replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[ScopInfo] Simplify inbounds assumptions under domain constraints 

Without this simplification for a loop nest:

  void foo(long n1_a, long n1_b, long n1_c, long n1_d,
           long p1_b, long p1_c, long p1_d,
           float A_1[][p1_b][p1_c][p1_d]) {
    for (long i = 0; i < n1_a; i++)
      for (long j = 0; j < n1_b; j++)
        for (long k = 0; k < n1_c; k++)
          for (long l = 0; l < n1_d; l++)
            A_1[i][j][k][l] += i + j + k + l;
 }

the assumption:

  n1_a <= 0 or (n1_a > 0 and n1_b <= 0) or
  (n1_a > 0 and n1_b > 0 and n1_c <= 0) or
  (n1_a > 0 and n1_b > 0 and n1_c > 0 and n1_d <= 0) or
  (n1_a > 0 and n1_b > 0 and n1_c > 0 and n1_d > 0 and
   p1_b >= n1_b and p1_c >= n1_c and p1_d >= n1_d)

is taken rather than the simpler assumption:

  p9_b >= n9_b and p9_c >= n9_c and p9_d >= n9_d.

The former is less strict, as it allows arbitrary values of p1_* in case, the
loop is not executed at all. However, in practice these precise constraints
explode when combined across different accesses and loops. For now it seems
to make more sense to take less precise, but more scalable constraints by
default. In case we find a practical example where more precise constraints
are needed, we can think about allowing such precise constraints in specific
situations where they help.

This change speeds up the new test case from taking very long (waited at least
a minute, but it probably takes a lot more) to below a second.

llvm-svn: 296456
This commit is contained in:
Tobias Grosser 2017-02-28 09:45:54 +00:00
parent 566a15d749
commit d7c4975349
6 changed files with 786 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
polly/lib/Analysis/ScopInfo.cpp
View File

@ -137,6 +137,11 @@ static cl::opt<bool> UnprofitableScalarAccs(
cl::desc("Count statements with scalar accesses as not optimizable"),
cl::Hidden, cl::init(true), cl::cat(PollyCategory));
static cl::opt<bool> PollyPreciseInbounds(
"polly-precise-inbounds",
cl::desc("Take more precise inbounds assumptions (do not scale well)"),
cl::Hidden, cl::init(false), cl::cat(PollyCategory));
//===----------------------------------------------------------------------===//
// Create a sequence of two schedules. Either argument may be null and is
@ -703,6 +708,8 @@ void MemoryAccess::assumeNoOutOfBound() {
const auto &Loc = getAccessInstruction()
? getAccessInstruction()->getDebugLoc()
: DebugLoc();
if (!PollyPreciseInbounds)
Outside = isl_set_gist(Outside, isl_set_params(Statement->getDomain()));
Statement->getParent()->recordAssumption(INBOUNDS, Outside, Loc,
AS_ASSUMPTION);
isl_space_free(Space);

4
polly/test/Isl/Ast/simple-run-time-condition.ll
View File

@ -1,4 +1,6 @@
; RUN: opt %loadPolly -polly-ast -analyze < %s | FileCheck %s
; RUN: opt %loadPolly -polly-ast -analyze -polly-precise-inbounds < %s \
; RUN: | 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-S128"
; void foo(long n, long m, int o, double A[n][m], long p, long q) {

3
polly/test/ScopInfo/assume_gep_bounds_2.ll
View File

@ -1,4 +1,5 @@
; RUN: opt %loadPolly -basicaa -polly-scops -analyze < %s | FileCheck %s
; RUN: opt %loadPolly -basicaa -polly-scops -analyze < %s \
; RUN: -polly-precise-inbounds | FileCheck %s
;
; void foo(float A[restrict][20], float B[restrict][20], long n, long m,
; long p) {

