diff options
Diffstat (limited to 'sysdeps/arm/armv7/multiarch')
-rw-r--r-- | sysdeps/arm/armv7/multiarch/memcpy_impl.S | 581 |
1 files changed, 196 insertions, 385 deletions
diff --git a/sysdeps/arm/armv7/multiarch/memcpy_impl.S b/sysdeps/arm/armv7/multiarch/memcpy_impl.S index 5d5a3cefaa..c1b9fb0ab5 100644 --- a/sysdeps/arm/armv7/multiarch/memcpy_impl.S +++ b/sysdeps/arm/armv7/multiarch/memcpy_impl.S @@ -226,71 +226,40 @@ #ifdef USE_VFP .macro cpy_line_vfp vreg, base - sfi_breg dst, \ - vstr \vreg, [\B, #\base] - sfi_breg src, \ - vldr \vreg, [\B, #\base] - sfi_breg dst, \ - vstr d0, [\B, #\base + 8] - sfi_breg src, \ - vldr d0, [\B, #\base + 8] - sfi_breg dst, \ - vstr d1, [\B, #\base + 16] - sfi_breg src, \ - vldr d1, [\B, #\base + 16] - sfi_breg dst, \ - vstr d2, [\B, #\base + 24] - sfi_breg src, \ - vldr d2, [\B, #\base + 24] - sfi_breg dst, \ - vstr \vreg, [\B, #\base + 32] - sfi_breg src, \ - vldr \vreg, [\B, #\base + prefetch_lines * 64 - 32] - sfi_breg dst, \ - vstr d0, [\B, #\base + 40] - sfi_breg src, \ - vldr d0, [\B, #\base + 40] - sfi_breg dst, \ - vstr d1, [\B, #\base + 48] - sfi_breg src, \ - vldr d1, [\B, #\base + 48] - sfi_breg dst, \ - vstr d2, [\B, #\base + 56] - sfi_breg src, \ - vldr d2, [\B, #\base + 56] + vstr \vreg, [dst, #\base] + vldr \vreg, [src, #\base] + vstr d0, [dst, #\base + 8] + vldr d0, [src, #\base + 8] + vstr d1, [dst, #\base + 16] + vldr d1, [src, #\base + 16] + vstr d2, [dst, #\base + 24] + vldr d2, [src, #\base + 24] + vstr \vreg, [dst, #\base + 32] + vldr \vreg, [src, #\base + prefetch_lines * 64 - 32] + vstr d0, [dst, #\base + 40] + vldr d0, [src, #\base + 40] + vstr d1, [dst, #\base + 48] + vldr d1, [src, #\base + 48] + vstr d2, [dst, #\base + 56] + vldr d2, [src, #\base + 56] .endm .macro cpy_tail_vfp vreg, base - sfi_breg dst, \ - vstr \vreg, [\B, #\base] - sfi_breg src, \ - vldr \vreg, [\B, #\base] - sfi_breg dst, \ - vstr d0, [\B, #\base + 8] - sfi_breg src, \ - vldr d0, [\B, #\base + 8] - sfi_breg dst, \ - vstr d1, [\B, #\base + 16] - sfi_breg src, \ - vldr d1, [\B, #\base + 16] - sfi_breg dst, \ - vstr d2, [\B, #\base + 24] - sfi_breg src, \ - vldr d2, [\B, #\base + 24] - sfi_breg dst, \ - vstr \vreg, [\B, #\base + 32] - sfi_breg dst, \ - vstr d0, [\B, #\base + 40] - sfi_breg src, \ - vldr d0, [\B, #\base + 40] - sfi_breg dst, \ - vstr d1, [\B, #\base + 48] - sfi_breg src, \ - vldr d1, [\B, #\base + 48] - sfi_breg dst, \ - vstr d2, [\B, #\base + 56] - sfi_breg src, \ - vldr d2, [\B, #\base + 56] + vstr \vreg, [dst, #\base] + vldr \vreg, [src, #\base] + vstr