diff options
author | Geert Uytterhoeven <geert@linux-m68k.org> | 2011-06-12 11:01:21 +0200 |
---|---|---|
committer | Geert Uytterhoeven <geert@linux-m68k.org> | 2011-07-30 21:21:40 +0200 |
commit | ffe6c42aa3f731f8707e49c39b4b37310ed363e9 (patch) | |
tree | 5feac6358f9bddf949122074f939f3106de24b5f | |
parent | b2cb92417d301f46801695243df5061a9bd31dd5 (diff) |
m68k/math-emu: Remove commented out old code
It's been unused for ages, and contains bugs (e.g. incorrect shifts in
lsl64()).
Reported-by: Jonathan Elchison <jelchison@gmail.com>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
-rw-r--r-- | arch/m68k/math-emu/multi_arith.h | 530 |
1 files changed, 0 insertions, 530 deletions
diff --git a/arch/m68k/math-emu/multi_arith.h b/arch/m68k/math-emu/multi_arith.h index 4ad0ca918e2..4b5eb3d8563 100644 --- a/arch/m68k/math-emu/multi_arith.h +++ b/arch/m68k/math-emu/multi_arith.h @@ -19,246 +19,6 @@ #ifndef MULTI_ARITH_H #define MULTI_ARITH_H -#if 0 /* old code... */ - -/* Unsigned only, because we don't need signs to multiply and divide. */ -typedef unsigned int int128[4]; - -/* Word order */ -enum { - MSW128, - NMSW128, - NLSW128, - LSW128 -}; - -/* big-endian */ -#define LO_WORD(ll) (((unsigned int *) &ll)[1]) -#define HI_WORD(ll) (((unsigned int *) &ll)[0]) - -/* Convenience functions to stuff various integer values into int128s */ - -static inline void zero128(int128 a) -{ - a[LSW128] = a[NLSW128] = a[NMSW128] = a[MSW128] = 0; -} - -/* Human-readable word order in the arguments */ -static inline void set128(unsigned int i3, unsigned int i2, unsigned int i1, - unsigned int i0, int128 a) -{ - a[LSW128] = i0; - a[NLSW128] = i1; - a[NMSW128] = i2; - a[MSW128] = i3; -} - -/* Convenience functions (for testing as well) */ -static inline void int64_to_128(unsigned long long src, int128 dest) -{ - dest[LSW128] = (unsigned int) src; - dest[NLSW128] = src >> 32; - dest[NMSW128] = dest[MSW128] = 0; -} - -static inline void int128_to_64(const int128 src, unsigned long long *dest) -{ - *dest = src[LSW128] | (long long) src[NLSW128] << 32; -} - -static inline void put_i128(const int128 a) -{ - printk("%08x %08x %08x %08x\n", a[MSW128], a[NMSW128], - a[NLSW128], a[LSW128]); -} - -/* Internal shifters: - - Note that these are only good for 0 < count < 32. - */ - -static inline void _lsl128(unsigned int count, int128 a) -{ - a[MSW128] = (a[MSW128] << count) | (a[NMSW128] >> (32 - count)); - a[NMSW128] = (a[NMSW128] << count) | (a[NLSW128] >> (32 - count)); - a[NLSW128] = (a[NLSW128] << count) | (a[LSW128] >> (32 - count)); - a[LSW128] <<= count; -} - -static inline void _lsr128(unsigned int count, int128 a) -{ - a[LSW128] = (a[LSW128] >> count) | (a[NLSW128] << (32 - count)); - a[NLSW128] = (a[NLSW128] >> count) | (a[NMSW128] << (32 - count)); - a[NMSW128] = (a[NMSW128] >> count) | (a[MSW128] << (32 - count)); - a[MSW128] >>= count; -} - -/* Should be faster, one would hope */ - -static inline void lslone128(int128 a) -{ - asm volatile ("lsl.l #1,%0\n" - "roxl.l #1,%1\n" - "roxl.l #1,%2\n" - "roxl.l #1,%3\n" - : - "=d" (a[LSW128]), - "=d"(a[NLSW128]), - "=d"(a[NMSW128]), - "=d"(a[MSW128]) - : - "0"(a[LSW128]), - "1"(a[NLSW128]), - "2"(a[NMSW128]), - "3"(a[MSW128])); -} - -static inline void lsrone128(int128 a) -{ - asm volatile ("lsr.l #1,%0\n" - "roxr.l #1,%1\n" - "roxr.l #1,%2\n" - "roxr.l #1,%3\n" - : - "=d" (a[MSW128]), - "=d"(a[NMSW128]), - "=d"(a[NLSW128]), - "=d"(a[LSW128]) - : - "0"(a[MSW128]), - "1"(a[NMSW128]), - "2"(a[NLSW128]), - "3"(a[LSW128])); -} - -/* Generalized 128-bit shifters: - - These bit-shift to a multiple of 