771
polly/test/ScopInfo/assume_gep_bounds_many.ll Normal file
View File

@ -0,0 +1,771 @@
; RUN: opt %loadPolly -analyze -polly-scops -polly-ignore-aliasing \
; RUN: < %s | FileCheck %s
; CHECK: Assumed Context:
; CHECK-NEXT: [n1_a, n1_b, n1_c, n1_d, n2_a, n2_b, n2_c, n2_d, n3_a, n3_b, n3_c, n3_d, n4_a, n4_b, n4_c, n4_d, n5_a, n5_b, n5_c, n5_d, n6_a, n6_b, n6_c, n6_d, n7_a, n7_b, n7_c, n7_d, n8_a, n8_b, n8_c, n8_d, n9_a, n9_b, n9_c, n9_d, p1_b, p1_c, p1_d, p2_b, p2_c, p2_d, p3_b, p3_c, p3_d, p4_b, p4_c, p4_d, p5_b, p5_c, p5_d, p6_b, p6_c, p6_d, p7_b, p7_c, p7_d, p8_b, p8_c, p8_d, p9_b, p9_c, p9_d] -> { : p1_b >= n1_b and p1_c >= n1_c and p1_d >= n1_d and p2_b >= n2_b and p2_c >= n2_c and p2_d >= n2_d and p3_b >= n3_b and p3_c >= n3_c and p3_d >= n3_d and p4_b >= n4_b and p4_c >= n4_c and p4_d >= n4_d and p5_b >= n5_b and p5_c >= n5_c and p5_d >= n5_d and p6_b >= n6_b and p6_c >= n6_c and p6_d >= n6_d and p7_b >= n7_b and p7_c >= n7_c and p7_d >= n7_d and p8_b >= n8_b and p8_c >= n8_c and p8_d >= n8_d and p9_b >= n9_b and p9_c >= n9_c and p9_d >= n9_d }
; CHECK-NEXT: Invalid Context:
; CHECK-NEXT: [n1_a, n1_b, n1_c, n1_d, n2_a, n2_b, n2_c, n2_d, n3_a, n3_b, n3_c, n3_d, n4_a, n4_b, n4_c, n4_d, n5_a, n5_b, n5_c, n5_d, n6_a, n6_b, n6_c, n6_d, n7_a, n7_b, n7_c, n7_d, n8_a, n8_b, n8_c, n8_d, n9_a, n9_b, n9_c, n9_d, p1_b, p1_c, p1_d, p2_b, p2_c, p2_d, p3_b, p3_c, p3_d, p4_b, p4_c, p4_d, p5_b, p5_c, p5_d, p6_b, p6_c, p6_d, p7_b, p7_c, p7_d, p8_b, p8_c, p8_d, p9_b, p9_c, p9_d] -> { : 1 = 0 }
;
; void foo(long n1_a, long n1_b, long n1_c, long n1_d, long n2_a, long n2_b,
; long n2_c, long n2_d, long n3_a, long n3_b, long n3_c, long n3_d,
; long n4_a, long n4_b, long n4_c, long n4_d, long n5_a, long n5_b,
; long n5_c, long n5_d, long n6_a, long n6_b, long n6_c, long n6_d,
; long n7_a, long n7_b, long n7_c, long n7_d, long n8_a, long n8_b,
; long n8_c, long n8_d, long n9_a, long n9_b, long n9_c, long n9_d,
; long p1_b, long p1_c, long p1_d, long p2_b, long p2_c, long p2_d,
; long p3_b, long p3_c, long p3_d, long p4_b, long p4_c, long p4_d,
; long p5_b, long p5_c, long p5_d, long p6_b, long p6_c, long p6_d,
; long p7_b, long p7_c, long p7_d, long p8_b, long p8_c, long p8_d,
; long p9_b, long p9_c, long p9_d, float A_1[][p1_b][p1_c][p1_d],
; float A_2[][p2_b][p2_c][p2_d], float A_3[][p3_b][p3_c][p3_d],
; float A_4[][p4_b][p4_c][p4_d], float A_5[][p5_b][p5_c][p5_d],
; float A_6[][p6_b][p6_c][p6_d], float A_7[][p7_b][p7_c][p7_d],
; float A_8[][p8_b][p8_c][p8_d], float A_9[][p9_b][p9_c][p9_d]) {
; for (long i = 0; i < n1_a; i++)
; for (long j = 0; j < n1_b; j++)
; for (long k = 0; k < n1_c; k++)
; for (long l = 0; l < n1_d; l++)
; A_1[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n2_a; i++)
; for (long j = 0; j < n2_b; j++)
; for (long k = 0; k < n2_c; k++)
; for (long l = 0; l < n2_d; l++)
; A_2[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n3_a; i++)
; for (long j = 0; j < n3_b; j++)
; for (long k = 0; k < n3_c; k++)
; for (long l = 0; l < n3_d; l++)
; A_3[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n4_a; i++)
; for (long j = 0; j < n4_b; j++)
; for (long k = 0; k < n4_c; k++)
; for (long l = 0; l < n4_d; l++)
; A_4[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n5_a; i++)
; for (long j = 0; j < n5_b; j++)
; for (long k = 0; k < n5_c; k++)
; for (long l = 0; l < n5_d; l++)
; A_5[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n6_a; i++)
; for (long j = 0; j < n6_b; j++)
; for (long k = 0; k < n6_c; k++)
; for (long l = 0; l < n6_d; l++)
; A_6[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n7_a; i++)
; for (long j = 0; j < n7_b; j++)
; for (long k = 0; k < n7_c; k++)
; for (long l = 0; l < n7_d; l++)
; A_7[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n8_a; i++)
; for (long j = 0; j < n8_b; j++)
; for (long k = 0; k < n8_c; k++)