d0, [dst, #\base + 8] + vldr d0, [src, #\base + 8] + vstr d1, [dst, #\base + 16] + vldr d1, [src, #\base + 16] + vstr d2, [dst, #\base + 24] + vldr d2, [src, #\base + 24] + vstr \vreg, [dst, #\base + 32] + vstr d0, [dst, #\base + 40] + vldr d0, [src, #\base + 40] + vstr d1, [dst, #\base + 48] + vldr d1, [src, #\base + 48] + vstr d2, [dst, #\base + 56] + vldr d2, [src, #\base + 56] .endm #endif @@ -316,26 +285,16 @@ ENTRY(memcpy) vst1.8 {d0}, [\reg]! .endm - /* These are used by the NaCl sfi_breg macro. */ - .macro _sfi_breg_dmask_neon_load_d0 reg - _sfi_dmask \reg - .endm - .macro _sfi_breg_dmask_neon_store_d0 reg - _sfi_dmask \reg - .endm - and tmp1, count, #0x38 .macro dispatch_step i - sfi_breg src, neon_load_d0 \B - sfi_breg dst, neon_store_d0 \B + neon_load_d0 src + neon_store_d0 dst .endm dispatch_7_dword tst count, #4 - sfi_breg src, \ - ldrne tmp1, [\B], #4 - sfi_breg dst, \ - strne tmp1, [\B], #4 + ldrne tmp1, [src], #4 + strne tmp1, [dst], #4 #else /* Copy up to 15 full words of data. May not be aligned. */ /* Cannot use VFP for unaligned data. */ @@ -344,23 +303,17 @@ ENTRY(memcpy) add src, src, tmp1 /* Jump directly into the sequence below at the correct offset. */ .macro dispatch_step i - sfi_breg src, \ - ldr tmp1, [\B, #-(\i * 4)] - sfi_breg dst, \ - str tmp1, [\B, #-(\i * 4)] + ldr tmp1, [src, #-(\i * 4)] + str tmp1, [dst, #-(\i * 4)] .endm dispatch_15_word #endif lsls count, count, #31 - sfi_breg src, \ - ldrhcs tmp1, [\B], #2 - sfi_breg src, \ - ldrbne src, [\B] /* Src is dead, use as a scratch. */ - sfi_breg dst, \ - strhcs tmp1, [\B], #2 - sfi_breg dst, \ - strbne src, [\B] + ldrhcs tmp1, [src], #2 + ldrbne src, [src] /* Src is dead, use as a scratch. */ + strhcs tmp1, [dst], #2 + strbne src, [dst] bx lr .Lcpy_not_short: @@ -388,19 +341,13 @@ ENTRY(memcpy) beq 1f rsbs tmp2, tmp2, #0 sub count, count, tmp2, lsr #29 - sfi_breg src, \ - ldrmi tmp1, [\B], #4 - sfi_breg dst, \ - strmi tmp1, [\B], #4 + ldrmi tmp1, [src], #4 + strmi tmp1, [dst], #4 lsls tmp2, tmp2, #2 - sfi_breg src, \ - ldrhcs tmp1, [\B], #2 - sfi_breg src, \ - ldrbne tmp2, [\B], #1 - sfi_breg dst, \ - strhcs tmp1, [\B], #2 - sfi_breg dst, \ - strbne tmp2, [\B], #1 + ldrhcs tmp1, [src], #2 + ldrbne tmp2, [src], #1 + strhcs tmp1, [dst], #2 + strbne tmp2, [dst], #1 1: subs tmp2, count, #64 /* Use tmp2 for count. */ @@ -412,40 +359,24 @@ ENTRY(memcpy) .Lcpy_body_medium: /* Count in tmp2. */ #ifdef USE_VFP 1: - sfi_breg src, \ - vldr d0, [\B, #0] + vldr d0, [src, #0] subs tmp2, tmp2, #64 - sfi_breg src, \ - vldr d1, [\B, #8] - sfi_breg