ffitarget.h: Remove obsolete sgidefs.h include.

2005-12-07  Thiemo Seufer  <ths@networkno.de>

	* src/mips/ffitarget.h: Remove obsolete sgidefs.h include. Add
	missing parentheses.
	* src/mips/o32.S (ffi_call_O32): Code formatting. Define
	and use A3_OFF, FP_OFF, RA_OFF. Micro-optimizations.
	(ffi_closure_O32): Likewise, but with newly defined A3_OFF2,
	A2_OFF2, A1_OFF2, A0_OFF2, RA_OFF2, FP_OFF2, S0_OFF2, GP_OFF2,
	V1_OFF2, V0_OFF2, FA_1_1_OFF2, FA_1_0_OFF2, FA_0_1_OFF2,
	FA_0_0_OFF2.
	* src/mips/ffi.c (ffi_prep_args): Code formatting. Fix
	endianness bugs.
	(ffi_prep_closure): Improve trampoline instruction scheduling.
	(ffi_closure_mips_inner_O32): Fix endianness bugs.

From-SVN: r108165
This commit is contained in:
Thiemo Seufer 2005-12-07 14:56:38 +00:00 committed by Matthias Klose
parent 7f416ffb28
commit 8a39029d19
4 changed files with 207 additions and 168 deletions

View File

@ -1,3 +1,18 @@
2005-12-07 Thiemo Seufer <ths@networkno.de>
* src/mips/ffitarget.h: Remove obsolete sgidefs.h include. Add
missing parentheses.
* src/mips/o32.S (ffi_call_O32): Code formatting. Define
and use A3_OFF, FP_OFF, RA_OFF. Micro-optimizations.
(ffi_closure_O32): Likewise, but with newly defined A3_OFF2,
A2_OFF2, A1_OFF2, A0_OFF2, RA_OFF2, FP_OFF2, S0_OFF2, GP_OFF2,
V1_OFF2, V0_OFF2, FA_1_1_OFF2, FA_1_0_OFF2, FA_0_1_OFF2,
FA_0_0_OFF2.
* src/mips/ffi.c (ffi_prep_args): Code formatting. Fix
endianness bugs.
(ffi_prep_closure): Improve trampoline instruction scheduling.
(ffi_closure_mips_inner_O32): Fix endianness bugs.
2005-12-03 Alan Modra <amodra@bigpond.net.au>
* src/powerpc/ffi.c: Formatting.

