Hello!
Attached patch fixes a typo in ix86_expand_set_or_movmem, where a
wrong define was used in a condition. The patch also adds additional
condition (as proposed by H.J.) as a correctness improvement (although
patched gcc bootstraps and regression tests OK without).
I took the liberty to rename a couple of (long) constants and to fix
some typos while there.
2013-11-11 Uros Bizjak <ubiz...@gmail.com>
H.J. Lu <hongjiu...@intel.com>
PR target/58853
* config/i386/x86-tune.def
(X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES): Rename from
TARGET_MISALIGNED_MOVE_STRING_PROLOGUES.
* config/i386/i386.h
(TARGET_MISALIGNED_MOVE_STRING_PRO_EPILOGUES): Rename from
TARGET_MISALIGNED_MOVE_STRING_PROLOGUES_EPILOGUES. Update for renamed
X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES.
* config/i386/i386.c (ix86_expand_set_or_movmem): Use
TARGET_MISALIGNED_MOVE_STRING_PRO_EPILOGUES to calculate
misaligned_prologue_used. Check that
desired_aling <= epilogue_size_needed.
testsuite/ChangeLog:
2013-11-11 Uros Bizjak <ubiz...@gmail.com>
PR target/58853
* gcc.target/i386/pr58853.c: New test.
Patch was approved by Honza in the PR.
Patch was bootstrapped and regression tested on x86_64-pc-linux-gnu
{,-m32} and committed to mainline SVN.
Uros.
Index: config/i386/i386.c
===================================================================
--- config/i386/i386.c (revision 204676)
+++ config/i386/i386.c (working copy)
@@ -23761,13 +23761,15 @@ ix86_expand_set_or_movmem (rtx dst, rtx src, rtx c
}
gcc_assert (desired_align >= 1 && align >= 1);
- /* Misaligned move sequences handles both prologues and epilogues at once.
- Default code generation results in smaller code for large alignments and
- also avoids redundant job when sizes are known precisely. */
- misaligned_prologue_used = (TARGET_MISALIGNED_MOVE_STRING_PROLOGUES
- && MAX (desired_align, epilogue_size_needed) <= 32
- && ((desired_align > align && !align_bytes)
- || (!count && epilogue_size_needed > 1)));
+ /* Misaligned move sequences handle both prologue and epilogue at once.
+ Default code generation results in a smaller code for large alignments
+ and also avoids redundant job when sizes are known precisely. */
+ misaligned_prologue_used
+ = (TARGET_MISALIGNED_MOVE_STRING_PRO_EPILOGUES
+ && MAX (desired_align, epilogue_size_needed) <= 32
+ && desired_align <= epilogue_size_needed
+ && ((desired_align > align && !align_bytes)
+ || (!count && epilogue_size_needed > 1)));
/* Do the cheap promotion to allow better CSE across the
main loop and epilogue (ie one load of the big constant in the
Index: config/i386/i386.h
===================================================================
--- config/i386/i386.h (revision 204676)
+++ config/i386/i386.h (working copy)
@@ -353,8 +353,8 @@ extern unsigned char ix86_tune_features[X86_TUNE_L
#define TARGET_PROMOTE_QImode ix86_tune_features[X86_TUNE_PROMOTE_QIMODE]
#define TARGET_FAST_PREFIX ix86_tune_features[X86_TUNE_FAST_PREFIX]
#define TARGET_SINGLE_STRINGOP ix86_tune_features[X86_TUNE_SINGLE_STRINGOP]
-#define TARGET_MISALIGNED_MOVE_STRING_PROLOGUES_EPILOGUES \
- ix86_tune_features[TARGET_MISALIGNED_MOVE_STRING_PROLOGUES]
+#define TARGET_MISALIGNED_MOVE_STRING_PRO_EPILOGUES \
+ ix86_tune_features[X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES]
#define TARGET_QIMODE_MATH ix86_tune_features[X86_TUNE_QIMODE_MATH]
#define TARGET_HIMODE_MATH ix86_tune_features[X86_TUNE_HIMODE_MATH]
#define TARGET_PROMOTE_QI_REGS ix86_tune_features[X86_TUNE_PROMOTE_QI_REGS]
Index: config/i386/x86-tune.def
===================================================================
--- config/i386/x86-tune.def (revision 204676)
+++ config/i386/x86-tune.def (working copy)
@@ -257,13 +257,13 @@ DEF_TUNE (X86_TUNE_AVOID_MEM_OPND_FOR_CMOVE, "avoi
as MOVS and STOS (without a REP prefix) to move/set sequences of bytes. */
DEF_TUNE (X86_TUNE_SINGLE_STRINGOP, "single_stringop", m_386 | m_P4_NOCONA)
-/* TARGET_MISALIGNED_MOVE_STRING_PROLOGUES: Enable generation of compace
- prologues and epilogues by issuing a misaligned moves. This require
- target to handle misaligned moves and partial memory stalls resonably
- well.
- FIXME: This actualy may be a win on more targets than listed here. */
-DEF_TUNE (TARGET_MISALIGNED_MOVE_STRING_PROLOGUES,
- "misaligned_move_string_prologues",
+/* X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES: Enable generation of
+ compact prologues and epilogues by issuing a misaligned moves. This
+ requires target to handle misaligned moves and partial memory stalls
+ reasonably well.
+ FIXME: This may actualy be a win on more targets than listed here. */
+DEF_TUNE (X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES,
+ "misaligned_move_string_pro_epilogues",
m_386 | m_486 | m_CORE_ALL | m_AMD_MULTIPLE | m_GENERIC)
/* X86_TUNE_USE_SAHF: Controls use of SAHF. */
Index: testsuite/gcc.target/i386/pr58853.c
===================================================================
--- testsuite/gcc.target/i386/pr58853.c (revision 0)
+++ testsuite/gcc.target/i386/pr58853.c (working copy)
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-minline-all-stringops" } */
+/* { dg-additional-options "-mtune=pentiumpro" { target { ia32 } } } */
+
+void
+my_memcpy (char *dest, const char *src, int n)
+{
+ __builtin_memcpy (dest, src, n);
+}