; for (long l = 0; l < n8_d; l++)
; A_8[i][j][k][l] += i + j + k + l;
; for (long i = 0; i < n9_a; i++)
; for (long j = 0; j < n9_b; j++)
; for (long k = 0; k < n9_c; k++)
; for (long l = 0; l < n9_d; l++)
; A_9[i][j][k][l] += i + j + k + l;
; }
;
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
define void @foo(i64 %n1_a, i64 %n1_b, i64 %n1_c, i64 %n1_d, i64 %n2_a, i64 %n2_b, i64 %n2_c, i64 %n2_d, i64 %n3_a, i64 %n3_b, i64 %n3_c, i64 %n3_d, i64 %n4_a, i64 %n4_b, i64 %n4_c, i64 %n4_d, i64 %n5_a, i64 %n5_b, i64 %n5_c, i64 %n5_d, i64 %n6_a, i64 %n6_b, i64 %n6_c, i64 %n6_d, i64 %n7_a, i64 %n7_b, i64 %n7_c, i64 %n7_d, i64 %n8_a, i64 %n8_b, i64 %n8_c, i64 %n8_d, i64 %n9_a, i64 %n9_b, i64 %n9_c, i64 %n9_d, i64 %p1_b, i64 %p1_c, i64 %p1_d, i64 %p2_b, i64 %p2_c, i64 %p2_d, i64 %p3_b, i64 %p3_c, i64 %p3_d, i64 %p4_b, i64 %p4_c, i64 %p4_d, i64 %p5_b, i64 %p5_c, i64 %p5_d, i64 %p6_b, i64 %p6_c, i64 %p6_d, i64 %p7_b, i64 %p7_c, i64 %p7_d, i64 %p8_b, i64 %p8_c, i64 %p8_d, i64 %p9_b, i64 %p9_c, i64 %p9_d, float* %A_1, float* %A_2, float* %A_3, float* %A_4, float* %A_5, float* %A_6, float* %A_7, float* %A_8, float* %A_9) {
bb:
br label %bb1
bb1: ; preds = %bb37, %bb
%i.0 = phi i64 [ 0, %bb ], [ %tmp38, %bb37 ]
%tmp = icmp slt i64 %i.0, %n1_a
br i1 %tmp, label %bb2, label %bb39
bb2: ; preds = %bb1
br label %bb3
bb3: ; preds = %bb34, %bb2
%j.0 = phi i64 [ 0, %bb2 ], [ %tmp35, %bb34 ]
%tmp4 = icmp slt i64 %j.0, %n1_b
br i1 %tmp4, label %bb5, label %bb36
bb5: ; preds = %bb3
br label %bb6
bb6: ; preds = %bb31, %bb5
%k.0 = phi i64 [ 0, %bb5 ], [ %tmp32, %bb31 ]
%tmp7 = icmp slt i64 %k.0, %n1_c
br i1 %tmp7, label %bb8, label %bb33
bb8: ; preds = %bb6
br label %bb9
bb9: ; preds = %bb28, %bb8
%l.0 = phi i64 [ 0, %bb8 ], [ %tmp29, %bb28 ]
%tmp10 = icmp slt i64 %l.0, %n1_d
br i1 %tmp10, label %bb11, label %bb30
bb11: ; preds = %bb9
%tmp12 = add nuw nsw i64 %i.0, %j.0
%tmp13 = add nsw i64 %tmp12, %k.0
%tmp14 = add nsw i64 %tmp13, %l.0
%tmp15 = sitofp i64 %tmp14 to float
%tmp16 = mul nuw i64 %p1_b, %p1_c
%tmp17 = mul nuw i64 %tmp16, %p1_d
%tmp18 = mul nsw i64 %i.0, %tmp17
%tmp19 = getelementptr inbounds float, float* %A_1, i64 %tmp18
%tmp20 = mul nuw i64 %p1_c, %p1_d
%tmp21 = mul nsw i64 %j.0, %tmp20
%tmp22 = getelementptr inbounds float, float* %tmp19, i64 %tmp21
%tmp23 = mul nsw i64 %k.0, %p1_d
%tmp24 = getelementptr inbounds float, float* %tmp22, i64 %tmp23
%tmp25 = getelementptr inbounds float, float* %tmp24, i64 %l.0
%tmp26 = load float, float* %tmp25, align 4
%tmp27 = fadd float %tmp26, %tmp15
store float %tmp27, float* %tmp25, align 4
br label %bb28
bb28: ; preds = %bb11
%tmp29 = add nuw nsw i64 %l.0, 1
br label %bb9
bb30: ; preds = %bb9
br label %bb31
bb31: ; preds = %bb30
%tmp32 = add nuw nsw i64 %k.0, 1
br label %bb6
bb33: ; preds = %bb6
br label %bb34
bb34: ; preds = %bb33
%tmp35 = add nuw nsw i64 %j.0, 1
br label %bb3
bb36: ; preds = %bb3
br label %bb37
bb37: ; preds = %bb36
%tmp38 = add nuw nsw i64 %i.0, 1
br label %bb1
bb39: ; preds = %bb1
br label %bb40
bb40: ; preds = %bb77, %bb39
%i1.0 = phi i64 [ 0, %bb39 ], [ %tmp78, %bb77 ]
%tmp41 = icmp slt i64 %i1.0, %n2_a
br i1 %tmp41, label %bb42, label %bb79
bb42: ; preds = %bb40
br label %bb43
bb43: ; preds = %bb74, %bb42
%j2.0 = phi i64 [ 0, %bb42 ], [ %tmp75, %bb74 ]
%tmp44 = icmp slt i64 %j2.0, %n2_b
br i1 %tmp44, label %bb45, label %bb76
bb45: ; preds = %bb43
br label %bb46
bb46: ; preds = %bb71, %bb45
%k3.0 = phi i64 [ 0, %bb45 ], [ %tmp72, %bb71 ]
%tmp47 = icmp slt i64 %k3.0, %n2_c
br i1 %tmp47, label %bb48, label %bb73
bb48: ; preds = %bb46
br label %bb49
bb49: ; preds = %bb68, %bb48
%l4.0 = phi i64 [ 0, %bb48 ], [ %tmp69, %bb68 ]
%tmp50 = icmp slt i64 %l4.0, %n2_d
br i1 %tmp50, label %bb51, label %bb70
bb51: ; preds = %bb49
%tmp52 = add nuw nsw i64 %i1.0, %j2.0
%tmp53 = add nsw i64 %tmp52, %k3.0
%tmp54 = add nsw i64 %tmp53, %l4.0
