Skip to content

Commit 15c645f

Browse files
preamespreames
committed
[RISCV] Enable (scalable) vectorization by default
This change enables vectorization (using scalable vectorization only, fixed vectors are not yet enabled) for RISCV when vector instructions are available for the target configuration. At this point, the resulting configuration should be both stable (e.g. no crashes), and profitable (i.e. few cases where scalar loops beat vector ones), but is not going to be particularly well tuned (i.e. we emit the best possible vector loop). The goal of this change is to align testing across organizations and ensure the default configuration matches what downstreams are using as closely as possible. This exposes a large amount of code which hasn't otherwise been on by default, and thus may not have been fully exercised. Given that, having issues fall out is not unexpected. If you find issues, please make sure to include as much information as you can when reverting this change. Differential Revision: https://reviews.llvm.org/D129013
1 parent 17fb879 commit 15c645f

File tree

2 files changed

+64
-9
lines changed

2 files changed

+64
-9
lines changed

‎llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -57,6 +57,7 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
5757

5858
bool shouldExpandReduction(const IntrinsicInst *II) const;
5959
bool supportsScalableVectors() const { return ST->hasVInstructions(); }
60+
bool enableScalableVectorization() const { return ST->hasVInstructions(); }
6061
PredicationStyle emitGetActiveLaneMask() const {
6162
return ST->hasVInstructions() ? PredicationStyle::Data
6263
: PredicationStyle::None;

‎llvm/test/Transforms/LoopVectorize/RISCV/defaults.ll

Lines changed: 63 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -13,16 +13,43 @@ target triple = "riscv64"
1313
define void @vector_add(ptr noalias nocapture %a, i64 %v) {
1414
; CHECK-LABEL: @vector_add(
1515
; CHECK-NEXT: entry:
16+
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
17+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP0]]
18+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
19+
; CHECK: vector.ph:
20+
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
21+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]]
22+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
23+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[V:%.*]], i32 0
24+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 1 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer
25+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
26+
; CHECK: vector.body:
27+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
28+
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0
29+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP2]]
30+
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP3]], i32 0
31+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 1 x i64>, ptr [[TMP4]], align 8
32+
; CHECK-NEXT: [[TMP5:%.*]] = add <vscale x 1 x i64> [[WIDE_LOAD]], [[BROADCAST_SPLAT]]
33+
; CHECK-NEXT: store <vscale x 1 x i64> [[TMP5]], ptr [[TMP4]], align 8
34+
; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
35+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
36+
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
37+
; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
38+
; CHECK: middle.block:
39+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
40+
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
41+
; CHECK: scalar.ph:
42+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1643
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
1744
; CHECK: for.body:
18-
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
19-
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[IV]]
45+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
46+
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]]
2047
; CHECK-NEXT: [[ELEM:%.*]] = load i64, ptr [[ARRAYIDX]], align 8
21-
; CHECK-NEXT: [[ADD:%.*]] = add i64 [[ELEM]], [[V:%.*]]
48+
; CHECK-NEXT: [[ADD:%.*]] = add i64 [[ELEM]], [[V]]
2249
; CHECK-NEXT: store i64 [[ADD]], ptr [[ARRAYIDX]], align 8
2350
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
2451
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024
25-
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY]]
52+
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
2653
; CHECK: for.end:
2754
; CHECK-NEXT: ret void
2855
;
@@ -46,18 +73,45 @@ for.end:
4673
define i64 @vector_add_reduce(ptr noalias nocapture %a) {
4774
; CHECK-LABEL: @vector_add_reduce(
4875
; CHECK-NEXT: entry:
76+
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
77+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP0]]
78+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
79+
; CHECK: vector.ph:
80+
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
81+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]]
82+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
83+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
84+
; CHECK: vector.body:
85+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
86+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 1 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
87+
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0
88+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP2]]
89+
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP3]], i32 0
90+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 1 x i64>, ptr [[TMP4]], align 8
91+
; CHECK-NEXT: [[TMP5]] = add <vscale x 1 x i64> [[VEC_PHI]], [[WIDE_LOAD]]
92+
; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
93+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
94+
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
95+
; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
96+
; CHECK: middle.block:
97+
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vector.reduce.add.nxv1i64(<vscale x 1 x i64> [[TMP5]])
98+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
99+
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
100+
; CHECK: scalar.ph:
101+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
102+
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ]
49103
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
50104
; CHECK: for.body:
51-
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
52-
; CHECK-NEXT: [[SUM:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[SUM_NEXT:%.*]], [[FOR_BODY]] ]
53-
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[IV]]
105+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
106+
; CHECK-NEXT: [[SUM:%.*]] = phi i64 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[SUM_NEXT:%.*]], [[FOR_BODY]] ]
107+
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]]
54108
; CHECK-NEXT: [[ELEM:%.*]] = load i64, ptr [[ARRAYIDX]], align 8
55109
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
56110
; CHECK-NEXT: [[SUM_NEXT]] = add i64 [[SUM]], [[ELEM]]
57111
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024
58-
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY]]
112+
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
59113
; CHECK: for.end:
60-
; CHECK-NEXT: [[SUM_NEXT_LCSSA:%.*]] = phi i64 [ [[SUM_NEXT]], [[FOR_BODY]] ]
114+
; CHECK-NEXT: [[SUM_NEXT_LCSSA:%.*]] = phi i64 [ [[SUM_NEXT]], [[FOR_BODY]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ]
61115
; CHECK-NEXT: ret i64 [[SUM_NEXT_LCSSA]]
62116
;
63117
entry:

0 commit comments

Comments
 (0)