* expr.c (MOVE_BY_PIECES_P, CLEAR_BY_PIECES_P, SET_BY_PIECES_P): Pass speed operand.
* expr.h (MOVE_RATIO, CLEAR_RATIO, SET_RATIO): Update.
* gimplify.c (gimplify_init_constructor): Add speed operand.
* tree-sra.c (decide_block_copy): Likewise.
* tree-inline.c (estimate_move_cost): Likewise.
* config/alpha/alpha.h (MOVE_RATIO): Update.
* config/frv/frv.c (MOVE_RATIO): Update.
* config/spu/spu.h (MOVE_RATIO): Update.
* config/sparc/sparc.h (MOVE_RATIO): Update.
* config/i386/i386.h (MOVE_RATIO, CLEAR_RATIO): Update.
* config/m68hc11/m68hc11.h (MOVE_RATIO): Update.
* config/cris/cris.h (MOVE_RATIO): Update.
* config/mn10300/mn10300.h (MOVE_RATIO): Update.
* config/arm/arm.h (MOVE_RATIO): Update.
* config/pa/pa.md: Update uses of MOVE_RATIO
* config/pa/pa.h (MOVE_RATIO): Update.
* config/mips/mips.h (MOVE_RATIO, MOVE_BY_PIECES, CLEAR_RATIO, SET_RATIO): Update.
* config/h8300/h8300.h (MOVE_RATIO): Update.
* config/v850/v850.h (MOVE_RATIO): Update.
* config/bfin/bfin.h (MOVE_RATIO): Update.
From-SVN: r139855
+2008-00-01 Jan Hubicka <jh@suse.cz>
+
+ * expr.c (MOVE_BY_PIECES_P, CLEAR_BY_PIECES_P, SET_BY_PIECES_P): Pass speed operand.
+ * expr.h (MOVE_RATIO, CLEAR_RATIO, SET_RATIO): Update.
+ * gimplify.c (gimplify_init_constructor): Add speed operand.
+ * tree-sra.c (decide_block_copy): Likewise.
+ * tree-inline.c (estimate_move_cost): Likewise.
+ * config/alpha/alpha.h (MOVE_RATIO): Update.
+ * config/frv/frv.c (MOVE_RATIO): Update.
+ * config/spu/spu.h (MOVE_RATIO): Update.
+ * config/sparc/sparc.h (MOVE_RATIO): Update.
+ * config/i386/i386.h (MOVE_RATIO, CLEAR_RATIO): Update.
+ * config/m68hc11/m68hc11.h (MOVE_RATIO): Update.
+ * config/cris/cris.h (MOVE_RATIO): Update.
+ * config/mn10300/mn10300.h (MOVE_RATIO): Update.
+ * config/arm/arm.h (MOVE_RATIO): Update.
+ * config/pa/pa.md: Update uses of MOVE_RATIO
+ * config/pa/pa.h (MOVE_RATIO): Update.
+ * config/mips/mips.h (MOVE_RATIO, MOVE_BY_PIECES, CLEAR_RATIO, SET_RATIO): Update.
+ * config/h8300/h8300.h (MOVE_RATIO): Update.
+ * config/v850/v850.h (MOVE_RATIO): Update.
+ * config/bfin/bfin.h (MOVE_RATIO): Update.
+
2008-08-31 Andrey Belevantsev <abel@ispras.ru>
Dmitry Melnik <dm@ispras.ru>
Dmitry Zhurikhin <zhur@ispras.ru>
Without byte/word accesses, we want no more than four instructions;
with, several single byte accesses are better. */
-#define MOVE_RATIO (TARGET_BWX ? 7 : 2)
+#define MOVE_RATIO(speed) (TARGET_BWX ? 7 : 2)
/* Largest number of bytes of an object that can be placed in a register.
