diff options
Diffstat (limited to 'sysdeps/alpha/stxcpy.S')
| -rw-r--r-- | sysdeps/alpha/stxcpy.S | 33 |
1 files changed, 17 insertions, 16 deletions
diff --git a/sysdeps/alpha/stxcpy.S b/sysdeps/alpha/stxcpy.S index e381b704cd..49dd8e5c64 100644 --- a/sysdeps/alpha/stxcpy.S +++ b/sysdeps/alpha/stxcpy.S @@ -36,7 +36,7 @@ Cambridge, MA 02139, USA. */ Furthermore, v0, a3-a5, t11, and t12 are untouched. */ -/* This is generally scheduled for the EV5, but should still be pretty +/* This is generally scheduled for the EV5, but should still be pretty good for the EV4 too. */ #include <sysdep.h> @@ -65,7 +65,7 @@ stxcpy_aligned: lda t2, -1 # e1 : build a mask against false zero mskqh t2, a1, t2 # e0 : detection in the src word mskqh t1, a1, t3 # e0 : - ornot t1, t2, t2 # .. e1 : + ornot t1, t2, t2 # .. e1 : mskql t0, a1, t0 # e0 : assemble the first output word cmpbge zero, t2, t7 # .. e1 : bits set iff null found or t0, t3, t1 # e0 : @@ -99,9 +99,9 @@ $a_eos: /* We're doing a partial word store and so need to combine our source and original destination words. */ ldq_u t0, 0(a0) # e0 : - subq t8, 1, t6 # .. e1 : + subq t8, 1, t6 # .. e1 : zapnot t1, t6, t1 # e0 : clear src bytes >= null - or t8, t6, t7 # .. e1 : + or t8, t6, t7 # .. e1 : zap t0, t7, t0 # e0 : clear dst bytes <= null or t0, t1, t1 # e1 : @@ -156,13 +156,13 @@ $u_head: or t1, t4, t1 # .. e1 : mskqh t1, a0, t1 # e0 : or t0, t1, t1 # e1 : - + or t1, t6, t6 # e0 : cmpbge zero, t6, t7 # .. e1 : lda t6, -1 # e0 : for masking just below bne t7, $u_final # .. e1 : - mskql t6, a1, t6 # e0 : mask out the bits we have + mskql t6, a1, t6 # e0 : mask out the bits we have or t6, t2, t2 # e1 : already extracted before cmpbge zero, t2, t7 # e0 : testing eos bne t7, $u_late_head_exit # .. e1 (zdb) @@ -181,7 +181,7 @@ $u_head: /* Unaligned copy main loop. In order to avoid reading too much, the loop is structured to detect zeros in aligned source words. - This has, unfortunately, effectively pulled half of a loop + This has, unfortunately, effectively pulled half of a loop iteration out into the head and half into the tail, but it does prevent nastiness from accumulating in the very thing we want to run as fast as possible. @@ -207,7 +207,7 @@ $u_loop: /* We've found a zero somewhere in the source word we just read. If it resides in the lower half, we have one (probably partial) - word to write out, and if it resides in the upper half, we + word to write out, and if it resides in the upper half, we have one full and one partial word left to write out. On entry to this basic block: @@ -234,7 +234,7 @@ $u_final: negq t7, t6 # e0 : isolate low bit set and t6, t7, t8 # e1 : - and t8, 0x80, t6 # e0 : avoid dest word load if we can + and t8, 0x80, t6 # e0 : avoid dest word load if we can bne t6, 1f # .. e1 (zdb) ldq_u t0, 0(a0) # e0 : @@ -256,7 +256,7 @@ $unaligned: and a0, 7, t4 # .. e1 : find dest misalignment and a1, 7, t5 # e0 : find src misalignment - /* Conditionally load the first destination word and a bytemask + /* Conditionally load the first destination word and a bytemask with 0xff indicating that the destination byte is sacrosanct. */ mov zero, t0 # .. e1 : @@ -290,18 +290,19 @@ $unaligned: negq t7, t6 # .. e1 : build bitmask of bytes <= zero and t6, t7, t8 # e0 : - nop # .. e1 : + and a1, 7, t5 # .. e1 : subq t8, 1, t6 # e0 : or t6, t8, t7 # e1 : + srl t8, t5, t8 # e0 : adjust final null return value - zapnot t2, t7, t2 # e0 : prepare source word; mirror changes + zapnot t2, t7, t2 # .. e1 : prepare source word; mirror changes and t1, t2, t1 # e1 : to source validity mask - extql t2, a1, t2 # e0 : + extql t2, a1, t2 # .. e0 : extql t1, a1, t1 # e0 : - andnot t0, t2, t0 # e0 : zero place for source to reside + andnot t0, t2, t0 # .. e1 : zero place for source to reside or t0, t1, t1 # e1 : and put it there - stq_u t1, 0(a0) # e0 : - ret (t9) # .. e1 : + stq_u t1, 0(a0) # .. e0 : + ret (t9) .end __stxcpy |