diff options
Diffstat (limited to 'libc/string/ia64/memcmp.S')
| -rw-r--r-- | libc/string/ia64/memcmp.S | 94 |
1 files changed, 47 insertions, 47 deletions
diff --git a/libc/string/ia64/memcmp.S b/libc/string/ia64/memcmp.S index 8b0c096ce..adb1a20de 100644 --- a/libc/string/ia64/memcmp.S +++ b/libc/string/ia64/memcmp.S @@ -28,7 +28,7 @@ In this form, it assumes little endian mode. For big endian mode, the the two shifts in .l2 must be inverted: - shl tmp1[0] = r[1 + MEMLAT], sh1 // tmp1 = w0 << sh1 + shl tmp1[0] = r[1 + MEMLAT], sh1 // tmp1 = w0 << sh1 shr.u tmp2[0] = r[0 + MEMLAT], sh2 // tmp2 = w1 >> sh2 and all the mux1 instructions should be replaced by plain mov's. */ @@ -36,8 +36,8 @@ #include "sysdep.h" #undef ret -#define OP_T_THRES 16 -#define OPSIZ 8 +#define OP_T_THRES 16 +#define OPSIZ 8 #define MEMLAT 2 #define start r15 @@ -56,85 +56,85 @@ ENTRY(memcmp) .prologue - alloc r2 = ar.pfs, 3, 37, 0, 40 + alloc r2 = ar.pfs, 3, 37, 0, 40 .rotr r[MEMLAT + 2], q[MEMLAT + 5], tmp1[4], tmp2[4], val[2] .rotp p[MEMLAT + 4 + 1] - mov ret0 = r0 // by default return value = 0 + mov ret0 = r0 /* by default return value = 0 */ .save pr, saved_pr - mov saved_pr = pr // save the predicate registers + mov saved_pr = pr /* save the predicate registers */ .save ar.lc, saved_lc - mov saved_lc = ar.lc // save the loop counter + mov saved_lc = ar.lc /* save the loop counter */ .body - mov dest = in0 // dest - mov src = in1 // src - mov len = in2 // len - sub tmp = r0, in0 // tmp = -dest + mov dest = in0 /* dest */ + mov src = in1 /* src */ + mov len = in2 /* len */ + sub tmp = r0, in0 /* tmp = -dest */ ;; - and loopcnt = 7, tmp // loopcnt = -dest % 8 - cmp.ge p6, p0 = OP_T_THRES, len // is len <= OP_T_THRES -(p6) br.cond.spnt .cmpfew // compare byte by byte + and loopcnt = 7, tmp /* loopcnt = -dest % 8 */ + cmp.ge p6, p0 = OP_T_THRES, len /* is len <= OP_T_THRES */ +(p6) br.cond.spnt .cmpfew /* compare byte by byte */ ;; cmp.eq p6, p0 = loopcnt, r0 (p6) br.cond.sptk .dest_aligned - sub len = len, loopcnt // len -= -dest % 8 - adds loopcnt = -1, loopcnt // --loopcnt + sub len = len, loopcnt /* len -= -dest % 8 */ + adds loopcnt = -1, loopcnt /* --loopcnt */ ;; mov ar.lc = loopcnt -.l1: // copy -dest % 8 bytes - ld1 value1 = [src], 1 // value = *src++ +.l1: /* copy -dest % 8 bytes */ + ld1 value1 = [src], 1 /* value = *src++ */ ld1 value2 = [dest], 1 ;; cmp.ne p6, p0 = value1, value2 (p6) br.cond.spnt .done br.cloop.dptk .l1 .dest_aligned: - and sh1 = 7, src // sh1 = src % 8 - and tmp = -8, len // tmp = len & -OPSIZ - and asrc = -8, src // asrc = src & -OPSIZ -- align src - shr.u loopcnt = len, 3 // loopcnt = len / 8 - and len = 7, len ;; // len = len % 8 - shl sh1 = sh1, 3 // sh1 = 8 * (src % 8) - adds loopcnt = -1, loopcnt // --loopcnt - mov pr.rot = 1 << 16 ;; // set rotating predicates - sub sh2 = 64, sh1 // sh2 = 64 - sh1 - mov ar.lc = loopcnt // set LC - cmp.eq p6, p0 = sh1, r0 // is the src aligned? + and sh1 = 7, src /* sh1 = src % 8 */ + and tmp = -8, len /* tmp = len & -OPSIZ */ + and asrc = -8, src /* asrc = src & -OPSIZ -- align src */ + shr.u loopcnt = len, 3 /* loopcnt = len / 8 */ + and len = 7, len ;; /* len = len % 8 */ + shl sh1 = sh1, 3 /* sh1 = 8 * (src % 8) */ + adds loopcnt = -1, loopcnt /* --loopcnt */ + mov pr.rot = 1 << 16 ;; /* set rotating predicates */ + sub sh2 = 64, sh1 /* sh2 = 64 - sh1 */ + mov ar.lc = loopcnt /* set LC */ + cmp.eq p6, p0 = sh1, r0 /* is the src aligned? */ (p6) br.cond.sptk .src_aligned - add src = src, tmp // src += len & -OPSIZ - mov ar.ec = MEMLAT + 4 + 1 // four more passes needed - ld8 r[1] = [asrc], 8 ;; // r[1] = w0 + add src = src, tmp /* src += len & -OPSIZ */ + mov ar.ec = MEMLAT + 4 + 1 /* four more passes needed */ + ld8 r[1] = [asrc], 8 ;; /* r[1] = w0 */ .align 32 -// We enter this loop with p6 cleared by the above comparison +/* We enter this loop with p6 cleared by the above comparison */ .l2: -(p[0]) ld8 r[0] = [asrc], 8 // r[0] = w1 +(p[0]) ld8 r[0] = [asrc], 8 /* r[0] = w1 */ (p[0]) ld8 q[0] = [dest], 8 -(p[MEMLAT]) shr.u tmp1[0] = r[1 + MEMLAT], sh1 // tmp1 = w0 >> sh1 -(p[MEMLAT]) shl tmp2[0] = r[0 + MEMLAT], sh2 // tmp2 = w1 << sh2 +(p[MEMLAT]) shr.u tmp1[0] = r[1 + MEMLAT], sh1 /* tmp1 = w0 >> sh1 */ +(p[MEMLAT]) shl tmp2[0] = r[0 + MEMLAT], sh2 /* tmp2 = w1 << sh2 */ (p[MEMLAT+4]) cmp.ne p6, p0 = q[MEMLAT + 4], val[1] -(p[MEMLAT+3]) or val[0] = tmp1[3], tmp2[3] // val = tmp1 | tmp2 +(p[MEMLAT+3]) or val[0] = tmp1[3], tmp2[3] /* val = tmp1 | tmp2 */ (p6) br.cond.spnt .l2exit br.ctop.sptk .l2 br.cond.sptk .cmpfew .l3exit: mux1 value1 = r[MEMLAT], @rev mux1 value2 = q[MEMLAT], @rev - cmp.ne p6, p0 = r0, r0 ;; // clear p6 + cmp.ne p6, p0 = r0, r0 ;; /* clear p6 */ .l2exit: (p6) mux1 value1 = val[1], @rev (p6) mux1 value2 = q[MEMLAT + 4], @rev ;; cmp.ltu p6, p7 = value2, value1 ;; (p6) mov ret0 = -1 (p7) mov ret0 = 1 - mov pr = saved_pr, -1 // restore the predicate registers - mov ar.lc = saved_lc // restore the loop counter + mov pr = saved_pr, -1 /* restore the predicate registers */ + mov ar.lc = saved_lc /* restore the loop counter */ br.ret.sptk.many b0 .src_aligned: - cmp.ne p6, p0 = r0, r0 // clear p6 - mov ar.ec = MEMLAT + 1 ;; // set EC + cmp.ne p6, p0 = r0, r0 /* clear p6 */ + mov ar.ec = MEMLAT + 1 ;; /* set EC */ .l3: (p[0]) ld8 r[0] = [src], 8 (p[0]) ld8 q[0] = [dest], 8 @@ -142,8 +142,8 @@ ENTRY(memcmp) (p6) br.cond.spnt .l3exit br.ctop.dptk .l3 ;; .cmpfew: - cmp.eq p6, p0 = len, r0 // is len == 0 ? - adds len = -1, len // --len; + cmp.eq p6, p0 = len, r0 /* is len == 0 ? */ + adds len = -1, len /* --len; */ (p6) br.cond.spnt .restore_and_exit ;; mov ar.lc = len .l4: @@ -154,10 +154,10 @@ ENTRY(memcmp) (p6) br.cond.spnt .done br.cloop.dptk .l4 ;; .done: -(p6) sub ret0 = value2, value1 // don't execute it if falling thru +(p6) sub ret0 = value2, value1 /* don't execute it if falling thru */ .restore_and_exit: - mov pr = saved_pr, -1 // restore the predicate registers - mov ar.lc = saved_lc // restore the loop counter + mov pr = saved_pr, -1 /* restore the predicate registers */ + mov ar.lc = saved_lc /* restore the loop counter */ br.ret.sptk.many b0 END(memcmp) libc_hidden_def (memcmp) |