<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - delinearization fails: casting issue?"
href="https://bugs.llvm.org/show_bug.cgi?id=33104">33104</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>delinearization fails: casting issue?
</td>
</tr>
<tr>
<th>Product</th>
<td>Polly
</td>
</tr>
<tr>
<th>Version</th>
<td>unspecified
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Linux
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Other
</td>
</tr>
<tr>
<th>Assignee</th>
<td>polly-dev@googlegroups.com
</td>
</tr>
<tr>
<th>Reporter</th>
<td>cychan@lbl.gov
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>The delinearization analysis fails on the sample codes (Listings 4 and 5)
presented in the Grosser IMPACT2015 paper due to possible casting issues.
The following code (Listing 4) was tested with "clang test.c -O3 -mllvm -polly
-c -mllvm -debug-only=polly-scops -mllvm -polly-process-unprofitable":
#include <stddef.h>
struct TwoDArray {
size_t size0; size_t size1; float *Base;
};
#define ACCESS_2D(A, x, y) \
*(A->Base + (x) * A->size1 + (y))
#define SIZE0_2D(A) A->size0
#define SIZE1_2D(A) A->size1
void gemm(struct TwoDArray *A, struct TwoDArray *B,
struct TwoDArray *C) {
for (int i = 0; i < SIZE0_2D(C); i++)
for (int j = 0; j < SIZE1_2D(C); j++)
for (int k = 0; k < SIZE1_2D(A); ++k)
ACCESS_2D(C, i, j) += ACCESS_2D(A, i, k) * ACCESS_2D(B, k, j);
}
There was no output produced. I tried running "opt -debug -basicaa
-delinearize -polly-process-unprofitable -polly-scops -analyze temp.s.ll" on
the canonicalized IR, but i was unable to decipher the output.
The following code (Listing 5) was tested with "clang test.c -O3 -mllvm -polly
-c -mllvm -debug-only=polly-scops -mllvm -polly-process-unprofitable":
void set_subarray(unsigned n1, unsigned n2,
unsigned o0, unsigned o1, unsigned o2,
unsigned s0, unsigned s1, unsigned s2,
float A[]) {
for (unsigned i = 0; i < s0; i++)
for (unsigned j = 0; j < s1; j++)
for (unsigned k = 0; k < s2; k++)
A[(n2 * (n1 * o0 + o1) + o2) + n1 * n2 * i + n2 * j + k] = 1;
}
Which produces the following output:
Function: set_subarray
Region: %for.cond4.preheader---%for.inc15
Max Loop Depth: 1
Invariant Accesses: {
}
Context:
[s2, p_1, p_2, p_3, o2] -> { : -2147483648 <= s2 <= 2147483647 and
-2147483648 <= p_1 <= 2147483647 and -2147483648 <= p_2 <= 2147483647 and
-2147483648 <= p_3 <= 2147483647 and -2147483648 <= o2 <= 2147483647 }
Assumed Context:
[s2, p_1, p_2, p_3, o2] -> { : }
Invalid Context:
[s2, p_1, p_2, p_3, o2] -> { : s2 > 0 and (o2 <= -2147483649 - p_2 - p_3
or o2 <= -2147483649 - p_3 or o2 >= 2147483649 - s2 - p_1 - p_2 - p_3 or o2 >=
2147483648 - p_3 or o2 >= 2147483648 - p_2 - p_3 or o2 < -p_1 - p_2 - p_3) }
p0: %s2
p1: {0,+,%n2}<%for.cond4.preheader>
p2: {0,+,(%n1 * %n2)}<%for.cond1.preheader>
p3: (((%n1 * %o0) + %o1) * %n2)
p4: %o2
Arrays {
i32 MemRef_j_08; // Element size 4
float MemRef_A[*]; // Element size 4
}
Arrays (Bounds as pw_affs) {