32, then move whole longwords. */ - -static inline void lsl128(unsigned int count, int128 a) -{ - int wordcount, i; - - if (count % 32) - _lsl128(count % 32, a); - - if (0 == (wordcount = count / 32)) - return; - - /* argh, gak, endian-sensitive */ - for (i = 0; i < 4 - wordcount; i++) { - a[i] = a[i + wordcount]; - } - for (i = 3; i >= 4 - wordcount; --i) { - a[i] = 0; - } -} - -static inline void lsr128(unsigned int count, int128 a) -{ - int wordcount, i; - - if (count % 32) - _lsr128(count % 32, a); - - if (0 == (wordcount = count / 32)) - return; - - for (i = 3; i >= wordcount; --i) { - a[i] = a[i - wordcount]; - } - for (i = 0; i < wordcount; i++) { - a[i] = 0; - } -} - -static inline int orl128(int a, int128 b) -{ - b[LSW128] |= a; -} - -static inline int btsthi128(const int128 a) -{ - return a[MSW128] & 0x80000000; -} - -/* test bits (numbered from 0 = LSB) up to and including "top" */ -static inline int bftestlo128(int top, const int128 a) -{ - int r = 0; - - if (top > 31) - r |= a[LSW128]; - if (top > 63) - r |= a[NLSW128]; - if (top > 95) - r |= a[NMSW128]; - - r |= a[3 - (top / 32)] & ((1 << (top % 32 + 1)) - 1); - - return (r != 0); -} - -/* Aargh. We need these because GCC is broken */ -/* FIXME: do them in assembly, for goodness' sake! */ -static inline void mask64(int pos, unsigned long long *mask) -{ - *mask = 0; - - if (pos < 32) { - LO_WORD(*mask) = (1 << pos) - 1; - return; - } - LO_WORD(*mask) = -1; - HI_WORD(*mask) = (1 << (pos - 32)) - 1; -} - -static inline void bset64(int pos, unsigned long long *dest) -{ - /* This conditional will be optimized away. Thanks, GCC! */ - if (pos < 32) - asm volatile ("bset %1,%0":"=m" - (LO_WORD(*dest)):"id"(pos)); - else - asm volatile ("bset %1,%0":"=m" - (HI_WORD(*dest)):"id"(pos - 32)); -} - -static inline int btst64(int pos, unsigned long long dest) -{ - if (pos < 32) - return (0 != (LO_WORD(dest) & (1 << pos))); - else - return (0 != (HI_WORD(dest) & (1 << (pos - 32)))); -} - -static inline void lsl64(int count, unsigned long long *dest) -{ - if (count < 32) { - HI_WORD(*dest) = (HI_WORD(*dest) << count) - | (LO_WORD(*dest) >> count); - LO_WORD(*dest) <<= count; - return; - } - count -= 32; - HI_WORD(*dest) = LO_WORD(*dest) << count; - LO_WORD(*dest) = 0; -} - -static inline void lsr64(int count, unsigned long long *dest) -{ - if (count < 32) { - LO_WORD(*dest) = (LO_WORD(*dest) >> count) - | (HI_WORD(*dest) << (32 - count)); - HI_WORD(*dest) >>= count; - return; - } - count -= 32; - LO_WORD(*dest) = HI_WORD(*dest) >> count; - HI_WORD(*dest) = 0; -} -#endif - static inline void fp_denormalize(struct fp_ext *reg, unsigned int cnt) { reg->exp += cnt; @@ -481,117 +241,6 @@ static inline void fp_dividemant(union fp_mant128 *dest, struct fp_ext *src, } } -#if 0 -static inline unsigned int fp_fls128(union fp_mant128 *src) -{ - unsigned long data; - unsigned int res, off; - - if ((data = src->m32[0])) - off = 0; - else if ((data = src->m32[1])) - off = 32; - else if ((data = src->m32[2])) - off = 64; - else if ((data = src->m32[3])) - off = 96; - else - return 128; - - asm ("bfffo %1{#0,#32},%0" : "=d" (res) : "dm" (data)); - return res + off; -} - -static inline void fp_shiftmant128(union fp_mant128 *src, int shift) -{ - unsigned long sticky; - - switch (shift) { - case 0: - return; - case 1: - asm volatile ("lsl.l #1,%0" - : "=d" (src->m32[3]) : "0" (src->m32[3])); - asm volatile ("roxl.l #1,%0" - : "=d" (src->m32[2]) : "0" (src->m32[2])); - asm volatile ("roxl.l #1,%0" - : "=d" (src->m32[1]) : "0" (src->m32[1])); - asm volatile ("roxl.l #1,%0" - : "=d" (src->m32[0]) : "0" (src->m32[0])); - return; - case 2 ... 