View File

@ -50,17 +50,17 @@ static void ffi_prep_args(char *stack,
int bytes,
int flags)
{
register int i;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
int i;
void **p_argv;
char *argp;
ffi_type **p_arg;
#if _MIPS_SIM == _ABIN32
/* If more than 8 double words are used, the remainder go
on the stack. We reorder stuff on the stack here to
support this easily. */
if (bytes > 8 * FFI_SIZEOF_ARG)
argp = &stack[bytes - (8 * FFI_SIZEOF_ARG)];
if (bytes > 8 * sizeof(ffi_arg))
argp = &stack[bytes - (8 * sizeof(ffi_arg))];
else
argp = stack;
#else
@ -85,102 +85,93 @@ static void ffi_prep_args(char *stack,
for (i = ecif->cif->nargs, p_arg = ecif->cif->arg_types; i; i--, p_arg++)
{
size_t z;
unsigned short a;
unsigned int a;
/* Align if necessary */
/* Align if necessary. */
a = (*p_arg)->alignment;
if (a < FFI_SIZEOF_ARG)
a = FFI_SIZEOF_ARG;
if (a < sizeof(ffi_arg))
a = sizeof(ffi_arg);
if ((a - 1) & (unsigned) argp) {
argp = (char *) ALIGN(argp, a);
FIX_ARGP;
}
#if _MIPS_SIM == _ABIO32
#define OFFSET 0
#else
#define OFFSET sizeof(int)
#endif
z = (*p_arg)->size;
if (z < sizeof(ffi_arg))
{
z = sizeof(ffi_arg);
switch ((*p_arg)->type)
{
case FFI_TYPE_SINT8:
*(SINT32 *) &argp[OFFSET] = (SINT32)*(SINT8 *)(* p_argv);
break;
case FFI_TYPE_UINT8:
*(UINT32 *) &argp[OFFSET] = (UINT32)*(UINT8 *)(* p_argv);
break;
case FFI_TYPE_SINT16:
*(SINT32 *) &argp[OFFSET] = (SINT32)*(SINT16 *)(* p_argv);
break;
case FFI_TYPE_UINT16:
*(UINT32 *) &argp[OFFSET] = (UINT32)*(UINT16 *)(* p_argv);
break;
case FFI_TYPE_SINT32:
*(SINT32 *) &argp[OFFSET] = (SINT32)*(SINT32 *)(* p_argv);
break;
case FFI_TYPE_UINT32:
case FFI_TYPE_POINTER:
*(UINT32 *) &argp[OFFSET] = (UINT32)*(UINT32 *)(* p_argv);
break;
/* This can only happen with 64bit slots */
case FFI_TYPE_FLOAT:
*(float *) argp = *(float *)(* p_argv);
break;
/* Handle small structures */
case FFI_TYPE_STRUCT:
memcpy(argp, *p_argv, (*p_arg)->size);
break;
default:
FFI_ASSERT(0);
}
}
else
{
#if _MIPS_SIM == _ABIO32
memcpy(argp, *p_argv, z);
#else
{
unsigned end = (unsigned) argp+z;
unsigned cap = (unsigned) stack+bytes;
/* Check if the data will fit within the register
space. Handle it if it doesn't. */
if (end <= cap)
memcpy(argp, *p_argv, z);
else
{
unsigned portion = end - cap;
memcpy(argp, *p_argv, portion);
argp = stack;
memcpy(argp,
(void*)((unsigned)(*p_argv)+portion), z - portion);
}
}
#endif
}
p_argv++;
argp += z;
if ((a - 1) & (unsigned int) argp)
{
argp = (char *) ALIGN(argp, a);
FIX_ARGP;
}
z = (*p_arg)->size;
if (z <= sizeof(ffi_arg))
{
z = sizeof(ffi_arg);
switch ((*p_arg)->type)
{
case FFI_TYPE_SINT8:
*(ffi_arg *)argp = *(SINT8 *)(* p_argv);
break;
case FFI_TYPE_UINT8:
*(ffi_arg *)argp = *(UINT8 *)(* p_argv);
break;
case FFI_TYPE_SINT16:
*(ffi_arg *)argp = *(SINT16 *)(* p_argv);
break;
case FFI_TYPE_UINT16:
*(ffi_arg *)argp = *(UINT16 *)(* p_argv);
break;
case FFI_TYPE_SINT32:
*(ffi_arg *)argp = *(SINT32 *)(* p_argv);
break;
case FFI_TYPE_UINT32:
case FFI_TYPE_POINTER:
*(ffi_arg *)argp = *(UINT32 *)(* p_argv);