dst, \ - vstr d0, [\B, #0] - sfi_breg src, \ - vldr d0, [\B, #16] - sfi_breg dst, \ - vstr d1, [\B, #8] - sfi_breg src, \ - vldr d1, [\B, #24] - sfi_breg dst, \ - vstr d0, [\B, #16] - sfi_breg src, \ - vldr d0, [\B, #32] - sfi_breg dst, \ - vstr d1, [\B, #24] - sfi_breg src, \ - vldr d1, [\B, #40] - sfi_breg dst, \ - vstr d0, [\B, #32] - sfi_breg src, \ - vldr d0, [\B, #48] - sfi_breg dst, \ - vstr d1, [\B, #40] - sfi_breg src, \ - vldr d1, [\B, #56] - sfi_breg dst, \ - vstr d0, [\B, #48] + vldr d1, [src, #8] + vstr d0, [dst, #0] + vldr d0, [src, #16] + vstr d1, [dst, #8] + vldr d1, [src, #24] + vstr d0, [dst, #16] + vldr d0, [src, #32] + vstr d1, [dst, #24] + vldr d1, [src, #40] + vstr d0, [dst, #32] + vldr d0, [src, #48] + vstr d1, [dst, #40] + vldr d1, [src, #56] + vstr d0, [dst, #48] add src, src, #64 - sfi_breg dst, \ - vstr d1, [\B, #56] + vstr d1, [dst, #56] add dst, dst, #64 bge 1b tst tmp2, #0x3f @@ -456,48 +387,30 @@ ENTRY(memcpy) add dst, dst, tmp1 add src, src, tmp1 .macro dispatch_step i - sfi_breg src, \ - vldr d0, [\B, #-(\i * 8)] - sfi_breg dst, \ - vstr d0, [\B, #-(\i * 8)] + vldr d0, [src, #-(\i * 8)] + vstr d0, [dst, #-(\i * 8)] .endm dispatch_7_dword #else sub src, src, #8 sub dst, dst, #8 1: - sfi_breg src, \ - ldrd A_l, A_h, [\B, #8] - sfi_breg dst, \ - strd A_l, A_h, [\B, #8] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #16] - sfi_breg dst, \ - strd A_l, A_h, [\B, #16] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #24] - sfi_breg dst, \ - strd A_l, A_h, [\B, #24] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #32] - sfi_breg dst, \ - strd A_l, A_h, [\B, #32] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #40] - sfi_breg dst, \ - strd A_l, A_h, [\B, #40] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #48] - sfi_breg dst, \ - strd A_l, A_h, [\B, #48] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #56] - sfi_breg dst, \ - strd A_l, A_h, [\B, #56] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #64]! - sfi_breg dst, \ - strd A_l, A_h, [\B, #64]! + ldrd A_l, A_h, [src, #8] + strd A_l, A_h, [dst, #8] + ldrd A_l, A_h, [src, #16] + strd A_l, A_h, [dst, #16] + ldrd A_l, A_h, [src, #24] + strd A_l, A_h, [dst, #24] + ldrd A_l, A_h, [src, #32] + strd A_l, A_h, [dst, #32] + ldrd A_l, A_h, [src, #40] + strd A_l, A_h, [dst, #40] + ldrd A_l, A_h, [src, #48] + strd A_l, A_h, [dst, #48] + ldrd A_l, A_h, [src, #56] + strd A_l, A_h, [dst, #56] + ldrd A_l, A_h, [src, #64]! + strd A_l, A_h, [dst, #64]! subs tmp2, tmp2, #64 bge 1b tst tmp2, #0x3f @@ -524,28 +437,20 @@ ENTRY(memcpy) add dst, dst, tmp1 add src, src, tmp1 .macro dispatch_step i - sfi_breg src, \ - ldrd A_l, A_h, [\B, #-(\i * 8)] - sfi_breg dst, \ - strd A_l, A_h, [\B, #-(\i * 8)] + ldrd A_l, A_h, [src, #-(\i * 8)] + strd A_l, A_h, [dst, #-(\i * 8)] .endm dispatch_7_dword #endif tst tmp2, #4 - sfi_breg src, \ - ldrne tmp1, [\B], #4 - sfi_breg dst, \ - strne tmp1, [\B], #4 + ldrne tmp1, [src], #4 + strne tmp1, [dst], #4 lsls tmp2, tmp2, #31 /* Count (tmp2) now dead. */ - sfi_breg src, \ - ldrhcs tmp1, [\B], #2 - sfi_breg src, \ - ldrbne tmp2, [\B] - sfi_breg dst, \ - strhcs tmp1, [\B], #2 - sfi_breg dst, \ - strbne tmp2, [\B] + ldrhcs tmp1, [src], #2 + ldrbne tmp2, [src] + strhcs tmp1, [dst], #2 + strbne tmp2, [dst] .Ldone: ldr tmp2, [sp], #FRAME_SIZE @@ -565,23 +470,15 @@ ENTRY(memcpy) copy position into a register. This should act like a PLD operation but we won't have to repeat the transfer. */ - sfi_breg src, \ - vldr d3, [\B, #0] - sfi_breg src, \ - vldr d4, [\B, #64] - sfi_breg src, \ - vldr d5, [\B, #128] - sfi_breg src, \ - vldr d6, [\B, #192] - sfi_breg src, \ - vldr d7, [\B, #256] - - sfi_breg src, \ - vldr d0, [\B, #8] - sfi_breg src, \ - vldr d1, [\B, #16] - sfi_breg src, \ - vldr d2, [\B, #24] + vldr d3, [src, #0] + vldr d4, [src, #64] + vldr d5, [src, #128] + vldr d6, [src, #192] + vldr d7, [src, #256] + + vldr d0, [src, #8] + vldr d1, [src, #16] + vldr d2, [src, #24] add src, src, #32 subs tmp2, tmp2, #prefetch_lines * 64 * 2 @@ -606,31 +503,19 @@ ENTRY(memcpy) add src, src, #3 * 64 add dst, dst, #3 * 64 cpy_tail_vfp d6, 0 - sfi_breg dst, \ - vstr d7, [\B, #64] - sfi_breg src, \ - vldr d7, [\B, #64] - sfi_breg dst, \ - vstr d0, [\B, #64 + 8] - sfi_breg src, \ - vldr d0, [\B, #64 + 8] - sfi_breg dst, \ - vstr d1, [\B, #64 + 16] - sfi_breg src, \ - vldr d1, [\B, #64 + 16] - sfi_breg dst, \ - vstr d2, [\B, #64 + 24] - sfi_breg src, \ - vldr d2, [\B, #64 + 24] - sfi_breg dst, \ - vstr d7, [\B, #64 + 32] + vstr d7, [dst, #64] + vldr d7, [src, #64] + vstr d0, [dst, #64 + 8] + vldr d0, [src, #64 + 8] + vstr d1, [dst, #64 + 16] + vldr d1, [src, #64 + 16] + vstr d2, [dst, #64 + 24] + vldr d2, [src, #64 + 24] + vstr d7, [dst, #64 + 32] add src, src, #96 - sfi_breg dst, \ - vstr d0, [\B, #64 + 40] - sfi_breg dst, \ - vstr d1, [\B, #64 + 48] - sfi_breg dst, \ - vstr d2, [\B, #64 + 56] + vstr d0, [dst, #64 + 40] + vstr d1, [dst, #64 + 48] + vstr d2, [dst, #64 + 56] add dst, dst, #128 add tmp2, tmp2, #prefetch_lines * 64 b .Lcpy_body_medium @@ -641,83 +526,59 @@ ENTRY(memcpy) /* Pre-bias src and dst. */ sub src, src, #8 