%tmp55 = sitofp i64 %tmp54 to float
%tmp56 = mul nuw i64 %p2_b, %p2_c
%tmp57 = mul nuw i64 %tmp56, %p2_d
%tmp58 = mul nsw i64 %i1.0, %tmp57
%tmp59 = getelementptr inbounds float, float* %A_2, i64 %tmp58
%tmp60 = mul nuw i64 %p2_c, %p2_d
%tmp61 = mul nsw i64 %j2.0, %tmp60
%tmp62 = getelementptr inbounds float, float* %tmp59, i64 %tmp61
%tmp63 = mul nsw i64 %k3.0, %p2_d
%tmp64 = getelementptr inbounds float, float* %tmp62, i64 %tmp63
%tmp65 = getelementptr inbounds float, float* %tmp64, i64 %l4.0
%tmp66 = load float, float* %tmp65, align 4
%tmp67 = fadd float %tmp66, %tmp55
store float %tmp67, float* %tmp65, align 4
br label %bb68
bb68: ; preds = %bb51
%tmp69 = add nuw nsw i64 %l4.0, 1
br label %bb49
bb70: ; preds = %bb49
br label %bb71
bb71: ; preds = %bb70
%tmp72 = add nuw nsw i64 %k3.0, 1
br label %bb46
bb73: ; preds = %bb46
br label %bb74
bb74: ; preds = %bb73
%tmp75 = add nuw nsw i64 %j2.0, 1
br label %bb43
bb76: ; preds = %bb43
br label %bb77
bb77: ; preds = %bb76
%tmp78 = add nuw nsw i64 %i1.0, 1
br label %bb40
bb79: ; preds = %bb40
br label %bb80
bb80: ; preds = %bb117, %bb79
%i5.0 = phi i64 [ 0, %bb79 ], [ %tmp118, %bb117 ]
%tmp81 = icmp slt i64 %i5.0, %n3_a
br i1 %tmp81, label %bb82, label %bb119
bb82: ; preds = %bb80
br label %bb83
bb83: ; preds = %bb114, %bb82
%j6.0 = phi i64 [ 0, %bb82 ], [ %tmp115, %bb114 ]
%tmp84 = icmp slt i64 %j6.0, %n3_b
br i1 %tmp84, label %bb85, label %bb116
bb85: ; preds = %bb83
br label %bb86
bb86: ; preds = %bb111, %bb85
%k7.0 = phi i64 [ 0, %bb85 ], [ %tmp112, %bb111 ]
%tmp87 = icmp slt i64 %k7.0, %n3_c
br i1 %tmp87, label %bb88, label %bb113
bb88: ; preds = %bb86
br label %bb89
bb89: ; preds = %bb108, %bb88
%l8.0 = phi i64 [ 0, %bb88 ], [ %tmp109, %bb108 ]
%tmp90 = icmp slt i64 %l8.0, %n3_d
br i1 %tmp90, label %bb91, label %bb110
bb91: ; preds = %bb89
%tmp92 = add nuw nsw i64 %i5.0, %j6.0
%tmp93 = add nsw i64 %tmp92, %k7.0
%tmp94 = add nsw i64 %tmp93, %l8.0
%tmp95 = sitofp i64 %tmp94 to float
%tmp96 = mul nuw i64 %p3_b, %p3_c
%tmp97 = mul nuw i64 %tmp96, %p3_d
%tmp98 = mul nsw i64 %i5.0, %tmp97
%tmp99 = getelementptr inbounds float, float* %A_3, i64 %tmp98
%tmp100 = mul nuw i64 %p3_c, %p3_d
%tmp101 = mul nsw i64 %j6.0, %tmp100
%tmp102 = getelementptr inbounds float, float* %tmp99, i64 %tmp101
%tmp103 = mul nsw i64 %k7.0, %p3_d
%tmp104 = getelementptr inbounds float, float* %tmp102, i64 %tmp103
%tmp105 = getelementptr inbounds float, float* %tmp104, i64 %l8.0
%tmp106 = load float, float* %tmp105, align 4
%tmp107 = fadd float %tmp106, %tmp95
store float %tmp107, float* %tmp105, align 4
br label %bb108
bb108: ; preds = %bb91
%tmp109 = add nuw nsw i64 %l8.0, 1
br label %bb89
bb110: ; preds = %bb89
br label %bb111
bb111: ; preds = %bb110
%tmp112 = add nuw nsw i64 %k7.0, 1
br label %bb86
bb113: ; preds = %bb86
br label %bb114
bb114: ; preds = %bb113
%tmp115 = add nuw nsw i64 %j6.0, 1
br label %bb83
bb116: ; preds = %bb83
br label %bb117
bb117: ; preds = %bb116
%tmp118 = add nuw nsw i64 %i5.0, 1
br label %bb80
bb119: ; preds = %bb80
br label %bb120
bb120: ; preds = %bb157, %bb119
%i9.0 = phi i64 [ 0, %bb119 ], [ %tmp158, %bb157 ]
%tmp121 = icmp slt i64 %i9.0, %n4_a
br i1 %tmp121, label %bb122, label %bb159
bb122: ; preds = %bb120
br label %bb123
bb123: ; preds = %bb154, %bb122
%j10.0 = phi i64 [ 0, %bb122 ], [ %tmp155, %bb154 ]
%tmp124 = icmp slt i64 %j10.0, %n4_b
br i1 %tmp124, label %bb125, label %bb156
bb125: ; preds = %bb123
br label %bb126
bb126: ; preds = %bb151, %bb125
%k11.0 = phi i64 [ 0, %bb125 ], [ %tmp152, %bb151 ]
%tmp127 = icmp slt i64 %k11.0, %n4_c
br i1 %tmp127, label %bb128, label %bb153
bb128: ; preds = %bb126
br label %bb129
bb129: ; preds = %bb148, %bb128
%l12.0 = phi i64 [ 0, %bb128 ], [ %tmp149, %bb148 ]
%tmp130 = icmp slt i64 %l12.0, %n4_d
br i1 %tmp130, label %bb131, label %bb150