On the Alpha we have plenty of registers, so use TImode. */
#define MOVE_MAX 4
#undef MOVE_RATIO
-#define MOVE_RATIO (arm_tune_xscale ? 4 : 2)
+#define MOVE_RATIO(speed) (arm_tune_xscale ? 4 : 2)
/* Define if operations between registers always perform the operation
on the full register even if a narrower mode is specified. */
/* If a memory-to-memory move would take MOVE_RATIO or more simple
move-instruction pairs, we will do a movmem or libcall instead. */
-#define MOVE_RATIO 5
+#define MOVE_RATIO(speed) 5
/* STORAGE LAYOUT: target machine storage layout
Define this macro as a C expression which is nonzero if accessing
word-length sizes will be emitted. The "9" will translate to
(9 - 1) * 4 = 32 bytes maximum moved, but using 16 instructions
(8 instruction sequences) or less. */
-#define MOVE_RATIO 9
+#define MOVE_RATIO(speed) 9
/* Node: Sections */
#define FINAL_PRESCAN_INSN(insn, operand, nop) \
final_prescan_insn (insn, operand, nop)
-#define MOVE_RATIO 3
extern int h8300_move_ratio;
-#undef MOVE_RATIO
-#define MOVE_RATIO h8300_move_ratio
+#define MOVE_RATIO(speed) h8300_move_ratio
/* Machine-specific symbol_ref flags. */
#define SYMBOL_FLAG_FUNCVEC_FUNCTION (SYMBOL_FLAG_MACH_DEP << 0)
If you don't define this, a reasonable default is used. */
-#define MOVE_RATIO (optimize_size ? 3 : ix86_cost->move_ratio)
+#define MOVE_RATIO(speed) ((speed) ? ix86_cost->move_ratio : 3)
/* If a clear memory operation would take CLEAR_RATIO or more simple
move-instruction sequences, we will do a clrmem or libcall instead. */
-#define CLEAR_RATIO (optimize_size ? 2 : MIN (6, ix86_cost->move_ratio))
+#define CLEAR_RATIO(speed) ((speed) ? MIN (6, ix86_cost->move_ratio) : 2)
/* Define if shifts truncate the shift count
which implies one can omit a sign-extension or zero-extension
/* MOVE_RATIO is the number of move instructions that is better than a
block move. Make this small on 6811, since the code size grows very
large with each move. */
-#define MOVE_RATIO 3
+#define MOVE_RATIO(speed) 3
/* Define if shifts truncate the shift count which implies one can omit
a sign-extension or zero-extension of a shift count. */
we'll have to generate a load/store pair for each, halve the
value of MIPS_CALL_RATIO to take that into account. */
-#define MOVE_RATIO \
+#define MOVE_RATIO(speed) \
(HAVE_movmemsi \
? MIPS_MAX_MOVE_BYTES_STRAIGHT / MOVE_MAX \
: MIPS_CALL_RATIO / 2)
? (SIZE) < UNITS_PER_WORD \
: (SIZE) <= MIPS_MAX_MOVE_BYTES_STRAIGHT)) \
: (move_by_pieces_ninsns (SIZE, ALIGN, MOVE_MAX_PIECES + 1) \
- < (unsigned int) MOVE_RATIO))
+ < (unsigned int) MOVE_RATIO (false)))
/* For CLEAR_RATIO, when optimizing for size, give a better estimate
of the length of a memset call, but use the default otherwise. */
-#define CLEAR_RATIO \
- (optimize_size ? MIPS_CALL_RATIO : 15)
+#define CLEAR_RATIO(speed)\
+ ((speed) ? 15 : MIPS_CALL_RATIO)
/* This is similar to CLEAR_RATIO, but for a non-zero constant, so when
optimizing for size adjust the ratio to account for the overhead of
loading the constant and replicating it across the word. */
-#define SET_RATIO \