break;
/* This can only happen with 64bit slots. */
case FFI_TYPE_FLOAT:
*(float *) argp = *(float *)(* p_argv);
break;
/* Handle small structures. */
case FFI_TYPE_STRUCT:
default:
memcpy(argp, *p_argv, (*p_arg)->size);
break;
}
}
else
{
#if _MIPS_SIM == _ABIO32
memcpy(argp, *p_argv, z);
#else
{
unsigned end = (unsigned) argp+z;
unsigned cap = (unsigned) stack+bytes;
/* Check if the data will fit within the register space.
Handle it if it doesn't. */
if (end <= cap)
memcpy(argp, *p_argv, z);
else
{
unsigned portion = end - cap;
memcpy(argp, *p_argv, portion);
argp = stack;
memcpy(argp,
(void*)((unsigned)(*p_argv)+portion), z - portion);
}
}
#endif
}
p_argv++;
argp += z;
FIX_ARGP;
}
return;
}
#if _MIPS_SIM == _ABIN32
@ -524,8 +515,8 @@ ffi_prep_closure (ffi_closure *closure,
#endif /* FFI_MIPS_O32 */
tramp[0] = 0x3c190000 | (fn >> 16); /* lui $25,high(fn) */
tramp[1] = 0x3c080000 | (ctx >> 16); /* lui $8,high(ctx) */
tramp[2] = 0x37390000 | (fn & 0xffff); /* ori $25,low(fn) */
tramp[1] = 0x37390000 | (fn & 0xffff); /* ori $25,low(fn) */
tramp[2] = 0x3c080000 | (ctx >> 16); /* lui $8,high(ctx) */
tramp[3] = 0x03200008; /* jr $25 */
tramp[4] = 0x35080000 | (ctx & 0xffff); /* ori $8,low(ctx) */
@ -558,16 +549,18 @@ ffi_prep_closure (ffi_closure *closure,
*/
int
ffi_closure_mips_inner_O32 (ffi_closure *closure,
void *rvalue, unsigned long *ar,
void *rvalue, ffi_arg *ar,
double *fpr)
{
ffi_cif *cif;
void **avalue;
void **avaluep;
ffi_arg *avalue;
ffi_type **arg_types;
int i, avn, argn, seen_int;
cif = closure->cif;
avalue = alloca (cif->nargs * sizeof (void *));
avalue = alloca (cif->nargs * sizeof (ffi_arg));
avaluep = alloca (cif->nargs * sizeof (ffi_arg));
seen_int = (cif->abi == FFI_O32_SOFT_FLOAT);
argn = 0;
@ -588,13 +581,43 @@ ffi_closure_mips_inner_O32 (ffi_closure *closure,
(arg_types[i]->type == FFI_TYPE_FLOAT ||
arg_types[i]->type == FFI_TYPE_DOUBLE))
{
avalue[i] = ((char *) &fpr[i]);
#ifdef __MIPSEB__
if (arg_types[i]->type == FFI_TYPE_FLOAT)
avaluep[i] = ((char *) &fpr[i]) + sizeof (float);
else
#endif
avaluep[i] = (char *) &fpr[i];
}
else
{
if (arg_types[i]->alignment == 8 && (argn & 0x1))
argn++;
avalue[i] = ((char *) &ar[argn]);
switch (arg_types[i]->type)
{
case FFI_TYPE_SINT8:
avaluep[i] = &avalue[i];
*(SINT8 *) &avalue[i] = (SINT8) ar[argn];
break;
case FFI_TYPE_UINT8:
avaluep[i] = &avalue[i];
*(UINT8 *) &avalue[i] = (UINT8) ar[argn];
break;
case FFI_TYPE_SINT16:
avaluep[i] = &avalue[i];
*(SINT16 *) &avalue[i] = (SINT16) ar[argn];
break;
case FFI_TYPE_UINT16:
avaluep[i] = &avalue[i];
*(UINT16 *) &avalue[i] = (UINT16) ar[argn];
break;
default:
avaluep[i] = (char *) &ar[argn];
break;
}
seen_int = 1;
}
argn += ALIGN(arg_types[i]->size, FFI_SIZEOF_ARG) / FFI_SIZEOF_ARG;
@ -602,7 +625,7 @@ ffi_closure_mips_inner_O32 (ffi_closure *closure,
}
/* Invoke the closure. */
(closure->fun) (cif, rvalue, avalue, closure->user_data);
(closure->fun) (cif, rvalue, avaluep, closure->user_data);
if (cif->abi == FFI_O32_SOFT_FLOAT)
{

