %bb2, %bb
|
||||
ret void
|
||||
}
|
||||
|
||||
; CHECK: if (M >= 1) {
|
||||
; CHECK: for (c1=0;c1<=M-1;c1+=32) {
|
||||
; CHECK: for (c2=c1;c2<=min(M-1,c1+31);c2++) {
|
||||
; CHECK: Stmt_bb2(c2);
|
||||
; CHECK: }
|
||||
; CHECK: }
|
||||
; CHECK: }
|
||||
|
||||
; VECTOR: if (M >= 1) {
|
||||
; VECTOR: for (c1=0;c1<=M-1;c1+=32) {
|
||||
; VECTOR: for (c2=-4*floord(-c1,4);c2<=min(M-1,c1+31);c2+=4) {
|
||||
; VECTOR: for (c3=c2;c3<=min(M-1,c2+3);c3++) {
|
||||
; VECTOR: Stmt_bb2(c3);
|
||||
; VECTOR: }
|
||||
; VECTOR: }
|
||||
; VECTOR: }
|
||||
; VECTOR: }
|
||||
|
||||
|
||||
Loading…
Reference in New Issue