- (optimize_size ? MIPS_CALL_RATIO - 2 : 15)
+#define SET_RATIO(speed) \
+ ((speed) ? 15 : MIPS_CALL_RATIO - 2)
/* STORE_BY_PIECES_P can be used when copying a constant string, but
in that case each word takes 3 insns (lui, ori, sw), or more in
/* According expr.c, a value of around 6 should minimize code size, and
for the MN10300 series, that's our primary concern. */
-#define MOVE_RATIO 6
+#define MOVE_RATIO(speed) 6
#define TEXT_SECTION_ASM_OP "\t.section .text"
#define DATA_SECTION_ASM_OP "\t.section .data"
arguments passed in registers to avoid infinite recursion during argument
setup for a function call. Why? Consider how we copy the stack slots
reserved for parameters when they may be trashed by a call. */
-#define MOVE_RATIO (TARGET_64BIT ? 8 : 4)
+#define MOVE_RATIO(speed) (TARGET_64BIT ? 8 : 4)
/* Define if operations between registers always perform the operation
on the full register even if a narrower mode is specified. */
FAIL;
/* This does happen, but not often enough to worry much about. */
- if (size / align < MOVE_RATIO)
+ if (size / align < MOVE_RATIO (optimize_insn_for_speed_p ()))
FAIL;
/* Fall through means we're going to use our block move pattern. */
FAIL;
/* This does happen, but not often enough to worry much about. */
- if (size / align < MOVE_RATIO)
+ if (size / align < MOVE_RATIO (optimize_insn_for_speed_p ()))
FAIL;
/* Fall through means we're going to use our block move pattern. */
FAIL;
/* This does happen, but not often enough to worry much about. */
- if (size / align < MOVE_RATIO)
+ if (size / align < MOVE_RATIO (optimize_insn_for_speed_p ()))
FAIL;
/* Fall through means we're going to use our block clear pattern. */
FAIL;
/* This does happen, but not often enough to worry much about. */
- if (size / align < MOVE_RATIO)
+ if (size / align < MOVE_RATIO (optimize_insn_for_speed_p ()))
FAIL;
/* Fall through means we're going to use our block clear pattern. */
in tree-sra with UNITS_PER_WORD to make a decision so we adjust it
here to compensate for that factor since mvc costs exactly the same
on 31 and 64 bit. */
-#define MOVE_RATIO (TARGET_64BIT? 2 : 4)
+#define MOVE_RATIO(speed) (TARGET_64BIT? 2 : 4)
/* Sections. */
/* If a memory-to-memory move would take MOVE_RATIO or more simple
move-instruction pairs, we will do a movmem or libcall instead. */
-#define MOVE_RATIO (optimize_size ? 3 : 8)
+#define MOVE_RATIO(speed) ((speed) ? 8 : 3)
/* Define if operations between registers always perform the operation
on the full register even if a narrower mode is specified. */
#define SLOW_BYTE_ACCESS 0
-#define MOVE_RATIO 32
+#define MOVE_RATIO(speed) 32
#define NO_FUNCTION_CSE
/* According expr.c, a value of around 6 should minimize code size, and
for the V850 series, that's our primary concern. */
-#define MOVE_RATIO 6
+#define MOVE_RATIO(speed) 6
/* Indirect calls are expensive, never turn a direct call
into an indirect call. */
#ifndef MOVE_BY_PIECES_P
#define MOVE_BY_PIECES_P(SIZE, ALIGN) \
(move_by_pieces_ninsns (SIZE, ALIGN, MOVE_MAX_PIECES + 1) \
- < (unsigned int) MOVE_RATIO)
+ < (unsigned int) MOVE_RATIO (optimize_insn_for_speed_p ()))