i32 MemRef_j_08; // Element size 4
float MemRef_A[*]; // Element size 4
}
Alias Groups (0):
n/a
Statements {
Stmt_for_cond4_preheader
Domain :=
[s2, p_1, p_2, p_3, o2] -> { Stmt_for_cond4_preheader[] };
Schedule :=
[s2, p_1, p_2, p_3, o2] -> { Stmt_for_cond4_preheader[] -> [0,
0] };
MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]
[s2, p_1, p_2, p_3, o2] -> { Stmt_for_cond4_preheader[] ->
MemRef_j_08[] };
Stmt_for_body6
Domain :=
[s2, p_1, p_2, p_3, o2] -> { Stmt_for_body6[i0] : 0 <= i0 < s2
};
Schedule :=
[s2, p_1, p_2, p_3, o2] -> { Stmt_for_body6[i0] -> [1, i0] };
MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
[s2, p_1, p_2, p_3, o2] -> { Stmt_for_body6[i0] -> MemRef_A[p_1
+ p_2 + p_3 + o2 + i0] };
}
Changing unsigned to long appears to correct the issue:
Function: set_subarray
Region: %entry.split---%for.cond.cleanup
Max Loop Depth: 3
Invariant Accesses: {
}
Context:
[s0, s1, s2, n1, n2, o0, o1, o2] -> { : -9223372036854775808 <= s0 <=
9223372036854775807 and -9223372036854775808 <= s1 <= 9223372036854775807 and
-9223372036854775808 <= s2 <= 9223372036854775807 and -9223372036854775808 <=
n1 <= 9223372036854775807 and -9223372036854775808 <= n2 <= 9223372036854775807
and -9223372036854775808 <= o0 <= 9223372036854775807 and -9223372036854775808
<= o1 <= 9223372036854775807 and -9223372036854775808 <= o2 <=
9223372036854775807 }
Assumed Context:
[s0, s1, s2, n1, n2, o0, o1, o2] -> { : -n1 <= o1 <= 1 - s1 + n1 and ((o1
<span class="quote">> -n1 and -n2 <= o2 <= -s2) or (o1 <= -s1 + n1 and 0 <= o2 <= -s2 + n2) or (-n1</span >
< o1 <= -s1 + n1 and o2 >= -n2 and -s2 < o2 <= -s2 + n2 and o2 < 0)) }
Invalid Context:
[s0, s1, s2, n1, n2, o0, o1, o2] -> { : (s1 > 0 and s2 > 0 and o0 >=
9223372036854775809 - s0) or (s0 > 0 and s2 > 0 and o1 >= 9223372036854775809 -
s1) or (s0 > 0 and s1 > 0 and o2 >= 9223372036854775809 - s2) }
p0: %s0
p1: %s1
p2: %s2
p3: %n1
p4: %n2
p5: %o0
p6: %o1
p7: %o2
Arrays {
float MemRef_A[*][%n1][%n2]; // Element size 4
}
Arrays (Bounds as pw_affs) {
float MemRef_A[*][ [n1] -> { [] -> [(n1)] } ][ [n2] -> { [] -> [(n2)] }
]; // Element size 4
}
Alias Groups (0):
n/a
Statements {
Stmt_for_body8
Domain :=
[s0, s1, s2, n1, n2, o0, o1, o2] -> { Stmt_for_body8[i0, i1,
i2] : 0 <= i0 < s0 and 0 <= i1 < s1 and 0 <= i2 < s2 };
Schedule :=
[s0, s1, s2, n1, n2, o0, o1, o2] -> { Stmt_for_body8[i0, i1,
i2] -> [i0, i1, i2] };
MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
[s0, s1, s2, n1, n2, o0, o1, o2] -> { Stmt_for_body8[i0, i1,
i2] -> MemRef_A[-1 + o0 + i0, -1 + n1 + o1 + i1, n2 + o2 + i2] : i1 <= -o1 and
i2 < -o2; Stmt_for_body8[i0, i1, i2] -> MemRef_A[o0 + i0, -1 + o1 + i1, n2 + o2
+ i2] : i1 > -o1 and i2 < -o2; Stmt_for_body8[i0, i1, i2] -> MemRef_A[-1 + o0 +
i0, n1 + o1 + i1, o2 + i2] : i1 < -o1 and i2 >= -o2; Stmt_for_body8[i0, i1, i2]
-> MemRef_A[o0 + i0, o1 + i1, o2 + i2] : i1 >= -o1 and i2 >= -o2 };
}</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>