View File

@ -26,17 +26,13 @@
#ifndef LIBFFI_TARGET_H
#define LIBFFI_TARGET_H
#ifndef LIBFFI_ASM
#include <sgidefs.h>
#endif
#if !defined(_MIPS_SIM)
-- something is very wrong --
#else
# if (_MIPS_SIM==_ABIN32 && defined(_ABIN32)) || (_MIPS_SIM==_ABI64 && defined(_ABI64))
# define FFI_MIPS_N32
# else
# if _MIPS_SIM==_ABIO32 && defined(_ABIO32)
# if (_MIPS_SIM==_ABIO32 && defined(_ABIO32))
# define FFI_MIPS_O32
# else
-- this is an unsupported platform --

View File

@ -35,7 +35,10 @@
#define bytes a2
#define flags a3
#define SIZEOF_FRAME ( 4 * FFI_SIZEOF_ARG + 2 * FFI_SIZEOF_ARG )
#define SIZEOF_FRAME (4 * FFI_SIZEOF_ARG + 2 * FFI_SIZEOF_ARG)
#define A3_OFF (SIZEOF_FRAME + 3 * FFI_SIZEOF_ARG)
#define FP_OFF (SIZEOF_FRAME - 2 * FFI_SIZEOF_ARG)
#define RA_OFF (SIZEOF_FRAME - 1 * FFI_SIZEOF_ARG)
.abicalls
.text
@ -45,48 +48,42 @@
ffi_call_O32:
$LFB0:
# Prologue
SUBU $sp, SIZEOF_FRAME # Frame size
SUBU $sp, SIZEOF_FRAME # Frame size
$LCFI0:
REG_S $fp, SIZEOF_FRAME - 2*FFI_SIZEOF_ARG($sp) # Save frame pointer
REG_S $fp, FP_OFF($sp) # Save frame pointer
$LCFI1:
REG_S ra, SIZEOF_FRAME - 1*FFI_SIZEOF_ARG($sp) # Save return address
REG_S ra, RA_OFF($sp) # Save return address
$LCFI2:
move $fp, $sp
$LCFI3:
move t9, callback # callback function pointer
REG_S flags, SIZEOF_FRAME + 3*FFI_SIZEOF_ARG($fp) # flags
move t9, callback # callback function pointer
REG_S flags, A3_OFF($fp) # flags
# Allocate at least 4 words in the argstack
move v0, bytes
bge bytes, 4 * FFI_SIZEOF_ARG, bigger
LI v0, 4 * FFI_SIZEOF_ARG
b sixteen
blt bytes, v0, sixteen
bigger:
ADDU t0, v0, 2 * FFI_SIZEOF_ARG -1 # make sure it is aligned
and v0, t0, -2 * FFI_SIZEOF_ARG # to an 8 byte boundry
ADDU v0, bytes, 7 # make sure it is aligned
and v0, -8 # to an 8 byte boundry
sixteen:
SUBU $sp, $sp, v0 # move the stack pointer to reflect the
SUBU $sp, v0 # move the stack pointer to reflect the
# arg space
ADDU a0, $sp, 4 * FFI_SIZEOF_ARG
ADDU a3, $fp, SIZEOF_FRAME + 3*FFI_SIZEOF_ARG
jalr t9
REG_L t0, SIZEOF_FRAME + 3*FFI_SIZEOF_ARG($fp) # load the flags word
add t2, t0, 0 # and copy it into t2
REG_L t0, A3_OFF($fp) # load the flags word
SRL t2, t0, 4 # shift our arg info
and t0, ((1<<4)-1) # mask out the return type
SRL t2, 4 # shift our arg info
ADDU $sp, $sp, 4 * FFI_SIZEOF_ARG # adjust $sp to new args
ADDU $sp, 4 * FFI_SIZEOF_ARG # adjust $sp to new args
