diff options
Diffstat (limited to 'mdk-stage1/dietlibc/sparc')
-rw-r--r-- | mdk-stage1/dietlibc/sparc/Makefile.add | 3 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/__longjmp.S | 66 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/fork.S | 19 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/mmap.c | 43 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/pipe.S | 20 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/setjmp.S | 35 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/start.S | 46 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/udiv.S | 363 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/umul.S | 170 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/unified.S | 28 | ||||
-rw-r--r-- | mdk-stage1/dietlibc/sparc/urem.S | 362 |
11 files changed, 1155 insertions, 0 deletions
diff --git a/mdk-stage1/dietlibc/sparc/Makefile.add b/mdk-stage1/dietlibc/sparc/Makefile.add new file mode 100644 index 000000000..a12a0446a --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/Makefile.add @@ -0,0 +1,3 @@ + +CFLAGS+=-mcpu=supersparc -Os +override VPATH=sparc:syscalls.s:lib diff --git a/mdk-stage1/dietlibc/sparc/__longjmp.S b/mdk-stage1/dietlibc/sparc/__longjmp.S new file mode 100644 index 000000000..81dd24af2 --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/__longjmp.S @@ -0,0 +1,66 @@ +#define _ASM +#define _SETJMP_H +#include <bits/setjmp.h> + + +#define ENV(base,reg) [%base + (reg * 4)] +#define ST_FLUSH_WINDOWS 3 +#define RW_FP [%fp + 0x48] + +.text +.global __longjmp +__longjmp: + /* Store our arguments in global registers so we can still + use them while unwinding frames and their register windows. */ + + ld ENV(o0,JB_FP), %g3 /* Cache target FP in register %g3. */ + mov %o0, %g1 /* ENV in %g1 */ + orcc %o1, %g0, %g2 /* VAL in %g2 */ + be,a 0f /* Branch if zero; else skip delay slot. */ + mov 1, %g2 /* Delay slot only hit if zero: VAL = 1. */ +0: + xor %fp, %g3, %o0 + add %fp, 512, %o1 + andncc %o0, 4095, %o0 + bne .Lthread + cmp %o1, %g3 + bl .Lthread + + /* Now we will loop, unwinding the register windows up the stack + until the restored %fp value matches the target value in %g3. */ + +.Lloop: + cmp %fp, %g3 /* Have we reached the target frame? */ + bl,a .Lloop /* Loop while current fp is below target. */ + restore /* Unwind register window in delay slot. */ + be,a .Lfound /* Better have hit it exactly. */ + ld ENV(g1,JB_SP), %o0 /* Delay slot: extract target SP. */ + +.Lthread: + /* + * Do a "flush register windows trap". The trap handler in the + * kernel writes all the register windows to their stack slots, and + * marks them all as invalid (needing to be sucked up from the + * stack when used). This ensures that all information needed to + * unwind to these callers is in memory, not in the register + * windows. + */ + ta ST_FLUSH_WINDOWS + ld ENV(g1,JB_PC), %o7 /* Set return PC. */ + ld ENV(g1,JB_SP), %fp /* Set saved SP on restore below. */ + sub %fp, 64, %sp /* Allocate a register frame. */ + st %g3, RW_FP /* Set saved FP on restore below. */ + retl + restore %g2, 0, %o0 /* Restore values from above register frame. */ + +.Lfound: + /* We have unwound register windows so %fp matches the target. */ + mov %o0, %sp /* OK, install new SP. */ + +.Lsp_ok: + ld ENV(g1,JB_PC), %o0 /* Extract target return PC. */ + jmp %o0 + 8 /* Return there. */ + mov %g2, %o0 /* Delay slot: set return value. */ + +.size __longjmp, . - __longjmp + diff --git a/mdk-stage1/dietlibc/sparc/fork.S b/mdk-stage1/dietlibc/sparc/fork.S new file mode 100644 index 000000000..150839971 --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/fork.S @@ -0,0 +1,19 @@ +#include "syscalls.h" + +.text +.global fork +fork: + mov 2, %g1 + ta 0x10 + bcc,a 1f + nop + + sethi %hi(errno), %o3 + or %o3, %lo(errno), %o3 + st %i0, [%o3] + + retl + mov -1, %o0 +1: dec %o1 + retl + and %o0, %o1, %o0 diff --git a/mdk-stage1/dietlibc/sparc/mmap.c b/mdk-stage1/dietlibc/sparc/mmap.c new file mode 100644 index 000000000..25ebdc24e --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/mmap.c @@ -0,0 +1,43 @@ +#include <linux/types.h> +#include <linux/unistd.h> + +#define __SYSCALL_STRING \ + "ta 0x10;" \ + "bcs 2f;" \ + " nop;" \ + "1:" \ + ".subsection 2;" \ + "2:" \ + "save %%sp, -192, %%sp;" \ + "call __errno_location;" \ + " nop;" \ + "st %%i0,[%%o0];" \ + "ba 1b;" \ + " restore %%g0, -1, %%o0;" \ + ".previous;" + +#define __SYSCALL_CLOBBERS "g2", "g3", "g4", "g5", "g7", \ + "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \ + "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", \ + "f16", "f17", "f18", "f19", "f20", "f21", "f22", "f23", \ + "f24", "f25", "f26", "f27", "f28", "f29", "f30", "f31", \ + "cc", "memory" + +#define inline_syscall6(name,arg1,arg2,arg3,arg4,arg5,arg6) \ +({ \ + register long __o0 __asm__ ("o0") = (long)(arg1); \ + register long __o1 __asm__ ("o1") = (long)(arg2); \ + register long __o2 __asm__ ("o2") = (long)(arg3); \ + register long __o3 __asm__ ("o3") = (long)(arg4); \ + register long __o4 __asm__ ("o4") = (long)(arg5); \ + register long __o5 __asm__ ("o5") = (long)(arg6); \ + register long __g1 __asm__ ("g1") = __NR_##name; \ + __asm__ (__SYSCALL_STRING : "=r" (__g1), "=r" (__o0) : \ + "0" (__g1), "1" (__o0), "r" (__o1), "r" (__o2), \ + "r" (__o3), "r" (__o4), "r" (__o5) : \ + __SYSCALL_CLOBBERS); \ + __o0; \ +}) +int mmap(void*start,size_t length,int prot,int flags,int fd,off_t offset) { + return inline_syscall6(mmap,start,length,prot,flags,fd,offset); +} diff --git a/mdk-stage1/dietlibc/sparc/pipe.S b/mdk-stage1/dietlibc/sparc/pipe.S new file mode 100644 index 000000000..b8ad1d251 --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/pipe.S @@ -0,0 +1,20 @@ +#include "syscalls.h" + +.text +.global pipe +pipe: + mov __NR_pipe, %g1 + ta 0x10 + bcc,a .Lnoerror + nop + save %sp, -96, %sp + call __errno_location + nop + st %i0, [ %o0 ] + ret + restore %g0, -1, %o0 +.Lnoerror: + st %o0, [ %o2 ] + st %o1, [ %o2 + 4 ] + retl + mov %g0, %o0 diff --git a/mdk-stage1/dietlibc/sparc/setjmp.S b/mdk-stage1/dietlibc/sparc/setjmp.S new file mode 100644 index 000000000..be77af3c4 --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/setjmp.S @@ -0,0 +1,35 @@ +#define _ASM +#define _SETJMP_H +#include <bits/setjmp.h> + +#define ST_FLUSH_WINDOWS 0x03 + +.section .rodata +.text +.globl __setjmp +__setjmp: + b 1f + set 0, %o1 +.size __setjmp,.-__setjmp + +.globl setjmp +setjmp: + set 1, %o1 +.size setjmp,.-setjmp + +.globl __sigsetjmp +__sigsetjmp: +1: + /* Save our PC, SP and FP. Save the signal mask if requested with + a tail-call for simplicity; it always returns zero. */ + ta ST_FLUSH_WINDOWS + + st %o7, [%o0 + (JB_PC * 4)] + st %sp, [%o0 + (JB_SP * 4)] + st %fp, [%o0 + (JB_FP * 4)] + + mov %o7, %g1 + call __sigjmp_save + mov %g1, %o7 +.size __sigsetjmp,.-__sigsetjmp + diff --git a/mdk-stage1/dietlibc/sparc/start.S b/mdk-stage1/dietlibc/sparc/start.S new file mode 100644 index 000000000..e948aaddc --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/start.S @@ -0,0 +1,46 @@ +#include "start.h" + +#ifdef __sparc__ + .section ".text" + .align 4 + .global _start + .type _start,@function +_start: + + /* Terminate the stack frame, and reserve space for functions to + drop their arguments. */ + mov %g0, %fp + sub %sp, 6*4, %sp + + /* Extract the arguments and environment as encoded on the stack. The + argument info starts after one register window (16 words) past the SP. */ + ld [%sp+22*4], %o0 + add %sp, 23*4, %o1 + add %o1, %o0, %o2 + add %o2, %o0, %o2 + add %o2, %o0, %o2 + add %o2, %o0, %o2 + add %o2, 4, %o2 + + sethi %hi(environ), %o3 + or %o3, %lo(environ), %o3 + st %o2, [%o3] + + /* When starting a binary via the dynamic linker, %g1 contains the + address of the shared library termination function, which will be + registered with atexit(). If we are statically linked, this will + be NULL. */ + + /* Let libc do the rest of the initialization, and call main. */ + call main + mov %g1, %o5 + + b exit + mov %o0, %i0 + + /* Die very horribly if exit returns. */ + unimp + + .size _start, .-_start +#endif + diff --git a/mdk-stage1/dietlibc/sparc/udiv.S b/mdk-stage1/dietlibc/sparc/udiv.S new file mode 100644 index 000000000..87479e7bd --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/udiv.S @@ -0,0 +1,363 @@ +#ifdef __sparc__ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .udiv name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#define C_LABEL(name) name: + +#define C_SYMBOL_NAME(name) name + +#define ENTRY(name) \ + .global C_SYMBOL_NAME(name); \ + .align 4;\ + C_LABEL(name);\ + .type name,@function; + +#define LOC(name) . ## L ## name + +#define END(name) \ + .size name, . - name + +#define ST_DIV0 0x02 + +ENTRY(.udiv) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu LOC(got_result) ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu LOC(not_really_big) + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g2 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g2. + 2: addcc %o5, %o5, %o5 + bcc LOC(not_too_big) + add %g2, 1, %g2 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b LOC(do_single_div) + sub %g2, 1, %g2 + + LOC(not_too_big): + 3: cmp %o5, %o3 + blu 2b + nop + be LOC(do_single_div) + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g2 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + LOC(do_single_div): + subcc %g2, 1, %g2 + bl LOC(end_regular_divide) + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b LOC(end_single_divloop) + nop + LOC(single_divloop): + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + LOC(end_single_divloop): + subcc %g2, 1, %g2 + bge LOC(single_divloop) + tst %o3 + b,a LOC(end_regular_divide) + +LOC(not_really_big): +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be LOC(got_result) + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +LOC(divloop): + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl LOC(1.16) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl LOC(2.17) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl LOC(3.19) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl LOC(4.23) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +LOC(4.23): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +LOC(3.19): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl LOC(4.21) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +LOC(4.21): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +LOC(2.17): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl LOC(3.17) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl LOC(4.19) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +LOC(4.19): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +LOC(3.17): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl LOC(4.17) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +LOC(4.17): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +LOC(1.16): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl LOC(2.15) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl LOC(3.15) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl LOC(4.15) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +LOC(4.15): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +LOC(3.15): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl LOC(4.13) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +LOC(4.13): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +LOC(2.15): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl LOC(3.13) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl LOC(4.11) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +LOC(4.11): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +LOC(3.13): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl LOC(4.9) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +LOC(4.9): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +LOC(end_regular_divide): + subcc %o4, 1, %o4 + bge LOC(divloop) + tst %o3 + bl,a LOC(got_result) + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +LOC(got_result): + + retl + mov %o2, %o0 + +END(.udiv) + +#endif diff --git a/mdk-stage1/dietlibc/sparc/umul.S b/mdk-stage1/dietlibc/sparc/umul.S new file mode 100644 index 000000000..15038ab2a --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/umul.S @@ -0,0 +1,170 @@ +#ifdef __sparc__ +/* + * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the + * upper 32 bits of the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. Short + * multiplies require 25 instruction cycles, and long ones require + * 45 instruction cycles. + * + * On return, overflow has occurred (%o1 is not zero) if and only if + * the Z condition code is clear, allowing, e.g., the following: + * + * call .umul + * nop + * bnz overflow (or tnz) + */ + +#define C_LABEL(name) name: + +#define C_SYMBOL_NAME(name) name + +#define ENTRY(name) \ + .global C_SYMBOL_NAME(name); \ + .align 4;\ + C_LABEL(name);\ + .type name,@function; + +#define LOC(name) . ## L ## name + +#define END(name) \ + .size name, . - name + +ENTRY(.umul) + or %o0, %o1, %o4 + mov %o0, %y ! multiplier -> Y + andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args + be LOC(mul_shortway) ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product; clear N & V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + /* + * Normally, with the shift-and-add approach, if both numbers are + * positive you get the correct result. With 32-bit two's-complement + * numbers, -x is represented as + * + * x 32 + * ( 2 - ------ ) mod 2 * 2 + * 32 + * 2 + * + * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s, + * we can treat this as if the radix point were just to the left + * of the sign bit (multiply by 2^32), and get + * + * -x = (2 - x) mod 2 + * + * Then, ignoring the `mod 2's for convenience: + * + * x * y = xy + * -x * y = 2y - xy + * x * -y = 2x - xy + * -x * -y = 4 - 2x - 2y + xy + * + * For signed multiplies, we subtract (x << 32) from the partial + * product to fix this problem for negative multipliers (see mul.s). + * Because of the way the shift into the partial product is calculated + * (N xor V), this term is automatically removed for the multiplicand, + * so we don't have to adjust. + * + * But for unsigned multiplies, the high order bit wasn't a sign bit, + * and the correction is wrong. So for unsigned multiplies where the + * high order bit is one, we end up with xy - (y << 32). To fix it + * we add y << 32. + */ +#if 0 + tst %o1 + bl,a 1f ! if %o1 < 0 (high order bit = 1), + add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half) +1: rd %y, %o0 ! get lower half of product + retl + addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0 +#else + /* Faster code from tege@sics.se. */ + sra %o1, 31, %o2 ! make mask from sign bit + and %o0, %o2, %o2 ! %o2 = 0 or %o0, depending on sign of %o1 + rd %y, %o0 ! get lower half of product + retl + addcc %o4, %o2, %o1 ! add compensation and put upper half in place +#endif + +LOC(mul_shortway): + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above), + * and overflow is impossible (the answer is at most 24 bits long). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the result; %y has + * the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * -----result----- + * + * The 12 bits of %o4 left of the `result' area are all zero; + * in fact, all top 20 bits of %o4 are zero. + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20 + or %o5, %o0, %o0 + retl + addcc %g0, %g0, %o1 ! %o1 = zero, and set Z + +END(.umul) +#endif diff --git a/mdk-stage1/dietlibc/sparc/unified.S b/mdk-stage1/dietlibc/sparc/unified.S new file mode 100644 index 000000000..53b1612a9 --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/unified.S @@ -0,0 +1,28 @@ +#include <dietfeatures.h> + +.text +.global __unified_syscall +__unified_syscall: + ta 0x10 + + bcc 1f + save %sp, -104, %sp + + neg %i0, %i0 +1: + add %i0, 0xff, %l2 + cmp %l2, 0xfe + bgu 2f + neg %i0, %l3 +#ifdef WANT_THREAD_SAVE + call __errno_location + nop +#else + sethi %hi(errno), %o0 + or %o0, %lo(errno), %o0 +#endif + st %l3, [ %o0 ] + mov -1, %o0 +2: + ret + restore diff --git a/mdk-stage1/dietlibc/sparc/urem.S b/mdk-stage1/dietlibc/sparc/urem.S new file mode 100644 index 000000000..943cb7873 --- /dev/null +++ b/mdk-stage1/dietlibc/sparc/urem.S @@ -0,0 +1,362 @@ +#ifdef __sparc__ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .urem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#define C_LABEL(name) name: + +#define C_SYMBOL_NAME(name) name + +#define ENTRY(name) \ + .global C_SYMBOL_NAME(name); \ + .align 4;\ + C_LABEL(name);\ + .type name,@function; + +#define LOC(name) . ## L ## name + +#define END(name) \ + .size name, . - name + +#define ST_DIV0 0x02 + +ENTRY(.urem) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu LOC(got_result) ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu LOC(not_really_big) + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g2 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g2. + 2: addcc %o5, %o5, %o5 + bcc LOC(not_too_big) + add %g2, 1, %g2 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b LOC(do_single_div) + sub %g2, 1, %g2 + + LOC(not_too_big): + 3: cmp %o5, %o3 + blu 2b + nop + be LOC(do_single_div) + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g2 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + LOC(do_single_div): + subcc %g2, 1, %g2 + bl LOC(end_regular_divide) + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b LOC(end_single_divloop) + nop + LOC(single_divloop): + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + LOC(end_single_divloop): + subcc %g2, 1, %g2 + bge LOC(single_divloop) + tst %o3 + b,a LOC(end_regular_divide) + +LOC(not_really_big): +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be LOC(got_result) + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +LOC(divloop): + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl LOC(1.16) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl LOC(2.17) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl LOC(3.19) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl LOC(4.23) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +LOC(4.23): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +LOC(3.19): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl LOC(4.21) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +LOC(4.21): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +LOC(2.17): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl LOC(3.17) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl LOC(4.19) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +LOC(4.19): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +LOC(3.17): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl LOC(4.17) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +LOC(4.17): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +LOC(1.16): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl LOC(2.15) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl LOC(3.15) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl LOC(4.15) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +LOC(4.15): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +LOC(3.15): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl LOC(4.13) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +LOC(4.13): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +LOC(2.15): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl LOC(3.13) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl LOC(4.11) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +LOC(4.11): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +LOC(3.13): + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl LOC(4.9) + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +LOC(4.9): + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +LOC(end_regular_divide): + subcc %o4, 1, %o4 + bge LOC(divloop) + tst %o3 + bl,a LOC(got_result) + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + + +LOC(got_result): + + retl + mov %o3, %o0 + +END(.urem) +#endif |