bb131: ; preds = %bb129
%tmp132 = add nuw nsw i64 %i9.0, %j10.0
%tmp133 = add nsw i64 %tmp132, %k11.0
%tmp134 = add nsw i64 %tmp133, %l12.0
%tmp135 = sitofp i64 %tmp134 to float
%tmp136 = mul nuw i64 %p4_b, %p4_c
%tmp137 = mul nuw i64 %tmp136, %p4_d
%tmp138 = mul nsw i64 %i9.0, %tmp137
%tmp139 = getelementptr inbounds float, float* %A_4, i64 %tmp138
%tmp140 = mul nuw i64 %p4_c, %p4_d
%tmp141 = mul nsw i64 %j10.0, %tmp140
%tmp142 = getelementptr inbounds float, float* %tmp139, i64 %tmp141
%tmp143 = mul nsw i64 %k11.0, %p4_d
%tmp144 = getelementptr inbounds float, float* %tmp142, i64 %tmp143
%tmp145 = getelementptr inbounds float, float* %tmp144, i64 %l12.0
%tmp146 = load float, float* %tmp145, align 4
%tmp147 = fadd float %tmp146, %tmp135
store float %tmp147, float* %tmp145, align 4
br label %bb148
bb148: ; preds = %bb131
%tmp149 = add nuw nsw i64 %l12.0, 1
br label %bb129
bb150: ; preds = %bb129
br label %bb151
bb151: ; preds = %bb150
%tmp152 = add nuw nsw i64 %k11.0, 1
br label %bb126
bb153: ; preds = %bb126
br label %bb154
bb154: ; preds = %bb153
%tmp155 = add nuw nsw i64 %j10.0, 1
br label %bb123
bb156: ; preds = %bb123
br label %bb157
bb157: ; preds = %bb156
%tmp158 = add nuw nsw i64 %i9.0, 1
br label %bb120
bb159: ; preds = %bb120
br label %bb160
bb160: ; preds = %bb197, %bb159
%i13.0 = phi i64 [ 0, %bb159 ], [ %tmp198, %bb197 ]
%tmp161 = icmp slt i64 %i13.0, %n5_a
br i1 %tmp161, label %bb162, label %bb199
bb162: ; preds = %bb160
br label %bb163
bb163: ; preds = %bb194, %bb162
%j14.0 = phi i64 [ 0, %bb162 ], [ %tmp195, %bb194 ]
%tmp164 = icmp slt i64 %j14.0, %n5_b
br i1 %tmp164, label %bb165, label %bb196
bb165: ; preds = %bb163
br label %bb166
bb166: ; preds = %bb191, %bb165
%k15.0 = phi i64 [ 0, %bb165 ], [ %tmp192, %bb191 ]
%tmp167 = icmp slt i64 %k15.0, %n5_c
br i1 %tmp167, label %bb168, label %bb193
bb168: ; preds = %bb166
br label %bb169
bb169: ; preds = %bb188, %bb168
%l16.0 = phi i64 [ 0, %bb168 ], [ %tmp189, %bb188 ]
%tmp170 = icmp slt i64 %l16.0, %n5_d
br i1 %tmp170, label %bb171, label %bb190
bb171: ; preds = %bb169
%tmp172 = add nuw nsw i64 %i13.0, %j14.0
%tmp173 = add nsw i64 %tmp172, %k15.0
%tmp174 = add nsw i64 %tmp173, %l16.0
%tmp175 = sitofp i64 %tmp174 to float
%tmp176 = mul nuw i64 %p5_b, %p5_c
%tmp177 = mul nuw i64 %tmp176, %p5_d
%tmp178 = mul nsw i64 %i13.0, %tmp177
%tmp179 = getelementptr inbounds float, float* %A_5, i64 %tmp178
%tmp180 = mul nuw i64 %p5_c, %p5_d
%tmp181 = mul nsw i64 %j14.0, %tmp180
%tmp182 = getelementptr inbounds float, float* %tmp179, i64 %tmp181
%tmp183 = mul nsw i64 %k15.0, %p5_d
%tmp184 = getelementptr inbounds float, float* %tmp182, i64 %tmp183
%tmp185 = getelementptr inbounds float, float* %tmp184, i64 %l16.0
%tmp186 = load float, float* %tmp185, align 4
%tmp187 = fadd float %tmp186, %tmp175
store float %tmp187, float* %tmp185, align 4
br label %bb188
bb188: ; preds = %bb171
%tmp189 = add nuw nsw i64 %l16.0, 1
br label %bb169
bb190: ; preds = %bb169
br label %bb191
bb191: ; preds = %bb190
%tmp192 = add nuw nsw i64 %k15.0, 1
br label %bb166
bb193: ; preds = %bb166
br label %bb194
bb194: ; preds = %bb193
%tmp195 = add nuw nsw i64 %j14.0, 1
br label %bb163
bb196: ; preds = %bb163
br label %bb197
bb197: ; preds = %bb196
%tmp198 = add nuw nsw i64 %i13.0, 1
br label %bb160
bb199: ; preds = %bb160
br label %bb200
bb200: ; preds = %bb237, %bb199
%i17.0 = phi i64 [ 0, %bb199 ], [ %tmp238, %bb237 ]
%tmp201 = icmp slt i64 %i17.0, %n6_a
br i1 %tmp201, label %bb202, label %bb239
bb202: ; preds = %bb200
br label %bb203
bb203: ; preds = %bb234, %bb202
%j18.0 = phi i64 [ 0, %bb202 ], [ %tmp235, %bb234 ]
%tmp204 = icmp slt i64 %j18.0, %n6_b
br i1 %tmp204, label %bb205, label %bb236
bb205: ; preds = %bb203
br label %bb206
bb206: ; preds = %bb231, %bb205
%k19.0 = phi i64 [ 0, %bb205 ], [ %tmp232, %bb231 ]
%tmp207 = icmp slt i64 %k19.0, %n6_c