31: - src->m32[0] = (src->m32[0] << shift) | (src->m32[1] >> (32 - shift)); - src->m32[1] = (src->m32[1] << shift) | (src->m32[2] >> (32 - shift)); - src->m32[2] = (src->m32[2] << shift) | (src->m32[3] >> (32 - shift)); - src->m32[3] = (src->m32[3] << shift); - return; - case 32 ... 63: - shift -= 32; - src->m32[0] = (src->m32[1] << shift) | (src->m32[2] >> (32 - shift)); - src->m32[1] = (src->m32[2] << shift) | (src->m32[3] >> (32 - shift)); - src->m32[2] = (src->m32[3] << shift); - src->m32[3] = 0; - return; - case 64 ... 95: - shift -= 64; - src->m32[0] = (src->m32[2] << shift) | (src->m32[3] >> (32 - shift)); - src->m32[1] = (src->m32[3] << shift); - src->m32[2] = src->m32[3] = 0; - return; - case 96 ... 127: - shift -= 96; - src->m32[0] = (src->m32[3] << shift); - src->m32[1] = src->m32[2] = src->m32[3] = 0; - return; - case -31 ... -1: - shift = -shift; - sticky = 0; - if (src->m32[3] << (32 - shift)) - sticky = 1; - src->m32[3] = (src->m32[3] >> shift) | (src->m32[2] << (32 - shift)) | sticky; - src->m32[2] = (src->m32[2] >> shift) | (src->m32[1] << (32 - shift)); - src->m32[1] = (src->m32[1] >> shift) | (src->m32[0] << (32 - shift)); - src->m32[0] = (src->m32[0] >> shift); - return; - case -63 ... -32: - shift = -shift - 32; - sticky = 0; - if ((src->m32[2] << (32 - shift)) || src->m32[3]) - sticky = 1; - src->m32[3] = (src->m32[2] >> shift) | (src->m32[1] << (32 - shift)) | sticky; - src->m32[2] = (src->m32[1] >> shift) | (src->m32[0] << (32 - shift)); - src->m32[1] = (src->m32[0] >> shift); - src->m32[0] = 0; - return; - case -95 ... -64: - shift = -shift - 64; - sticky = 0; - if ((src->m32[1] << (32 - shift)) || src->m32[2] || src->m32[3]) - sticky = 1; - src->m32[3] = (src->m32[1] >> shift) | (src->m32[0] << (32 - shift)) | sticky; - src->m32[2] = (src->m32[0] >> shift); - src->m32[1] = src->m32[0] = 0; - return; - case -127 ... -96: - shift = -shift - 96; - sticky = 0; - if ((src->m32[0] << (32 - shift)) || src->m32[1] || src->m32[2] || src->m32[3]) - sticky = 1; - src->m32[3] = (src->m32[0] >> shift) | sticky; - src->m32[2] = src->m32[1] = src->m32[0] = 0; - return; - } - - if (shift < 0 && (src->m32[0] || src->m32[1] || src->m32[2] || src->m32[3])) - src->m32[3] = 1; - else - src->m32[3] = 0; - src->m32[2] = 0; - src->m32[1] = 0; - src->m32[0] = 0; -} -#endif - static inline void fp_putmant128(struct fp_ext *dest, union fp_mant128 *src, int shift) { @@ -637,183 +286,4 @@ static inline void fp_putmant128(struct fp_ext *dest, union fp_mant128 *src, } } -#if 0 /* old code... */ -static inline int fls(unsigned int a) -{ - int r; - - asm volatile ("bfffo %1{#0,#32},%0" - : "=d" (r) : "md" (a)); - return r; -} - -/* fls = "find last set" (cf. ffs(3)) */ -static inline int fls128(const int128 a) -{ - if (a[MSW128]) - return fls(a[MSW128]); - if (a[NMSW128]) - return fls(a[NMSW128]) + 32; - /* XXX: it probably never gets beyond this point in actual - use, but that's indicative of a more general problem in the - algorithm (i.e. as per the actual 68881 implementation, we - really only need at most 67 bits of precision [plus - overflow]) so I'm not going to fix it. */ - if (a[NLSW128]) - return fls(a[NLSW128]) + 64; - if (a[LSW128]) - return fls(a[LSW128]) + 96; - else - return -1; -} - -static inline int zerop128(const int128 a) -{ - return !