#endif
/* This macro is used to determine whether clear_by_pieces should be
#ifndef CLEAR_BY_PIECES_P
#define CLEAR_BY_PIECES_P(SIZE, ALIGN) \
(move_by_pieces_ninsns (SIZE, ALIGN, STORE_MAX_PIECES + 1) \
- < (unsigned int) CLEAR_RATIO)
+ < (unsigned int) CLEAR_RATIO (optimize_insn_for_speed_p ()))
#endif
/* This macro is used to determine whether store_by_pieces should be
#ifndef SET_BY_PIECES_P
#define SET_BY_PIECES_P(SIZE, ALIGN) \
(move_by_pieces_ninsns (SIZE, ALIGN, STORE_MAX_PIECES + 1) \
- < (unsigned int) SET_RATIO)
+ < (unsigned int) SET_RATIO (optimize_insn_for_speed_p ()))
#endif
/* This macro is used to determine whether store_by_pieces should be
#ifndef STORE_BY_PIECES_P
#define STORE_BY_PIECES_P(SIZE, ALIGN) \
(move_by_pieces_ninsns (SIZE, ALIGN, STORE_MAX_PIECES + 1) \
- < (unsigned int) MOVE_RATIO)
+ < (unsigned int) MOVE_RATIO (optimize_insn_for_speed_p ()))
#endif
/* This array records the insn_code of insns to perform block moves. */
#ifndef MOVE_RATIO
#if defined (HAVE_movmemqi) || defined (HAVE_movmemhi) || defined (HAVE_movmemsi) || defined (HAVE_movmemdi) || defined (HAVE_movmemti)
-#define MOVE_RATIO 2
+#define MOVE_RATIO(speed) 2
#else
/* If we are optimizing for space (-Os), cut down the default move ratio. */
-#define MOVE_RATIO (optimize_size ? 3 : 15)
+#define MOVE_RATIO(speed) ((speed) ? 15 : 3)
#endif
#endif
#ifndef CLEAR_RATIO
#if defined (HAVE_setmemqi) || defined (HAVE_setmemhi) || defined (HAVE_setmemsi) || defined (HAVE_setmemdi) || defined (HAVE_setmemti)
-#define CLEAR_RATIO 2
+#define CLEAR_RATIO(speed) 2
#else
/* If we are optimizing for space, cut down the default clear ratio. */
-#define CLEAR_RATIO (optimize_size ? 3 : 15)
+#define CLEAR_RATIO(speed) ((speed) ? 15 :3)
#endif
#endif
SET_RATIO or more simple move-instruction sequences, we will do a movmem
or libcall instead. */
#ifndef SET_RATIO
-#define SET_RATIO MOVE_RATIO
+#define SET_RATIO(speed) MOVE_RATIO(speed)
#endif
\f
enum direction {none, upward, downward};
if (num_type_elements < 0 && int_size_in_bytes (type) >= 0)
cleared = true;
/* If there are "lots" of zeros, then block clear the object first. */
- else if (num_type_elements - num_nonzero_elements > CLEAR_RATIO
+ else if (num_type_elements - num_nonzero_elements
+ > CLEAR_RATIO (optimize_function_for_speed_p (cfun))
&& num_nonzero_elements < num_type_elements/4)
cleared = true;
/* ??? This bit ought not be needed. For any element not present
size = int_size_in_bytes (type);
- if (size < 0 || size > MOVE_MAX_PIECES * MOVE_RATIO)
+ if (size < 0 || size > MOVE_MAX_PIECES * MOVE_RATIO (!optimize_size))
/* Cost of a memcpy call, 3 arguments and the call. */
return 4;
else
sensible default. */
max_size = SRA_MAX_STRUCTURE_SIZE
? SRA_MAX_STRUCTURE_SIZE
- : MOVE_RATIO * UNITS_PER_WORD;
+ : MOVE_RATIO (optimize_function_for_speed_p (cfun)) * UNITS_PER_WORD;
max_count = SRA_MAX_STRUCTURE_COUNT
? SRA_MAX_STRUCTURE_COUNT
- : MOVE_RATIO;
+ : MOVE_RATIO (optimize_function_for_speed_p (cfun));
full_size = tree_low_cst (size_tree, 1);
full_count = count_type_elements (elt->type, false);