sub dst, dst, #8 - sfi_pld src, #8 - sfi_pld src, #72 + pld [src, #8] + pld [src, #72] subs tmp2, tmp2, #64 - sfi_pld src, #136 - sfi_breg src, \ - ldrd A_l, A_h, [\B, #8] + pld [src, #136] + ldrd A_l, A_h, [src, #8] strd B_l, B_h, [sp, #8] cfi_rel_offset (B_l, 8) cfi_rel_offset (B_h, 12) - sfi_breg src, \ - ldrd B_l, B_h, [\B, #16] + ldrd B_l, B_h, [src, #16] strd C_l, C_h, [sp, #16] cfi_rel_offset (C_l, 16) cfi_rel_offset (C_h, 20) - sfi_breg src, \ - ldrd C_l, C_h, [\B, #24] + ldrd C_l, C_h, [src, #24] strd D_l, D_h, [sp, #24] cfi_rel_offset (D_l, 24) cfi_rel_offset (D_h, 28) - sfi_pld src, #200 - sfi_breg src, \ - ldrd D_l, D_h, [\B, #32]! + pld [src, #200] + ldrd D_l, D_h, [src, #32]! b 1f .p2align 6 2: - sfi_pld src, #232 - sfi_breg dst, \ - strd A_l, A_h, [\B, #40] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #40] - sfi_breg dst, \ - strd B_l, B_h, [\B, #48] - sfi_breg src, \ - ldrd B_l, B_h, [\B, #48] - sfi_breg dst, \ - strd C_l, C_h, [\B, #56] - sfi_breg src, \ - ldrd C_l, C_h, [\B, #56] - sfi_breg dst, \ - strd D_l, D_h, [\B, #64]! - sfi_breg src, \ - ldrd D_l, D_h, [\B, #64]! + pld [src, #232] + strd A_l, A_h, [dst, #40] + ldrd A_l, A_h, [src, #40] + strd B_l, B_h, [dst, #48] + ldrd B_l, B_h, [src, #48] + strd C_l, C_h, [dst, #56] + ldrd C_l, C_h, [src, #56] + strd D_l, D_h, [dst, #64]! + ldrd D_l, D_h, [src, #64]! subs tmp2, tmp2, #64 1: - sfi_breg dst, \ - strd A_l, A_h, [\B, #8] - sfi_breg src, \ - ldrd A_l, A_h, [\B, #8] - sfi_breg dst, \ - strd B_l, B_h, [\B, #16] - sfi_breg src, \ - ldrd B_l, B_h, [\B, #16] - sfi_breg dst, \ - strd C_l, C_h, [\B, #24] - sfi_breg src, \ - ldrd C_l, C_h, [\B, #24] - sfi_breg dst, \ - strd D_l, D_h, [\B, #32] - sfi_breg src, \ - ldrd D_l, D_h, [\B, #32] + strd A_l, A_h, [dst, #8] + ldrd A_l, A_h, [src, #8] + strd B_l, B_h, [dst, #16] + ldrd B_l, B_h, [src, #16] + strd C_l, C_h, [dst, #24] + ldrd C_l, C_h, [src, #24] + strd D_l, D_h, [dst, #32] + ldrd D_l, D_h, [src, #32] bcs 2b /* Save the remaining bytes and restore the callee-saved regs. */ - sfi_breg dst, \ - strd A_l, A_h, [\B, #40] + strd A_l, A_h, [dst, #40] add src, src, #40 - sfi_breg dst, \ - strd B_l, B_h, [\B, #48] + strd B_l, B_h, [dst, #48] ldrd B_l, B_h, [sp, #8] cfi_restore (B_l) cfi_restore (B_h) - sfi_breg dst, \ - strd C_l, C_h, [\B, #56] + strd C_l, C_h, [dst, #56] ldrd C_l, C_h, [sp, #16] cfi_restore (C_l) cfi_restore (C_h) - sfi_breg dst, \ - strd D_l, D_h, [\B, #64] + strd D_l, D_h, [dst, #64] ldrd D_l, D_h, [sp, #24] cfi_restore (D_l) cfi_restore (D_h) @@ -734,35 +595,29 @@ ENTRY(memcpy) cfi_remember_state .Lcpy_notaligned: - sfi_pld src - sfi_pld src, #64 + pld [src, #0] + pld [src, #64] /* There's at least 64 bytes to copy, but there is no mutual alignment. */ /* Bring DST to 64-bit alignment. */ lsls tmp2, dst, #29 - sfi_pld src, #(2 * 64) + pld [src, #(2 * 64)] beq 1f rsbs tmp2, tmp2, #0 sub count, count, tmp2, lsr #29 - sfi_breg src, \ - ldrmi tmp1, [\B], #4 - sfi_breg dst, \ - strmi tmp1, [\B], #4 + ldrmi tmp1, [src], #4 + strmi tmp1, [dst], #4 lsls tmp2, tmp2, #2 - sfi_breg src, \ - ldrbne tmp1, [\B], #1 - sfi_breg src, \ - ldrhcs tmp2, [\B], #2 - sfi_breg dst, \ - strbne tmp1, [\B], #1 - sfi_breg dst, \ - strhcs tmp2, [\B], #2 + ldrbne tmp1, [src], #1 + ldrhcs tmp2, [src], #2 + strbne tmp1, [dst], #1 + strhcs tmp2, [dst], #2 1: - sfi_pld src, #(3 * 64) + pld [src, #(3 * 64)] subs count, count, #64 ldrmi tmp2, [sp], #FRAME_SIZE bmi .Ltail63unaligned - sfi_pld src, #(4 * 64) + pld [src, #(4 * 64)] #ifdef USE_NEON /* These need an extra layer of macro just to work around a @@ -775,132 +630,88 @@ ENTRY(memcpy) vst1.8 {\reglist}, [ALIGN (\basereg, 64)]! .endm - /* These are used by the NaCl sfi_breg macro. */ - .macro _sfi_breg_dmask_neon_load_multi reg - _sfi_dmask \reg - .endm - .macro _sfi_breg_dmask_neon_store_multi reg - _sfi_dmask \reg - .endm - - sfi_breg src, neon_load_multi d0-d3, \B - sfi_breg src, neon_load_multi d4-d7, \B + neon_load_multi d0-d3, src + neon_load_multi d4-d7, src subs count, count, #64 bmi 2f 1: - sfi_pld src, #(4 * 64) - sfi_breg dst, neon_store_multi d0-d3, \B - sfi_breg src, neon_load_multi d0-d3, \B - sfi_breg dst, neon_store_multi d4-d7, \B - sfi_breg src, neon_load_multi d4-d7, \B + pld [src, #(4 * 64)] + neon_store_multi d0-d3, dst + neon_load_multi d0-d3, src + neon_store_multi d4-d7, dst + neon_load_multi d4-d7, src subs count, count, #64 bpl 1b 2: - sfi_breg dst, neon_store_multi d0-d3, \B - sfi_breg dst, neon_store_multi d4-d7, \B + neon_store_multi d0-d3, dst + neon_store_multi d4-d7, dst ands count, count, #0x3f #else /* Use an SMS style loop to maximize the I/O bandwidth. */ sub src, src, #4 sub dst, dst, #8 subs tmp2, count, #64 /* Use tmp2 for count. */ - sfi_breg src, \ - ldr A_l, [\B, #4] - sfi_breg src, \ - ldr A_h, [\B, #8] + ldr A_l, [src, #4] + ldr A_h, [src, #8] strd B_l, B_h, [sp, #8] cfi_rel_offset (B_l, 8) cfi_rel_offset (B_h, 12) - sfi_breg src, \ - ldr B_l, [\B, #12] - sfi_breg src, \ - ldr B_h, [\B, #16] + ldr B_l, [src, #12] + ldr B_h, [src, #16] strd C_l, C_h, [sp, #16] cfi_rel_offset (C_l, 16) cfi_rel_offset (C_h, 