diff options
Diffstat (limited to 'zap/source/amd64/mem/cp.s')
| -rw-r--r-- | zap/source/amd64/mem/cp.s | 15 |
1 files changed, 8 insertions, 7 deletions
diff --git a/zap/source/amd64/mem/cp.s b/zap/source/amd64/mem/cp.s index 3c9f2e1..b0b91e5 100644 --- a/zap/source/amd64/mem/cp.s +++ b/zap/source/amd64/mem/cp.s @@ -14,38 +14,38 @@ zap_cp: .big02cp: # big02cp:; // We assume AVX. cmp rdx,0x20 - jl short .big01cp # if (num < 0x20u) goto big01cp; + jl short .big01cp # if (rem < 0x20u) goto big01cp; vmovdqu ymm0,[rsi] # val02 = *(unsigned int256_t *)src; vmovdqu [rdi],ymm0 # *(unsigned int256_t *)dest = val02; add rdi,0x20 # dest += 0x20u; add rsi,0x20 # src += 0x20u; - sub rdx,0x20 # num -= 0x20u; + sub rdx,0x20 # rem -= 0x20u; jmp short .big02cp # goto big02cp; .big01cp: # big01cp:; cmp rdx,0x10 - jl short .wrdcp # if (num < 0x10u) goto wrdcp; + jl short .wrdcp # if (rem < 0x10u) goto wrdcp; movdqu xmm0,[rsi] # val01 = *(unsigned int128_t *)src; movdqu [rdi],xmm0 # *(unsigned int128_t *)dest = val01; add rdi,0x10 # dest += 0x10u; add rsi,0x10 # src += 0x10u; - sub rdx,0x10 # num -= 0x10u; + sub rdx,0x10 # rem -= 0x10u; jmp short .big01cp # goto big01cp; .wrdcp: # wrdcp:; cmp rdx,0x8 - jl short .bytecp # if (num < 0x8u) goto bytecp; + jl short .bytecp # if (rem < 0x8u) goto bytecp; mov rcx,[rsi] # val8 = *(zap_i04 *)src; mov [rdi],rcx # *(zap_i04 *)dest = val8; add rsi,0x8 # dest += 0x8u; add rdi,0x8 # src += 0x8u; - sub rdx,0x8 # num -= 0x8u; + sub rdx,0x8 # rem -= 0x8u; jmp short .wrdcp # goto wrdcp .bytecp: # bytecp:; @@ -62,4 +62,5 @@ zap_cp: .done: mov rax,rdi - ret # return dest; + mov rdx,rsi + ret # return (zap_cpret) {.dest = dest,.src = src}; |