bnez t0, pass_d # make it quick for int
REG_L a0, 0*FFI_SIZEOF_ARG($sp) # just go ahead and load the
REG_L a1, 1*FFI_SIZEOF_ARG($sp) # four regs.
REG_L a0, 0*FFI_SIZEOF_ARG($sp) # just go ahead and load the
REG_L a1, 1*FFI_SIZEOF_ARG($sp) # four regs.
REG_L a2, 2*FFI_SIZEOF_ARG($sp)
REG_L a3, 3*FFI_SIZEOF_ARG($sp)
b call_it
@ -176,9 +173,9 @@ noretval:
# Epilogue
epilogue:
move $sp, $fp
REG_L $fp, SIZEOF_FRAME - 2*FFI_SIZEOF_ARG($sp) # Restore frame pointer
REG_L ra, SIZEOF_FRAME - 1*FFI_SIZEOF_ARG($sp) # Restore return address
ADDU $sp, SIZEOF_FRAME # Fix stack pointer
REG_L $fp, FP_OFF($sp) # Restore frame pointer
REG_L ra, RA_OFF($sp) # Restore return address
ADDU $sp, SIZEOF_FRAME # Fix stack pointer
j ra
$LFE0:
@ -209,7 +206,21 @@ $LFE0:
0 - Called function a0 save our sp, fp point here
*/
#define SIZEOF_FRAME2 ( 14 * FFI_SIZEOF_ARG )
#define SIZEOF_FRAME2 (14 * FFI_SIZEOF_ARG)
#define A3_OFF2 (SIZEOF_FRAME2 + 3 * FFI_SIZEOF_ARG)
#define A2_OFF2 (SIZEOF_FRAME2 + 2 * FFI_SIZEOF_ARG)
#define A1_OFF2 (SIZEOF_FRAME2 + 1 * FFI_SIZEOF_ARG)
#define A0_OFF2 (SIZEOF_FRAME2 + 0 * FFI_SIZEOF_ARG)
#define RA_OFF2 (SIZEOF_FRAME2 - 1 * FFI_SIZEOF_ARG)
#define FP_OFF2 (SIZEOF_FRAME2 - 2 * FFI_SIZEOF_ARG)
#define S0_OFF2 (SIZEOF_FRAME2 - 3 * FFI_SIZEOF_ARG)
#define GP_OFF2 (SIZEOF_FRAME2 - 4 * FFI_SIZEOF_ARG)
#define V1_OFF2 (SIZEOF_FRAME2 - 5 * FFI_SIZEOF_ARG)
#define V0_OFF2 (SIZEOF_FRAME2 - 6 * FFI_SIZEOF_ARG)
#define FA_1_1_OFF2 (SIZEOF_FRAME2 - 7 * FFI_SIZEOF_ARG)
#define FA_1_0_OFF2 (SIZEOF_FRAME2 - 8 * FFI_SIZEOF_ARG)
#define FA_0_1_OFF2 (SIZEOF_FRAME2 - 9 * FFI_SIZEOF_ARG)
#define FA_0_0_OFF2 (SIZEOF_FRAME2 - 10 * FFI_SIZEOF_ARG)
.text
.align 2
@ -218,28 +229,28 @@ $LFE0:
ffi_closure_O32:
$LFB1:
# Prologue
.frame $fp, SIZEOF_FRAME2, $31
.frame $fp, SIZEOF_FRAME2, ra
.set noreorder
.cpload $25
.cpload t9
.set reorder
SUBU $sp, SIZEOF_FRAME2
.cprestore SIZEOF_FRAME2 - 4*FFI_SIZEOF_ARG
.cprestore GP_OFF2
$LCFI4:
REG_S $16, SIZEOF_FRAME2 - 3*FFI_SIZEOF_ARG($sp) # Save s0
REG_S $fp, SIZEOF_FRAME2 - 2*FFI_SIZEOF_ARG($sp) # Save frame pointer
REG_S ra, SIZEOF_FRAME2 - 1*FFI_SIZEOF_ARG($sp) # Save return address
REG_S $16, S0_OFF2($sp) # Save s0
REG_S $fp, FP_OFF2($sp) # Save frame pointer
REG_S ra, RA_OFF2($sp) # Save return address
$LCFI6:
move $fp, $sp
$LCFI7:
