Merge pull request #6 from kernigh/pr-linuxppc

PowerPC fixes
This commit is contained in:
David Given 2016-10-19 20:39:10 +02:00 committed by GitHub
commit 34c4cfee8f
33 changed files with 1117 additions and 809 deletions

View file

@ -8,6 +8,7 @@ vars.plats = {
"cpm",
"linux386",
"linux68k",
"linuxppc",
"pc86",
"rpi",
}

View file

@ -8,7 +8,8 @@
program hilo(input, output);
type
string = array [0..255] of char;
string = packed array [0..255] of char;
charstar = packed array [0..0] of char;
var
playing : Boolean;
@ -34,18 +35,18 @@ function random(range : integer) : integer;
that conflicts with a Pascal keyword. Luckily there's a private function
uread() in the ACK Pascal library that we can use instead. }
function uread(fd : integer; var buffer : char; count : integer) : integer;
function uread(fd : integer; var buffer : charstar; count : integer) : integer;
extern;
function readchar : char;
var
c : char;
c : charstar;
dummy : integer;
begin
c := chr(0);
c[0] := chr(0);
dummy := uread(0, c, 1);
readchar := c;
readchar := c[0];
end;
procedure readstring(var buffer : string; var length : integer);

View file

@ -66,7 +66,7 @@ struct outname {
#define RELO2 2 /* 2 bytes */
#define RELO4 3 /* 4 bytes */
#define RELOPPC 4 /* PowerPC 26-bit address */
#define RELOH2 5 /* write top 2 bytes of 4 byte word */
/* relo 5 is unused */
#define RELOVC4 6 /* VideoCore IV address in 32-bit instruction */
#define RELPC 0x2000 /* pc relative */

View file

@ -80,7 +80,7 @@
%token <y_word> OP_TO_RA_RB
%token <y_word> OP_TO_RA_SI
%token <y_word> OP_la
%token <y_word> OP_LI32
/* Other token types */

View file

@ -98,7 +98,7 @@
/* Special instructions */
0, OP_la, 0, "la",
0, OP_LI32, 0, "li32",
/* Branch processor instructions (page 20) */

View file

@ -58,6 +58,7 @@ operation
| OP_LEV u7 { emit4($1 | ($2<<5)); }
| OP_LIA lia { emit4($1 | $2); }
| OP_LIL lil { emit4($1 | $2); }
| OP_LI32 li32 /* emitted in subrule */
;
c
@ -66,26 +67,12 @@ c
;
e16
: '<' expr
: absexp
{
DOTVAL += 2;
newrelo($2.typ, RELOH2 | FIXUPFLAGS);
DOTVAL -= 2;
$$ = ($2.val >> 16) & 0xFFFF;
}
| '>' expr
{
DOTVAL += 2;
newrelo($2.typ, RELO2 | FIXUPFLAGS);
DOTVAL -= 2;
$$ = $2.val & 0xFFFF;
}
| expr
{
DOTVAL += 2;
newrelo($1.typ, RELO2 | FIXUPFLAGS);
DOTVAL -= 2;
$$ = $1.val & 0xFFFF;
/* Allow signed or unsigned 16-bit values. */
if (($1 < -0x8000) || ($1 > 0xffff))
serror("16-bit value out of range");
$$ = (uint16_t) $1;
}
;
@ -206,6 +193,15 @@ bda
}
;
li32
: GPR ',' expr
{
newrelo($3.typ, RELOPPC | FIXUPFLAGS);
emit4((15<<26) | ($1<<21) | (0<<16) | ($3.val >> 16)); /* addis */
emit4((24<<26) | ($1<<21) | ($1<<16) | ($3.val & 0xffff)); /* ori */
}
;
lil
: expr
{

View file

@ -13,11 +13,14 @@
! r3 = ptr to descriptor
! r4 = index
! r5 = address of array
! Yields:
! r3 = address of element
! r0 = size of element (used by .lar4, .sar4)
! Preserves r10 for .lar4, .sar4
.define .aar4
.aar4:
addis r0, r0, <.trap_earray
ori r0, r0, >.trap_earray
li32 r0, .trap_earray
mtspr ctr, r0 ! load CTR with trap address
lwz r0, 0(r3)

View file

@ -0,0 +1,13 @@
for _, plat in ipairs(vars.plats) do
acklibrary {
name = "lib_"..plat,
srcs = {
"./*.s",
},
vars = { plat = plat },
deps = {
"h+emheaders"
}
}
end

View file

@ -13,15 +13,15 @@
.define .cfu8
.cfu8:
la(r3, .fd_00000000)
li32 r3, .fd_00000000
lfd f0, 0(r3) ! f0 = 0.0
lfd f1, 0(sp) ! value to be converted
la(r3, .fd_FFFFFFFF)
li32 r3, .fd_FFFFFFFF
lfd f3, 0(r3) ! f3 = 0xFFFFFFFF
la(r3, .fd_80000000)
li32 r3, .fd_80000000
lfd f4, 0(r3) ! f4 = 0x80000000
fsel f2, f1, f1, f0

View file

@ -24,7 +24,7 @@
lfd f0, 0(sp) ! load value
la (r3, pivot)
li32 r3, pivot
lfd f1, 0(r3) ! load pivot value
fsub f0, f0, f1 ! adjust

View file

@ -20,7 +20,7 @@
lfd f0, 0(sp) ! load value
la (r3, pivot)
li32 r3, pivot
lfd f1, 0(r3) ! load pivot value
fsub f0, f0, f1 ! adjust

View file

@ -1,46 +0,0 @@
/*
* $Source$
* $State$
* $Revision$
*/
/* no headers allowed! */
/* Given a double, calculates the mantissa and exponent.
*
* This function is intended to be called internally by the code generator,
* so the calling convention is odd.
*/
int __fef8(double* fp)
{
double f = *fp;
int exponent, sign;
if (f == 0.0)
return 0;
if (f < 0.0)
{
sign = -1;
f = -f;
}
else
sign = 0;
exponent = 0;
while (f >= 1.0)
{
f /= 2.0;
exponent++;
}
while (f < 0.5)
{
f *= 2.0;
exponent--;
}
*fp = (sign) ? -f : f;
return exponent;
}

58
mach/powerpc/libem/fef8.s Normal file
View file

@ -0,0 +1,58 @@
#include "powerpc.h"
.sect .text
! Split a double-precision float into fraction and exponent, like
! frexp(3) in C. On entry:
! r3 = float, high word (bits 0..31)
! r4 = float, low word (bits 32..63)
! Yields:
! r3 = fraction, high word (bits 0..31)
! r4 = fraction, low word (bits 32..63)
! r5 = exponent
! Kills: cr0 f0 f1 r6 r7
.define .fef8
.fef8:
! IEEE double-precision format:
! sign exponent fraction
! 0 1..11 12..63
rlwinm r6, r3, 12, 21, 31 ! r6 = IEEE exponent
addis r7, r0, 0x7ff0 ! r7 = exponent mask
addi r5, r6, -1022 ! r5 = true exponent
cmpi cr0, 0, r6, 2047
bclr IFTRUE, EQ, 0 ! return if infinity or NaN
cmpi cr0, 0, r6, 0
bc IFFALSE, EQ, 1f ! jump if normalized number
! Got denormalized number or zero, probably zero.
rlwinm r6, r3, 0, 12, 31
addi r5, r0, 0 ! r5 = true exponent = 0
or. r6, r6, r4 ! r6 = high|low fraction
bclr IFTRUE, EQ, 0 ! return if zero
! Got denormalized number, not zero.
stwu r4, -4(sp)
stwu r3, -4(sp)
li32 r6, _2_64
lfd f0, 0(sp)
lfd f1, 0(r6)
fmul f0, f0, f1 ! multiply it by 2**64
stfd f0, 0(sp)
lwz r3, 0(sp)
lwz r4, 4(sp)
rlwinm r6, r3, 12, 21, 31 ! r6 = IEEE exponent
addi sp, sp, 8
addi r5, r6, -1022 - 64 ! r5 = true exponent
1:
! Put fraction in [0.5, 1) or (-1, -0.5] by setting its
! exponent to true 0, IEEE 1022.
andc r3, r3, r7 ! clear old exponent
oris r3, r3, 1022 << 4 ! set new exponent
bclr ALWAYS, 0, 0
.sect .rom
_2_64:
! (double) 2**64
.data4 0x43f00000
.data4 0x00000000

View file

@ -1,38 +1,71 @@
#
! $Source$
! $State$
! $Revision$
#include "powerpc.h"
.sect .text
! Multiplies two floats, and returns the fraction and integer.
! Multiplies two double-precision floats, then splits the product into
! integer and fraction, like modf(3) in C. On entry:
! f1 = float
! f2 = other float
! Yields:
! f1 = fraction
! f2 = integer
! Kills: cr0 f1 f2 r3 r4 r5 r6
.define .fif8
.fif8:
lfd f0, 8(sp)
lfd f1, 0(sp)
fmul f0, f0, f1
fabs f1, f0 ! f0 = result
! The following chunk does f1 = floor(f1). See page 158 of the book.
mtfsfi cr7, 3 ! set rounding mode to -inf.
mtfsb0 23
fctid f2, f1
fcfid f2, f2
mcrfs cr7, cr5
bc IFFALSE, 31, toobig
fmr f1, f2
toobig:
fmul f1, f1, f2
stfdu f1, -8(sp) ! push f1 = product
lwz r3, 0(sp) ! r3 = high word
lwz r4, 4(sp) ! r4 = low word
fabs f2, f1 ! f2 = fabs(f1)
fsub f2, f2, f1
stfd f2, 8(sp)
fneg f2, f1
fsel f2, f0, f1, f2
stfd f2, 0(sp)
! IEEE double-precision format:
! sign exponent fraction
! 0 1..11 12..63
! Subtract 1023 from the IEEE exponent. If the result is from
! 0 to 51, then the IEEE fraction has that many integer bits.
! (IEEE has an implicit 1 before its fraction. If the IEEE
! fraction has 0 integer bits, we still have an integer.)
rlwinm r5, r3, 12, 21, 31 ! r5 = IEEE exponent
addic. r5, r5, -1023 ! r5 = nr of integer bits
bc IFTRUE, LT, no_int
cmpi cr0, 0, r5, 21
bc IFTRUE, LT, small_int
cmpi cr0, 0, r5, 52
bc IFTRUE, LT, big_int
! f1 is an integer without fraction. Jump to calculate
! fraction f1 = f2 - f1. It will be zero (or perhaps NaN).
fmr f2, f1
b subtract
no_int:
! f1 is a fraction without integer.
fsub f2, f1, f1 ! integer = zero
b done
small_int:
! f1 has r5 = 0 to 20 integer bits in the IEEE fraction.
! High word has 20 - r5 fraction bits.
addi r6, r0, 20
subf r6, r5, r6
srw r3, r3, r6
addi r4, r0, 0 ! clear low word
slw r3, r3, r6 ! clear fraction in high word
b move_int
big_int:
! f1 has r5 = 21 to 51 to integer bits.
! Low word has 52 - r5 fraction bits.
addi r6, r0, 52
subf r6, r5, r6
srw r4, r4, r6
slw r4, r4, r6 ! clear fraction in low word
move_int:
stw r3, 0(sp)
stw r4, 4(sp)
lfd f2, 0(sp) ! f2 = integer
subtract:
fsub f1, f1, f2 ! fraction = value - integer
done:
addi sp, sp, 8 ! restore stack pointer
bclr ALWAYS, 0, 0

43
mach/powerpc/libem/lar4.s Normal file
View file

@ -0,0 +1,43 @@
#
#include "powerpc.h"
.sect .text
! Load from bounds-checked array.
!
! On entry:
! r3 = ptr to descriptor
! r4 = index
! r5 = address of array
.define .lar4
.lar4:
mfspr r10, lr
bl .aar4
mtspr lr, r10
! r3 = ptr to element
! r0 = size of element
cmpi cr0, 0, r0, 1
bc IFFALSE, EQ, 1f
! Load 1 byte.
lbz r4, 0(r3)
stwu r4, -4(sp)
bclr ALWAYS, 0, 0
1:
cmpi cr0, 0, r0, 2
bc IFFALSE, EQ, 2f
! Load 2 bytes.
lhz r4, 0(r3)
stwu r4, -4(sp)
bclr ALWAYS, 0, 0
2:
! Load r0 bytes, where r0 must be a positive multiple of 4.
subf sp, r0, sp ! move stack pointer down
or r5, r0, r0 ! index r5 = length r0
3:
addic. r5, r5, -4 ! r5 -= 4
lwzx r4, r5, r3
stwx r4, r5, sp
bc IFTRUE, GT, 3b ! loop if r5 > 0
bclr ALWAYS, 0, 0

View file

@ -20,4 +20,3 @@
#define EQ 2
#define OV 3
#define la(reg, val) addis reg, r0, <val; ori reg, reg, >val

45
mach/powerpc/libem/sar4.s Normal file
View file

@ -0,0 +1,45 @@
#
#include "powerpc.h"
.sect .text
! Store to bounds-checked array.
!
! On entry:
! r3 = ptr to descriptor
! r4 = index
! r5 = address of array
.define .sar4
.sar4:
mfspr r10, lr
bl .aar4
mtspr lr, r10
! r3 = ptr to element
! r0 = size of element
cmpi cr0, 0, r0, 1
bc IFFALSE, EQ, 1f
! Store 1 byte.
lwz r4, 0(sp)
addi sp, sp, 4
stb r4, 0(r3)
bclr ALWAYS, 0, 0
1:
cmpi cr0, 0, r0, 2
bc IFFALSE, EQ, 2f
! Store 2 bytes.
lwz r4, 0(sp)
addi sp, sp, 4
sth r4, 0(r3)
bclr ALWAYS, 0, 0
2:
! Store r0 bytes, where r0 must be a positive multiple of 4.
or r5, r0, r0 ! index r5 = length r0
3:
addic. r5, r5, -4 ! r5 -= 4
lwzx r4, r5, sp
stwx r4, r5, r3
bc IFTRUE, GT, 3b ! loop if r5 > 0
add sp, r0, sp ! move stack pointer up
bclr ALWAYS, 0, 0

View file

@ -0,0 +1,8 @@
for _, plat in ipairs(vars.plats) do
acklibrary {
name = "lib_"..plat,
srcs = { "./*.s" },
vars = { plat = plat },
}
end

File diff suppressed because it is too large Load diff

View file

@ -909,23 +909,6 @@ normalfailed: if (stackpad!=tokpatlen) {
break;
}
#endif
#ifdef USE_NOFRAMEPOINTER
case DO_STACKADJUST: {
result_t result;
int nodeno;
DEBUG("STACKADJUST");
/* The offset is an expression, which we need to evaluate. */
getint(nodeno,codep);
compute(&enodes[nodeno], &result);
assert(result.e_typ==EV_INT);
if (toplevel)
stackoffset += result.e_v.e_con;
break;
}
#endif
}
}

View file

@ -20,9 +20,6 @@ extern rl_p curreglist; /* side effect of findcoerc() */
#ifndef NDEBUG
extern int Debug; /* on/off debug printout */
#endif
#ifdef USE_NOFRAMEPOINTER
extern int stackoffset; /* offset from localbase to sp */
#endif
/*
* Next descriptions are external declarations for tables created

View file

@ -41,7 +41,14 @@ begtext:
stwu r3, -4(sp)
b __m_a_i_n
! Define symbols at the beginning of our various segments, so that we can find
! them. (Except .text, which has already been done.)
.sect .data; begdata:
.sect .rom; begrom:
.sect .bss; begbss:
! Some magic data. All EM systems need these.
.define _errno
@ -50,7 +57,3 @@ begtext:
.define .trppc, .ignmask
.comm .trppc, 4 ! ptr to user trap handler
.comm .ignmask, 4 ! user trap ignore mask
.define .linenumber, .filename
.comm .linenumber, 4 ! current linenumber (used for debugging)
.comm .filename, 4 ! ptr to current filename (used for debugging)

View file

@ -0,0 +1,25 @@
include("plat/build.lua")
ackfile {
name = "boot",
srcs = { "./boot.s" },
vars = { plat = "linuxppc" }
}
build_plat_libs {
name = "libs",
arch = "powerpc",
plat = "linuxppc",
}
installable {
name = "pkg",
map = {
"+tools",
"+libs",
"./include+pkg",
["$(PLATIND)/linuxppc/boot.o"] = "+boot",
["$(PLATIND)/linuxppc/libsys.a"] = "./libsys+lib",
}
}

View file

@ -0,0 +1,21 @@
include("plat/build.lua")
build_as {
name = "as",
arch = "powerpc",
}
build_ncg {
name = "ncg",
arch = "powerpc",
}
return installable {
name = "tools",
map = {
["$(PLATDEP)/linuxppc/as"] = "+as",
["$(PLATDEP)/linuxppc/ncg"] = "+ncg",
["$(PLATIND)/descr/linuxppc"] = "./descr",
"util/opt+pkg",
}
}

View file

@ -1,6 +1,4 @@
# $Source: /cvsroot/tack/Ack/plat/linux386/descr,v $
# $State: Exp $
# $Revision: 1.1 $
# plat/linuxppc/descr
var w=4
var wa=4
@ -19,8 +17,8 @@ var xa={x}
var ARCH=powerpc
var PLATFORM=linuxppc
var PLATFORMDIR={EM}/share/ack/{PLATFORM}
var CPP_F=-D__unix -D__POWERPC
var ALIGN=-a0:4 -a1:4 -a2:4 -a3:4 -b0:0x80000054
var CPP_F=-D__unix
var ALIGN=-a0:4 -a1:4 -a2:4 -a3:4 -b0:0x10000054
var C_LIB={PLATFORMDIR}/libc-ansi.a
# bitfields reversed for compatibility with (g)cc.
var CC_ALIGN=-Vr
@ -35,25 +33,26 @@ var C_INCLUDES=-I{PLATFORMDIR}/include -I{EM}/share/ack/include/ansi
name be
from .m.g
to .s
program {EM}/lib/ack/{ARCH}/ncg
program {EM}/lib/ack/{PLATFORM}/ncg
mapflag -gdb GF=-gdb
args {GF?} <
stdout
need .e
end
name asopt
from .s
to .so
program {EM}/lib/ack/{ARCH}/top
args
optimizer
stdin
stdout
end
# FIXME(dtrg): not working yet
#name asopt
# from .s
# to .so
# program {EM}/lib/ack/{PLATFORM}/top
# args
# optimizer
# stdin
# stdout
#end
name as
from .s.so
to .o
program {EM}/lib/ack/{ARCH}/as
program {EM}/lib/ack/{PLATFORM}/as
args - -o > <
prep cond
end
@ -64,8 +63,9 @@ name led
mapflag -l* LNAME={PLATFORMDIR}/lib*
mapflag -fp FLOATS={EM}/{LIB}fp
args {ALIGN} {SEPID?} \
{PLATFORMDIR}/boot.o \
({RTS}:.ocm.b.c={PLATFORMDIR}/c-ansi.o) \
(.e:{HEAD}={PLATFORMDIR}/boot.o) \
({RTS}:.ocm.b={PLATFORMDIR}/c-ansi.o) \
({RTS}:.c={PLATFORMDIR}/c-ansi.o) \
({RTS}:.mod={PLATFORMDIR}/modula2.o) \
({RTS}:.p={PLATFORMDIR}/pascal.o) \
-o > < \
@ -74,10 +74,10 @@ name led
(.mod:{TAIL}={PLATFORMDIR}/libmodula2.a) \
(.ocm:{TAIL}={PLATFORMDIR}/liboccam.a) \
(.ocm.b.mod.c.p:{TAIL}={PLATFORMDIR}/libc.a) \
{PLATFORMDIR}/liblinux.a \
{PLATFORMDIR}/libem.a \
{PLATFORMDIR}/libsys.a \
{PLATFORMDIR}/libend.a
{FLOATS?} \
(.e:{TAIL}={PLATFORMDIR}/libem.a \
{PLATFORMDIR}/libsys.a \
{PLATFORMDIR}/libend.a)
linker
end
name cv

View file

@ -0,0 +1,24 @@
include("plat/build.lua")
headermap = {}
packagemap = {}
local function addheader(h)
headermap[h] = "./"..h
packagemap["$(PLATIND)/linuxppc/include/"..h] = "./"..h
end
addheader("ack/config.h")
addheader("sys/ioctl.h")
addheader("unistd.h")
acklibrary {
name = "headers",
hdrs = headermap
}
installable {
name = "pkg",
map = packagemap
}

View file

@ -42,8 +42,7 @@ __syscall:
bc IFTRUE, GT, 2f
3:
addis r4, r0, <_errno
ori r4, r4, >_errno
li32 r4, _errno
stw r3, 0(r4)
addi r3, r0, -1
bclr ALWAYS, 0, 0

View file

@ -0,0 +1,16 @@
acklibrary {
name = "lib",
srcs = {
"./*.s",
"plat/linux/libsys/*.c",
"plat/linux/libsys/*.s",
},
deps = {
"lang/cem/libcc.ansi/headers+headers",
"plat/linuxppc/include+headers",
},
vars = {
plat = "linuxppc"
}
}

View file

@ -65,15 +65,13 @@ EUNIMPL = 63 ! unimplemented em-instruction called
addi r4, r0, 1
rlwnm r4, r4, r3, 0, 31 ! calculate trap bit
addis r5, r0, <.ignmask
ori r5, r5, >.ignmask
li32 r5, .ignmask
lwz r5, 0(r5) ! load ignore mask
and. r4, r4, r5 ! compare
bclr IFFALSE, EQ, 0 ! return if non-zero
1:
addis r4, r0, <.trppc
ori r4, r4, >.trppc
li32 r4, .trppc
lwz r5, 0(r4) ! load user trap routine
or. r5, r5, r5 ! test
bc IFTRUE, EQ, fatal ! if no user trap routine, bail out
@ -94,8 +92,7 @@ EUNIMPL = 63 ! unimplemented em-instruction called
fatal:
addi r3, r0, 1
addis r4, r0, <message
ori r4, r4, >message
li32 r4, message
addi r5, r0, 6
addi r0, r0, 4 ! write()
sc 0

View file

@ -140,9 +140,6 @@ showrelo()
case RELOPPC:
printf("\tPowerPC 26-bit address\n");
break;
case RELOH2:
printf("\ttop 2 bytes of a 4 byte word\n");
break;
case RELOVC4:
printf("\tVideoCore IV address in 32-bit instruction\n");
break;

View file

@ -158,12 +158,11 @@ struct outrelo {
/*
* relocation type bits
*/
#define RELSZ 0x0fffi /* relocation length */
#define RELSZ 0x0fff /* relocation length */
#define RELO1 0x01 /* 1 byte */
#define RELO2 0x02 /* 2 bytes */
#define RELO4 0x03 /* 4 bytes */
#define RELOPPC 0x04 /* 26-bit PowerPC address */
#define RELOH2 0x05 /* write top 2 bytes of 4 byte word */
#define RELOVC4 0x06 /* VideoCore IV address in 32-bit insruction */
#define RELPC 0x2000 /* pc relative */
#define RELBR 0x4000 /* High order byte lowest address. */

View file

@ -18,7 +18,7 @@ static char rcsid[] = "$Id$";
#define UBYTE(x) ((x) & BYTEMASK)
static long read2(char* addr, int type)
static uint16_t read2(char* addr, int type)
{
unsigned short word0, word1;
@ -28,7 +28,7 @@ static long read2(char* addr, int type)
return (UBYTE(addr[1]) << WIDTH) + UBYTE(addr[0]);
}
static long read4(char* addr, int type)
static uint32_t read4(char* addr, int type)
{
unsigned short word0, word1;
@ -49,7 +49,7 @@ static long read4(char* addr, int type)
* one of several different ways (depending on what the instruction is).
*/
static long get_vc4_valu(char* addr)
static uint32_t get_vc4_valu(char* addr)
{
uint16_t opcode = read2(addr, 0);
@ -104,11 +104,36 @@ static long get_vc4_valu(char* addr)
assert(0 && "unrecognised VC4 instruction");
}
/* PowerPC fixups are complex as we need to patch up to the next two
* instructions in one of several different ways, depending on what the
* instructions area.
*/
static uint32_t get_powerpc_valu(char* addr, uint16_t type)
{
uint32_t opcode1 = read4(addr+0, type);
uint32_t opcode2 = read4(addr+4, type);
if ((opcode1 & 0xfc000000) == 0x48000000)
{
/* branch instruction */
return opcode1 & 0x03fffffd;
}
else if (((opcode1 & 0xfc1f0000) == 0x3c000000) &&
((opcode2 & 0xfc000000) == 0x60000000))
{
/* addis / ori instruction pair */
return ((opcode1 & 0xffff) << 16) | (opcode2 & 0xffff);
}
assert(0 && "unrecognised PowerPC instruction");
}
/*
* The bits in type indicate how many bytes the value occupies and what
* significance should be attributed to each byte.
*/
static long getvalu(char* addr, uint16_t type)
static uint32_t getvalu(char* addr, uint16_t type)
{
switch (type & RELSZ) {
case RELO1:
@ -118,9 +143,7 @@ static long getvalu(char* addr, uint16_t type)
case RELO4:
return read4(addr, type);
case RELOPPC:
return read4(addr, type) & 0x03FFFFFD;
case RELOH2:
return read2(addr, type) << 16;
return get_powerpc_valu(addr, type);
case RELOVC4:
return get_vc4_valu(addr);
default:
@ -129,7 +152,7 @@ static long getvalu(char* addr, uint16_t type)
/* NOTREACHED */
}
static void write2(long valu, char* addr, int type)
static void write2(uint16_t valu, char* addr, int type)
{
unsigned short word0, word1;
@ -142,7 +165,7 @@ static void write2(long valu, char* addr, int type)
}
}
static void write4(long valu, char* addr, int type)
static void write4(uint32_t valu, char* addr, int type)
{
unsigned short word0, word1;
@ -170,7 +193,7 @@ static void write4(long valu, char* addr, int type)
* one of several different ways (depending on what the instruction is).
*/
static void put_vc4_valu(char* addr, long value)
static void put_vc4_valu(char* addr, uint32_t value)
{
uint16_t opcode = read2(addr, 0);
@ -220,12 +243,42 @@ static void put_vc4_valu(char* addr, long value)
assert(0 && "unrecognised VC4 instruction");
}
/* PowerPC fixups are complex as we need to patch up to the next two
* instructions in one of several different ways, depending on what the
* instructions area.
*/
static void put_powerpc_valu(char* addr, uint32_t value, uint16_t type)
{
uint32_t opcode1 = read4(addr+0, type);
uint32_t opcode2 = read4(addr+4, type);
if ((opcode1 & 0xfc000000) == 0x48000000)
{
/* branch instruction */
uint32_t i = opcode1 & ~0x03fffffd;
i |= value & 0x03fffffd;
write4(i, addr, type);
}
else if (((opcode1 & 0xfc1f0000) == 0x3c000000) &&
((opcode2 & 0xfc000000) == 0x60000000))
{
uint16_t hi = value >> 16;
uint16_t lo = value & 0xffff;
write4((opcode1 & 0xffff0000) | hi, addr+0, type);
write4((opcode2 & 0xffff0000) | lo, addr+4, type);
}
else
assert(0 && "unrecognised PowerPC instruction");
}
/*
* The bits in type indicate how many bytes the value occupies and what
* significance should be attributed to each byte.
* We do not check for overflow.
*/
static putvalu(long valu, char* addr, uint16_t type)
static putvalu(uint32_t valu, char* addr, uint16_t type)
{
switch (type & RELSZ) {
@ -239,14 +292,7 @@ static putvalu(long valu, char* addr, uint16_t type)
write4(valu, addr, type);
break;
case RELOPPC:
{
long i = read4(addr, type) & ~0x03FFFFFD;
i |= valu & 0x03FFFFFD;
write4(i, addr, type);
break;
}
case RELOH2:
write2(valu>>16, addr, type);
put_powerpc_valu(addr, valu, type);
break;
case RELOVC4:
put_vc4_valu(addr, valu);

View file

@ -15,7 +15,7 @@
#define BORS(x,y) y
#endif
#define MAXREGS BORS(80,30)
#define MAXREGS BORS(200,30)
#define MAXPROPS BORS(120,20)
#define MAXTOKENS BORS(100,60)
#define MAXATT 6