br i1 %tmp207, label %bb208, label %bb233
bb208: ; preds = %bb206
br label %bb209
bb209: ; preds = %bb228, %bb208
%l20.0 = phi i64 [ 0, %bb208 ], [ %tmp229, %bb228 ]
%tmp210 = icmp slt i64 %l20.0, %n6_d
br i1 %tmp210, label %bb211, label %bb230
bb211: ; preds = %bb209
%tmp212 = add nuw nsw i64 %i17.0, %j18.0
%tmp213 = add nsw i64 %tmp212, %k19.0
%tmp214 = add nsw i64 %tmp213, %l20.0
%tmp215 = sitofp i64 %tmp214 to float
%tmp216 = mul nuw i64 %p6_b, %p6_c
%tmp217 = mul nuw i64 %tmp216, %p6_d
%tmp218 = mul nsw i64 %i17.0, %tmp217
%tmp219 = getelementptr inbounds float, float* %A_6, i64 %tmp218
%tmp220 = mul nuw i64 %p6_c, %p6_d
%tmp221 = mul nsw i64 %j18.0, %tmp220
%tmp222 = getelementptr inbounds float, float* %tmp219, i64 %tmp221
%tmp223 = mul nsw i64 %k19.0, %p6_d
%tmp224 = getelementptr inbounds float, float* %tmp222, i64 %tmp223
%tmp225 = getelementptr inbounds float, float* %tmp224, i64 %l20.0
%tmp226 = load float, float* %tmp225, align 4
%tmp227 = fadd float %tmp226, %tmp215
store float %tmp227, float* %tmp225, align 4
br label %bb228
bb228: ; preds = %bb211
%tmp229 = add nuw nsw i64 %l20.0, 1
br label %bb209
bb230: ; preds = %bb209
br label %bb231
bb231: ; preds = %bb230
%tmp232 = add nuw nsw i64 %k19.0, 1
br label %bb206
bb233: ; preds = %bb206
br label %bb234
bb234: ; preds = %bb233
%tmp235 = add nuw nsw i64 %j18.0, 1
br label %bb203
bb236: ; preds = %bb203
br label %bb237
bb237: ; preds = %bb236
%tmp238 = add nuw nsw i64 %i17.0, 1
br label %bb200
bb239: ; preds = %bb200
br label %bb240
bb240: ; preds = %bb277, %bb239
%i21.0 = phi i64 [ 0, %bb239 ], [ %tmp278, %bb277 ]
%tmp241 = icmp slt i64 %i21.0, %n7_a
br i1 %tmp241, label %bb242, label %bb279
bb242: ; preds = %bb240
br label %bb243
bb243: ; preds = %bb274, %bb242
%j22.0 = phi i64 [ 0, %bb242 ], [ %tmp275, %bb274 ]
%tmp244 = icmp slt i64 %j22.0, %n7_b
br i1 %tmp244, label %bb245, label %bb276
bb245: ; preds = %bb243
br label %bb246
bb246: ; preds = %bb271, %bb245
%k23.0 = phi i64 [ 0, %bb245 ], [ %tmp272, %bb271 ]
%tmp247 = icmp slt i64 %k23.0, %n7_c
br i1 %tmp247, label %bb248, label %bb273
bb248: ; preds = %bb246
br label %bb249
bb249: ; preds = %bb268, %bb248
%l24.0 = phi i64 [ 0, %bb248 ], [ %tmp269, %bb268 ]
%tmp250 = icmp slt i64 %l24.0, %n7_d
br i1 %tmp250, label %bb251, label %bb270
bb251: ; preds = %bb249
%tmp252 = add nuw nsw i64 %i21.0, %j22.0
%tmp253 = add nsw i64 %tmp252, %k23.0
%tmp254 = add nsw i64 %tmp253, %l24.0
%tmp255 = sitofp i64 %tmp254 to float
%tmp256 = mul nuw i64 %p7_b, %p7_c
%tmp257 = mul nuw i64 %tmp256, %p7_d
%tmp258 = mul nsw i64 %i21.0, %tmp257
%tmp259 = getelementptr inbounds float, float* %A_7, i64 %tmp258
%tmp260 = mul nuw i64 %p7_c, %p7_d
%tmp261 = mul nsw i64 %j22.0, %tmp260
%tmp262 = getelementptr inbounds float, float* %tmp259, i64 %tmp261
%tmp263 = mul nsw i64 %k23.0, %p7_d
%tmp264 = getelementptr inbounds float, float* %tmp262, i64 %tmp263
%tmp265 = getelementptr inbounds float, float* %tmp264, i64 %l24.0
%tmp266 = load float, float* %tmp265, align 4
%tmp267 = fadd float %tmp266, %tmp255
store float %tmp267, float* %tmp265, align 4
br label %bb268
bb268: ; preds = %bb251
%tmp269 = add nuw nsw i64 %l24.0, 1
br label %bb249
bb270: ; preds = %bb249
br label %bb271
bb271: ; preds = %bb270
%tmp272 = add nuw nsw i64 %k23.0, 1
br label %bb246
bb273: ; preds = %bb246
br label %bb274
bb274: ; preds = %bb273
%tmp275 = add nuw nsw i64 %j22.0, 1
br label %bb243
bb276: ; preds = %bb243
br label %bb277
bb277: ; preds = %bb276
%tmp278 = add nuw nsw i64 %i21.0, 1
br label %bb240
bb279: ; preds = %bb240
br label %bb280
bb280: ; preds = %bb317, %bb279
%i25.0 = phi i64 [ 0, %bb279 ], [ %tmp318, %bb317 ]
%tmp281 = icmp slt i64 %i25.0, %n8_a
br i1 %tmp281, label %bb282, label %bb319
bb282: ; preds = %bb280
br label %bb283
bb283: ; preds = %bb314, %bb282
%j26.0 = phi i64 [ 0, %bb282 ], [ %tmp315, %bb314 ]
%tmp284 = icmp slt i64 %j26.0, %n8_b
br i1 %tmp284, label %bb285, label %bb316
bb285: ; preds = %bb283
br label %bb286
bb286: ; preds = %bb311, %bb285
%k27.0 = phi i64 [ 0, %bb285 ], [ %tmp312, %bb311 ]
%tmp287 = icmp slt i64 %k27.0, %n8_c
br i1 %tmp287, label %bb288, label %bb313
bb288: ; preds = %bb286
br label %bb289
bb289: ; preds = %bb308, %bb288
%l28.0 = phi i64 [ 0, %bb288 ], [ %tmp309, %bb308 ]
%tmp290 = icmp slt i64 %l28.0, %n8_d
br i1 %tmp290, label %bb291, label %bb310
bb291: ; preds = %bb289
%tmp292 = add nuw nsw i64 %i25.0, %j26.0
%tmp293 = add nsw i64 %tmp292, %k27.0
%tmp294 = add nsw i64 %tmp293, %l28.0
%tmp295 = sitofp i64 %tmp294 to float
%tmp296 = mul nuw i64 %p8_b, %p8_c
%tmp297 = mul nuw i64 %tmp296, %p8_d
%tmp298 = mul nsw i64 %i25.0, %tmp297
%tmp299 = getelementptr inbounds float, float* %A_8, i64 %tmp298
%tmp300 = mul nuw i64 %p8_c, %p8_d
%tmp301 = mul nsw i64 %j26.0, %tmp300
%tmp302 = getelementptr inbounds float, float* %tmp299, i64 %tmp301
%tmp303 = mul nsw i64 %k27.0, %p8_d
%tmp304 = getelementptr inbounds float, float* %tmp302, i64 %tmp303
%tmp305 = getelementptr inbounds float, float* %tmp304, i64 %l28.0
%tmp306 = load float, float* %tmp305, align 4
%tmp307 = fadd float %tmp306, %tmp295
store float %tmp307, float* %tmp305, align 4
br label %bb308
bb308: ; preds = %bb291
%tmp309 = add nuw nsw i64 %l28.0, 1
br label %bb289
bb310: ; preds = %bb289
br label %bb311
bb311: ; preds = %bb310
%tmp312 = add nuw nsw i64 %k27.0, 1
br label %bb286
bb313: ; preds = %bb286
br label %bb314
bb314: ; preds = %bb313
%tmp315 = add nuw nsw i64 %j26.0, 1
br label %bb283
bb316: ; preds = %bb283
br label %bb317
bb317: ; preds = %bb316
%tmp318 = add nuw nsw i64 %i25.0, 1
br label %bb280
bb319: ; preds = %bb280
br label %bb320
bb320: ; preds = %bb357, %bb319
%i29.0 = phi i64 [ 0, %bb319 ], [ %tmp358, %bb357 ]
%tmp321 = icmp slt i64 %i29.0, %n9_a
br i1 %tmp321, label %bb322, label %bb359
bb322: ; preds = %bb320
br label %bb323
bb323: ; preds = %bb354, %bb322
%j30.0 = phi i64 [ 0, %bb322 ], [ %tmp355, %bb354 ]
%tmp324 = icmp slt i64 %j30.0, %n9_b
br i1 %tmp324, label %bb325, label %bb356
bb325: ; preds = %bb323
br label %bb326
bb326: ; preds = %bb351, %bb325
%k31.0 = phi i64 [ 0, %bb325 ], [ %tmp352, %bb351 ]
%tmp327 = icmp slt i64 %k31.0, %n9_c
br i1 %tmp327, label %bb328, label %bb353
bb328: ; preds = %bb326
br label %bb329
bb329: ; preds = %bb348, %bb328
%l32.0 = phi i64 [ 0, %bb328 ], [ %tmp349, %bb348 ]
%tmp330 = icmp slt i64 %l32.0, %n9_d
br i1 %tmp330, label %bb331, label %bb350
bb331: ; preds = %bb329
%tmp332 = add nuw nsw i64 %i29.0, %j30.0
%tmp333 = add nsw i64 %tmp332, %k31.0
%tmp334 = add nsw i64 %tmp333, %l32.0
%tmp335 = sitofp i64 %tmp334 to float
%tmp336 = mul nuw i64 %p9_b, %p9_c
%tmp337 = mul nuw i64 %tmp336, %p9_d
%tmp338 = mul nsw i64 %i29.0, %tmp337
%tmp339 = getelementptr inbounds float, float* %A_9, i64 %tmp338
%tmp340 = mul nuw i64 %p9_c, %p9_d
%tmp341 = mul nsw i64 %j30.0, %tmp340
%tmp342 = getelementptr inbounds float, float* %tmp339, i64 %tmp341
%tmp343 = mul nsw i64 %k31.0, %p9_d
%tmp344 = getelementptr inbounds float, float* %tmp342, i64 %tmp343
%tmp345 = getelementptr inbounds float, float* %tmp344, i64 %l32.0
%tmp346 = load float, float* %tmp345, align 4
%tmp347 = fadd float %tmp346, %tmp335
store float %tmp347, float* %tmp345, align 4
br label %bb348
bb348: ; preds = %bb331
%tmp349 = add nuw nsw i64 %l32.0, 1
br label %bb329
bb350: ; preds = %bb329
br label %bb351
bb351: ; preds = %bb350
%tmp352 = add nuw nsw i64 %k31.0, 1
br label %bb326
bb353: ; preds = %bb326
br label %bb354
bb354: ; preds = %bb353
%tmp355 = add nuw nsw i64 %j30.0, 1
br label %bb323
bb356: ; preds = %bb323
br label %bb357
bb357: ; preds = %bb356
%tmp358 = add nuw nsw i64 %i29.0, 1
br label %bb320
bb359: ; preds = %bb320
ret void
}

2
polly/test/ScopInfo/remarks.ll
View File

@ -2,7 +2,7 @@
; RUN: -polly-invariant-load-hoisting=true -disable-output < %s 2>&1 | FileCheck %s
;
; CHECK: remark: test/ScopInfo/remarks.c:4:7: SCoP begins here.
; CHECK: remark: test/ScopInfo/remarks.c:9:15: Inbounds assumption: [N, M, Debug] -> { : N <= 0 or (N > 0 and M <= 100) }
; CHECK: remark: test/ScopInfo/remarks.c:9:15: Inbounds assumption: [N, M, Debug] -> { : M <= 100 }
; CHECK: remark: test/ScopInfo/remarks.c:13:7: No-error restriction: [N, M, Debug] -> { : N > 0 and M >= 0 and (Debug < 0 or Debug > 0) }
; CHECK: remark: test/ScopInfo/remarks.c:8:5: Finite loop restriction: [N, M, Debug] -> { : N > 0 and (M <= -2 or M = -1) }
; CHECK: remark: test/ScopInfo/remarks.c:4:7: No-overflows restriction: [N, M, Debug] -> { : M <= -2147483649 - N or M >= 2147483648 - N }

3
polly/test/ScopInfo/user_provided_non_dominating_assumptions.ll
View File

@ -1,4 +1,5 @@
; RUN: opt %loadPolly -pass-remarks-analysis="polly-scops" -polly-scops -disable-output < %s 2>&1 | FileCheck %s
; RUN: opt %loadPolly -pass-remarks-analysis="polly-scops" -polly-scops \
; RUN: -polly-precise-inbounds -disable-output < %s 2>&1 | FileCheck %s
;
; CHECK: remark: <unknown>:0:0: SCoP begins here.
; CHECK-NEXT: remark: <unknown>:0:0: Use user assumption: [i, N, M] -> { : N <= i or (N > i and N >= 0) }