# Store all possible argument registers. If there are more than
# four arguments, then they should be stored above where we put $7.
REG_S $4, SIZEOF_FRAME2 + 0*FFI_SIZEOF_ARG($fp)
REG_S $5, SIZEOF_FRAME2 + 1*FFI_SIZEOF_ARG($fp)
REG_S $6, SIZEOF_FRAME2 + 2*FFI_SIZEOF_ARG($fp)
REG_S $7, SIZEOF_FRAME2 + 3*FFI_SIZEOF_ARG($fp)
# four arguments, then they are stored above where we put a3.
REG_S a0, A0_OFF2($fp)
REG_S a1, A1_OFF2($fp)
REG_S a2, A2_OFF2($fp)
REG_S a3, A3_OFF2($fp)
# Load ABI enum to $16
# Load ABI enum to s0
REG_L $16, 20($8) # cif pointer follows tramp.
REG_L $16, 0($16) # abi is first member.
@ -247,16 +258,16 @@ $LCFI7:
bne $16, $13, 1f # Skip fp save if FFI_O32_SOFT_FLOAT
# Store all possible float/double registers.
s.d $f12, SIZEOF_FRAME2 - 10*FFI_SIZEOF_ARG($fp)
s.d $f14, SIZEOF_FRAME2 - 8*FFI_SIZEOF_ARG($fp)
s.d $f12, FA_0_0_OFF2($fp)
s.d $f14, FA_1_0_OFF2($fp)
1:
# Call ffi_closure_mips_inner_O32 to do the work.
la $25, ffi_closure_mips_inner_O32
move $4, $8 # Pointer to the ffi_closure
addu $5, $fp, SIZEOF_FRAME2 - 6*FFI_SIZEOF_ARG
addu $6, $fp, SIZEOF_FRAME2 + 0*FFI_SIZEOF_ARG
addu $7, $fp, SIZEOF_FRAME2 - 10*FFI_SIZEOF_ARG
jalr $31, $25
la t9, ffi_closure_mips_inner_O32
move a0, $8 # Pointer to the ffi_closure
addu a1, $fp, V0_OFF2
addu a2, $fp, A0_OFF2
addu a3, $fp, FA_0_0_OFF2
jalr t9
# Load the return value into the appropriate register.
move $8, $2
@ -267,28 +278,22 @@ $LCFI7:
bne $16, $13, 1f # Skip fp restore if FFI_O32_SOFT_FLOAT
li $9, FFI_TYPE_FLOAT
l.s $f0, SIZEOF_FRAME2 - 6*FFI_SIZEOF_ARG($fp)
l.s $f0, V0_OFF2($fp)
beq $8, $9, closure_done
li $9, FFI_TYPE_DOUBLE
l.d $f0, SIZEOF_FRAME2 - 6*FFI_SIZEOF_ARG($fp)
l.d $f0, V0_OFF2($fp)
beq $8, $9, closure_done
1:
li $9, FFI_TYPE_SINT64
REG_L $3, SIZEOF_FRAME2 - 5*FFI_SIZEOF_ARG($fp)
beq $8, $9, integer
li $9, FFI_TYPE_UINT64
beq $8, $9, integer
integer:
REG_L $2, SIZEOF_FRAME2 - 6*FFI_SIZEOF_ARG($fp)
REG_L $3, V1_OFF2($fp)
REG_L $2, V0_OFF2($fp)
closure_done:
# Epilogue
move $sp, $fp
REG_L $16, SIZEOF_FRAME2 - 3*FFI_SIZEOF_ARG($sp) # Restore s0
REG_L $fp, SIZEOF_FRAME2 - 2*FFI_SIZEOF_ARG($sp) # Restore frame pointer
REG_L ra, SIZEOF_FRAME2 - 1*FFI_SIZEOF_ARG($sp) # Restore return address
REG_L $16, S0_OFF2($sp) # Restore s0
REG_L $fp, FP_OFF2($sp) # Restore frame pointer
REG_L ra, RA_OFF2($sp) # Restore return address
ADDU $sp, SIZEOF_FRAME2
j ra
$LFE1: