mirror/gcc
2
0
Fork 0
mirror of git://gcc.gnu.org/git/gcc.git synced 2025-02-11 15:10:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

re PR target/22432 (Wrong code generation using MMX intrinsics on amd64)

Browse Source 
./:	PR target/22432
	* combine.c (apply_distributive_law): Don't distribute across a
	vector mode subreg.
testsuite/:
	PR target/22432
	* gcc.target/i386/pr22432.c: New test.

From-SVN: r106555
This commit is contained in:
Ian Lance Taylor 2005-11-06 05:34:38 +00:00 committed by Ian Lance Taylor
parent d4a43a05fa
commit 1f2a3c8f5e
4 changed files with 41 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
gcc/ChangeLog
View File

@ -1,3 +1,9 @@
2005-11-05 Ian Lance Taylor <ian@airs.com>
PR target/22432
* combine.c (apply_distributive_law): Don't distribute across a
vector mode subreg.
2005-11-05 Kazu Hirata <kazu@codesourcery.com>
* c-typeck.c, config/i386/netware.h, config/m32c/cond.md,

18
gcc/combine.c
View File

@ -8117,14 +8117,15 @@ apply_distributive_law (rtx x)
break;
case SUBREG:
/* Non-paradoxical SUBREGs distributes over all operations, provided
the inner modes and byte offsets are the same, this is an extraction
of a low-order part, we don't convert an fp operation to int or
vice versa, and we would not be converting a single-word
operation into a multi-word operation. The latter test is not
required, but it prevents generating unneeded multi-word operations.
Some of the previous tests are redundant given the latter test, but
are retained because they are required for correctness.
/* Non-paradoxical SUBREGs distributes over all operations,
provided the inner modes and byte offsets are the same, this
is an extraction of a low-order part, we don't convert an fp
operation to int or vice versa, this is not a vector mode,
and we would not be converting a single-word operation into a
multi-word operation. The latter test is not required, but
it prevents generating unneeded multi-word operations. Some
of the previous tests are redundant given the latter test,
but are retained because they are required for correctness.
We produce the result slightly differently in this case. */
@ -8135,6 +8136,7 @@ apply_distributive_law (rtx x)
!= GET_MODE_CLASS (GET_MODE (SUBREG_REG (lhs))))
|| (GET_MODE_SIZE (GET_MODE (lhs))
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (lhs))))
|| VECTOR_MODE_P (GET_MODE (lhs))
|| GET_MODE_SIZE (GET_MODE (SUBREG_REG (lhs))) > UNITS_PER_WORD)
return x;

5
gcc/testsuite/ChangeLog
View File

@ -1,3 +1,8 @@
2005-11-05 Ian Lance Taylor <ian@airs.com>
PR target/22432
* gcc.target/i386/pr22432.c: New test.
2005-11-05 Richard Henderson <rth@redhat.com>
* gcc.target/alpha/asm-1.c: Move from gcc.dg/asm-5.c.

20
gcc/testsuite/gcc.target/i386/pr22432.c Normal file
View File

@ -0,0 +1,20 @@
/* { dg-do compile } */
/* { dg-options "-O2 -mmmx" } */
/* { dg-final { scan-assembler-not "paddb" } } */
typedef int v2si __attribute__ ((__vector_size__ (8)));
typedef short v4hi __attribute__ ((__vector_size__ (8)));
typedef char v8qi __attribute__ ((__vector_size__ (8)));
int
foo (unsigned int *a, unsigned int *b)
{
long long i, j, k;
i = (long long) __builtin_ia32_vec_init_v2si (*a, 0);
j = (long long) __builtin_ia32_vec_init_v2si (*b, 0);
i = (long long) __builtin_ia32_punpcklbw ((v8qi) i, (v8qi) 0ll);
j = (long long) __builtin_ia32_punpcklbw ((v8qi) j, (v8qi) 0ll);
k = (long long) __builtin_ia32_paddw ((v4hi) i, (v4hi) j);
return __builtin_ia32_vec_ext_v2si ((v2si) k, 0);
}