20) - sfi_breg src, \ - ldr C_l, [\B, #20] - sfi_breg src, \ - ldr C_h, [\B, #24] + ldr C_l, [src, #20] + ldr C_h, [src, #24] strd D_l, D_h, [sp, #24] cfi_rel_offset (D_l, 24) cfi_rel_offset (D_h, 28) - sfi_breg src, \ - ldr D_l, [\B, #28] - sfi_breg src, \ - ldr D_h, [\B, #32]! + ldr D_l, [src, #28] + ldr D_h, [src, #32]! b 1f .p2align 6 2: - sfi_pld src, #(5 * 64) - (32 - 4) - sfi_breg dst, \ - strd A_l, A_h, [\B, #40] - sfi_breg src, \ - ldr A_l, [\B, #36] - sfi_breg src, \ - ldr A_h, [\B, #40] - sfi_breg dst, \ - strd B_l, B_h, [\B, #48] - sfi_breg src, \ - ldr B_l, [\B, #44] - sfi_breg src, \ - ldr B_h, [\B, #48] - sfi_breg dst, \ - strd C_l, C_h, [\B, #56] - sfi_breg src, \ - ldr C_l, [\B, #52] - sfi_breg src, \ - ldr C_h, [\B, #56] - sfi_breg dst, \ - strd D_l, D_h, [\B, #64]! - sfi_breg src, \ - ldr D_l, [\B, #60] - sfi_breg src, \ - ldr D_h, [\B, #64]! + pld [src, #(5 * 64) - (32 - 4)] + strd A_l, A_h, [dst, #40] + ldr A_l, [src, #36] + ldr A_h, [src, #40] + strd B_l, B_h, [dst, #48] + ldr B_l, [src, #44] + ldr B_h, [src, #48] + strd C_l, C_h, [dst, #56] + ldr C_l, [src, #52] + ldr C_h, [src, #56] + strd D_l, D_h, [dst, #64]! + ldr D_l, [src, #60] + ldr D_h, [src, #64]! subs tmp2, tmp2, #64 1: - sfi_breg dst, \ - strd A_l, A_h, [\B, #8] - sfi_breg src, \ - ldr A_l, [\B, #4] - sfi_breg src, \ - ldr A_h, [\B, #8] - sfi_breg dst, \ - strd B_l, B_h, [\B, #16] - sfi_breg src, \ - ldr B_l, [\B, #12] - sfi_breg src, \ - ldr B_h, [\B, #16] - sfi_breg dst, \ - strd C_l, C_h, [\B, #24] - sfi_breg src, \ - ldr C_l, [\B, #20] - sfi_breg src, \ - ldr C_h, [\B, #24] - sfi_breg dst, \ - strd D_l, D_h, [\B, #32] - sfi_breg src, \ - ldr D_l, [\B, #28] - sfi_breg src, \ - ldr D_h, [\B, #32] + strd A_l, A_h, [dst, #8] + ldr A_l, [src, #4] + ldr A_h, [src, #8] + strd B_l, B_h, [dst, #16] + ldr B_l, [src, #12] + ldr B_h, [src, #16] + strd C_l, C_h, [dst, #24] + ldr C_l, [src, #20] + ldr C_h, [src, #24] + strd D_l, D_h, [dst, #32] + ldr D_l, [src, #28] + ldr D_h, [src, #32] bcs 2b /* Save the remaining bytes and restore the callee-saved regs. */ - sfi_breg dst, \ - strd A_l, A_h, [\B, #40] + strd A_l, A_h, [dst, #40] add src, src, #36 - sfi_breg dst, \ - strd B_l, B_h, [\B, #48] + strd B_l, B_h, [dst, #48] ldrd B_l, B_h, [sp, #8] cfi_restore (B_l) cfi_restore (B_h) - sfi_breg dst, \ - strd C_l, C_h, [\B, #56] + strd C_l, C_h, [dst, #56] ldrd C_l, C_h, [sp, #16] cfi_restore (C_l) cfi_restore (C_h) - sfi_breg dst, \ - strd D_l, D_h, [\B, #64] + strd D_l, D_h, [dst, #64] ldrd D_l, D_h, [sp, #24] cfi_restore (D_l) cfi_restore (D_h) |