view release on metacpan or  search on metacpan

ext/zstd/lib/decompress/huf_decompress_amd64.S  view on Meta::CPAN

.global _HUF_decompress4X2_usingDTable_internal_fast_asm_loop
.text

/* Sets up register mappings for clarity.
 * op[], bits[], dtable & ip[0] each get their own register.
 * ip[1,2,3] & olimit alias var[].
 * %rax is a scratch register.
 */

#define op0    rsi
#define op1    rbx
#define op2    rcx
#define op3    rdi

#define ip0    r8
#define ip1    r9
#define ip2    r10
#define ip3    r11

#define bits0  rbp
#define bits1  rdx
#define bits2  r12
#define bits3  r13
#define dtable r14
#define olimit r15

/* var[] aliases ip[1,2,3] & olimit
 * ip[1,2,3] are saved every iteration.
 * olimit is only used in compute_olimit.
 */
#define var0   r15
#define var1   r9
#define var2   r10
#define var3   r11

/* 32-bit var registers */
#define vard0  r15d
#define vard1  r9d
#define vard2  r10d
#define vard3  r11d

/* Calls X(N) for each stream 0, 1, 2, 3. */
#define FOR_EACH_STREAM(X) \
    X(0);                  \
    X(1);                  \
    X(2);                  \
    X(3)

/* Calls X(N, idx) for each stream 0, 1, 2, 3. */
#define FOR_EACH_STREAM_WITH_INDEX(X, idx) \
    X(0, idx);                             \
    X(1, idx);                             \
    X(2, idx);                             \
    X(3, idx)

/* Define both _HUF_* & HUF_* symbols because MacOS
 * C symbols are prefixed with '_' & Linux symbols aren't.
 */
_HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
    ZSTD_CET_ENDBRANCH
    /* Save all registers - even if they are callee saved for simplicity. */
    push %rax
    push %rbx
    push %rcx
    push %rdx
    push %rbp
    push %rsi
    push %rdi
    push %r8
    push %r9
    push %r10
    push %r11
    push %r12
    push %r13
    push %r14
    push %r15

    /* Read HUF_DecompressAsmArgs* args from %rax */
    movq %rdi, %rax
    movq  0(%rax), %ip0
    movq  8(%rax), %ip1
    movq 16(%rax), %ip2
    movq 24(%rax), %ip3
    movq 32(%rax), %op0
    movq 40(%rax), %op1
    movq 48(%rax), %op2
    movq 56(%rax), %op3
    movq 64(%rax), %bits0
    movq 72(%rax), %bits1
    movq 80(%rax), %bits2
    movq 88(%rax), %bits3
    movq 96(%rax), %dtable
    push %rax      /* argument */
    push 104(%rax) /* ilimit */
    push 112(%rax) /* oend */
    push %olimit   /* olimit space */

    subq $24, %rsp

.L_4X1_compute_olimit:
    /* Computes how many iterations we can do safely
     * %r15, %rax may be clobbered
     * rbx, rdx must be saved
     * op3 & ip0 mustn't be clobbered
     */
    movq %rbx, 0(%rsp)
    movq %rdx, 8(%rsp)

    movq 32(%rsp), %rax /* rax = oend */
    subq %op3,    %rax  /* rax = oend - op3 */

    /* r15 = (oend - op3) / 5 */
    movabsq $-3689348814741910323, %rdx
    mulq %rdx
    movq %rdx, %r15
    shrq $2, %r15

    movq %ip0,     %rax /* rax = ip0 */
    movq 40(%rsp), %rdx /* rdx = ilimit */
    subq %rdx,     %rax /* rax = ip0 - ilimit */

ext/zstd/lib/decompress/huf_decompress_amd64.S  view on Meta::CPAN

    /* If op3 < olimit: continue the loop */
    cmp %op3, 24(%rsp)
    ja .L_4X1_loop_body

    /* Reload ip[1,2,3] from stack */
    movq 0(%rsp), %ip1
    movq 8(%rsp), %ip2
    movq 16(%rsp), %ip3

    /* Re-compute olimit */
    jmp .L_4X1_compute_olimit

#undef GET_NEXT_DELT
#undef DECODE_FROM_DELT
#undef DECODE
#undef RELOAD_BITS
.L_4X1_exit:
    addq $24, %rsp

    /* Restore stack (oend & olimit) */
    pop %rax /* olimit */
    pop %rax /* oend */
    pop %rax /* ilimit */
    pop %rax /* arg */

    /* Save ip / op / bits */
    movq %ip0,  0(%rax)
    movq %ip1,  8(%rax)
    movq %ip2, 16(%rax)
    movq %ip3, 24(%rax)
    movq %op0, 32(%rax)
    movq %op1, 40(%rax)
    movq %op2, 48(%rax)
    movq %op3, 56(%rax)
    movq %bits0, 64(%rax)
    movq %bits1, 72(%rax)
    movq %bits2, 80(%rax)
    movq %bits3, 88(%rax)

    /* Restore registers */
    pop %r15
    pop %r14
    pop %r13
    pop %r12
    pop %r11
    pop %r10
    pop %r9
    pop %r8
    pop %rdi
    pop %rsi
    pop %rbp
    pop %rdx
    pop %rcx
    pop %rbx
    pop %rax
    ret

_HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
    ZSTD_CET_ENDBRANCH
    /* Save all registers - even if they are callee saved for simplicity. */
    push %rax
    push %rbx
    push %rcx
    push %rdx
    push %rbp
    push %rsi
    push %rdi
    push %r8
    push %r9
    push %r10
    push %r11
    push %r12
    push %r13
    push %r14
    push %r15

    movq %rdi, %rax
    movq  0(%rax), %ip0
    movq  8(%rax), %ip1
    movq 16(%rax), %ip2
    movq 24(%rax), %ip3
    movq 32(%rax), %op0
    movq 40(%rax), %op1
    movq 48(%rax), %op2
    movq 56(%rax), %op3
    movq 64(%rax), %bits0
    movq 72(%rax), %bits1
    movq 80(%rax), %bits2
    movq 88(%rax), %bits3
    movq 96(%rax), %dtable
    push %rax      /* argument */
    push %rax      /* olimit */
    push 104(%rax) /* ilimit */

    movq 112(%rax), %rax
    push %rax /* oend3 */

    movq %op3, %rax
    push %rax /* oend2 */

    movq %op2, %rax
    push %rax /* oend1 */

    movq %op1, %rax
    push %rax /* oend0 */

    /* Scratch space */
    subq $8, %rsp

.L_4X2_compute_olimit:
    /* Computes how many iterations we can do safely
     * %r15, %rax may be clobbered
     * rdx must be saved
     * op[1,2,3,4] & ip0 mustn't be clobbered
     */
    movq %rdx, 0(%rsp)

    /* We can consume up to 7 input bytes each iteration. */
    movq %ip0,     %rax  /* rax = ip0 */