(a[LSW128] | a[NLSW128] | a[NMSW128] | a[MSW128]); -} - -static inline int nonzerop128(const int128 a) -{ - return (a[LSW128] | a[NLSW128] | a[NMSW128] | a[MSW128]); -} - -/* Addition and subtraction */ -/* Do these in "pure" assembly, because "extended" asm is unmanageable - here */ -static inline void add128(const int128 a, int128 b) -{ - /* rotating carry flags */ - unsigned int carry[2]; - - carry[0] = a[LSW128] > (0xffffffff - b[LSW128]); - b[LSW128] += a[LSW128]; - - carry[1] = a[NLSW128] > (0xffffffff - b[NLSW128] - carry[0]); - b[NLSW128] = a[NLSW128] + b[NLSW128] + carry[0]; - - carry[0] = a[NMSW128] > (0xffffffff - b[NMSW128] - carry[1]); - b[NMSW128] = a[NMSW128] + b[NMSW128] + carry[1]; - - b[MSW128] = a[MSW128] + b[MSW128] + carry[0]; -} - -/* Note: assembler semantics: "b -= a" */ -static inline void sub128(const int128 a, int128 b) -{ - /* rotating borrow flags */ - unsigned int borrow[2]; - - borrow[0] = b[LSW128] < a[LSW128]; - b[LSW128] -= a[LSW128]; - - borrow[1] = b[NLSW128] < a[NLSW128] + borrow[0]; - b[NLSW128] = b[NLSW128] - a[NLSW128] - borrow[0]; - - borrow[0] = b[NMSW128] < a[NMSW128] + borrow[1]; - b[NMSW128] = b[NMSW128] - a[NMSW128] - borrow[1]; - - b[MSW128] = b[MSW128] - a[MSW128] - borrow[0]; -} - -/* Poor man's 64-bit expanding multiply */ -static inline void mul64(unsigned long long a, unsigned long long b, int128 c) -{ - unsigned long long acc; - int128 acc128; - - zero128(acc128); - zero128(c); - - /* first the low words */ - if (LO_WORD(a) && LO_WORD(b)) { - acc = (long long) LO_WORD(a) * LO_WORD(b); - c[NLSW128] = HI_WORD(acc); - c[LSW128] = LO_WORD(acc); - } - /* Next the high words */ - if (HI_WORD(a) && HI_WORD(b)) { - acc = (long long) HI_WORD(a) * HI_WORD(b); - c[MSW128] = HI_WORD(acc); - c[NMSW128] = LO_WORD(acc); - } - /* The middle words */ - if (LO_WORD(a) && HI_WORD(b)) { - acc = (long long) LO_WORD(a) * HI_WORD(b); - acc128[NMSW128] = HI_WORD(acc); - acc128[NLSW128] = LO_WORD(acc); - add128(acc128, c); - } - /* The first and last words */ - if (HI_WORD(a) && LO_WORD(b)) { - acc = (long long) HI_WORD(a) * LO_WORD(b); - acc128[NMSW128] = HI_WORD(acc); - acc128[NLSW128] = LO_WORD(acc); - add128(acc128, c); - } -} - -/* Note: unsigned */ -static inline int cmp128(int128 a, int128 b) -{ - if (a[MSW128] < b[MSW128]) - return -1; - if (a[MSW128] > b[MSW128]) - return 1; - if (a[NMSW128] < b[NMSW128]) - return -1; - if (a[NMSW128] > b[NMSW128]) - return 1; - if (a[NLSW128] < b[NLSW128]) - return -1; - if (a[NLSW128] > b[NLSW128]) - return 1; - - return (signed) a[LSW128] - b[LSW128]; -} - -inline void div128(int128 a, int128 b, int128 c) -{ - int128 mask; - - /* Algorithm: - - Shift the divisor until it's at least as big as the - dividend, keeping track of the position to which we've - shifted it, i.e. the power of 2 which we've multiplied it - by. - - Then, for this power of 2 (the mask), and every one smaller - than it, subtract the mask from the dividend and add it to - the quotient until the dividend is smaller than the raised - divisor. At this point, divide the dividend and the mask - by 2 (i.e. shift one place to the right). Lather, rinse, - and repeat, until there are no more powers of 2 left. */ - - /* FIXME: needless to say, there's room for improvement here too. */ - - /* Shift up */ - /* XXX: since it just has to be "at least as big", we can - probably eliminate this horribly wasteful loop. I will - have to prove this first, though */ - set128(0, 0, 0, 1, mask); - while (cmp128(b, a) < 0 && !btsthi128(b)) { - lslone128(b); - lslone128(mask); - } - - /* Shift down */ - zero128(c); - do { - if (cmp128(a, b) >= 0) { - sub128(b, a); - add128(mask, c); - } - lsrone128(mask); - lsrone128(b); - } while (nonzerop128(mask)); - - /* The remainder is in a... */ -} -#endif - #endif /* MULTI_ARITH_H */ |