diff options
Diffstat (limited to '')
| -rw-r--r-- | third_party/dav1d/src/arm/32/msac.S | 167 |
1 files changed, 90 insertions, 77 deletions
diff --git a/third_party/dav1d/src/arm/32/msac.S b/third_party/dav1d/src/arm/32/msac.S index b06e109dda..b16957fb7e 100644 --- a/third_party/dav1d/src/arm/32/msac.S +++ b/third_party/dav1d/src/arm/32/msac.S @@ -279,60 +279,67 @@ L(renorm): sub r4, r4, r3 // rng = u - v clz r5, r4 // clz(rng) eor r5, r5, #16 // d = clz(rng) ^ 16 - mvn r7, r7 // ~dif - add r7, r7, r3, lsl #16 // ~dif + (v << 16) + sub r7, r7, r3, lsl #16 // dif - (v << 16) L(renorm2): lsl r4, r4, r5 // rng << d subs r6, r6, r5 // cnt -= d - lsl r7, r7, r5 // (~dif + (v << 16)) << d + lsl r7, r7, r5 // (dif - (v << 16)) << d str r4, [r0, #RNG] - mvn r7, r7 // ~dif - bhs 9f + bhs 4f // refill ldr r3, [r0, #BUF_POS] // BUF_POS ldr r4, [r0, #BUF_END] // BUF_END add r5, r3, #4 - cmp r5, r4 - bgt 2f - - ldr r3, [r3] // next_bits - add r8, r6, #23 // shift_bits = cnt + 23 - add r6, r6, #16 // cnt += 16 - rev r3, r3 // next_bits = bswap(next_bits) - sub r5, r5, r8, lsr #3 // buf_pos -= shift_bits >> 3 - and r8, r8, #24 // shift_bits &= 24 - lsr r3, r3, r8 // next_bits >>= shift_bits - sub r8, r8, r6 // shift_bits -= 16 + cnt - str r5, [r0, #BUF_POS] - lsl r3, r3, r8 // next_bits <<= shift_bits - rsb r6, r8, #16 // cnt = cnt + 32 - shift_bits - eor r7, r7, r3 // dif ^= next_bits - b 9f - -2: // refill_eob - rsb r5, r6, #8 // c = 8 - cnt -3: - cmp r3, r4 - bge 4f - ldrb r8, [r3], #1 - lsl r8, r8, r5 - eor r7, r7, r8 - subs r5, r5, #8 - bge 3b - -4: // refill_eob_end + subs r5, r5, r4 + bhi 6f + + ldr r8, [r3] // next_bits + rsb r5, r6, #16 + add r4, r6, #16 // shift_bits = cnt + 16 + mvn r8, r8 + lsr r5, r5, #3 // num_bytes_read + rev r8, r8 // next_bits = bswap(next_bits) + lsr r8, r8, r4 // next_bits >>= shift_bits + +2: // refill_end + add r3, r3, r5 + add r6, r6, r5, lsl #3 // cnt += num_bits_read str r3, [r0, #BUF_POS] - rsb r6, r5, #8 // cnt = 8 - c -9: +3: // refill_end2 + orr r7, r7, r8 // dif |= next_bits + +4: // end str r6, [r0, #CNT] str r7, [r0, #DIF] - mov r0, lr add sp, sp, #48 - pop {r4-r10,pc} + +5: // pad_with_ones + add r8, r6, #-240 + lsr r8, r8, r8 + b 3b + +6: // refill_eob + cmp r3, r4 + bhs 5b + + ldr r8, [r4, #-4] + lsl r5, r5, #3 + lsr r8, r8, r5 + add r5, r6, #16 + mvn r8, r8 + sub r4, r4, r3 // num_bytes_left + rev r8, r8 + lsr r8, r8, r5 + rsb r5, r6, #16 + lsr r5, r5, #3 + cmp r5, r4 + it hs + movhs r5, r4 + b 2b endfunc function msac_decode_symbol_adapt8_neon, export=1 @@ -414,53 +421,38 @@ function msac_decode_hi_tok_neon, export=1 sub r4, r4, r3 // rng = u - v clz r5, r4 // clz(rng) eor r5, r5, #16 // d = clz(rng) ^ 16 - mvn r7, r7 // ~dif - add r7, r7, r3, lsl #16 // ~dif + (v << 16) + sub r7, r7, r3, lsl #16 // dif - (v << 16) lsl r4, r4, r5 // rng << d subs r6, r6, r5 // cnt -= d - lsl r7, r7, r5 // (~dif + (v << 16)) << d + lsl r7, r7, r5 // (dif - (v << 16)) << d str r4, [r0, #RNG] vdup.16 d1, r4 - mvn r7, r7 // ~dif - bhs 9f + bhs 5f // refill ldr r3, [r0, #BUF_POS] // BUF_POS ldr r4, [r0, #BUF_END] // BUF_END add r5, r3, #4 - cmp r5, r4 - bgt 2f - - ldr r3, [r3] // next_bits - add r8, r6, #23 // shift_bits = cnt + 23 - add r6, r6, #16 // cnt += 16 - rev r3, r3 // next_bits = bswap(next_bits) - sub r5, r5, r8, lsr #3 // buf_pos -= shift_bits >> 3 - and r8, r8, #24 // shift_bits &= 24 - lsr r3, r3, r8 // next_bits >>= shift_bits - sub r8, r8, r6 // shift_bits -= 16 + cnt - str r5, [r0, #BUF_POS] - lsl r3, r3, r8 // next_bits <<= shift_bits - rsb r6, r8, #16 // cnt = cnt + 32 - shift_bits - eor r7, r7, r3 // dif ^= next_bits - b 9f - -2: // refill_eob - rsb r5, r6, #8 // c = 40 - cnt -3: - cmp r3, r4 - bge 4f - ldrb r8, [r3], #1 - lsl r8, r8, r5 - eor r7, r7, r8 - subs r5, r5, #8 - bge 3b - -4: // refill_eob_end + subs r5, r5, r4 + bhi 7f + + ldr r8, [r3] // next_bits + rsb r5, r6, #16 + add r4, r6, #16 // shift_bits = cnt + 16 + mvn r8, r8 + lsr r5, r5, #3 // num_bytes_read + rev r8, r8 // next_bits = bswap(next_bits) + lsr r8, r8, r4 // next_bits >>= shift_bits + +3: // refill_end + add r3, r3, r5 + add r6, r6, r5, lsl #3 // cnt += num_bits_read str r3, [r0, #BUF_POS] - rsb r6, r5, #8 // cnt = 40 - c -9: +4: // refill_end2 + orr r7, r7, r8 // dif |= next_bits + +5: // end lsl lr, lr, #1 sub lr, lr, #5 lsr r12, r7, #16 @@ -473,6 +465,30 @@ function msac_decode_hi_tok_neon, export=1 str r7, [r0, #DIF] lsr r0, r2, #1 pop {r4-r10,pc} + +6: // pad_with_ones + add r8, r6, #-240 + lsr r8, r8, r8 + b 4b + +7: // refill_eob + cmp r3, r4 + bhs 6b + + ldr r8, [r4, #-4] + lsl r5, r5, #3 + lsr r8, r8, r5 + add r5, r6, #16 + mvn r8, r8 + sub r4, r4, r3 // num_bytes_left + rev r8, r8 + lsr r8, r8, r5 + rsb r5, r6, #16 + lsr r5, r5, #3 + cmp r5, r4 + it hs + movhs r5, r4 + b 3b endfunc function msac_decode_bool_equi_neon, export=1 @@ -493,7 +509,6 @@ function msac_decode_bool_equi_neon, export=1 movhs r7, r8 // if (ret) dif = dif - vw; clz r5, r4 // clz(rng) - mvn r7, r7 // ~dif eor r5, r5, #16 // d = clz(rng) ^ 16 mov lr, r2 b L(renorm2) @@ -519,7 +534,6 @@ function msac_decode_bool_neon, export=1 movhs r7, r8 // if (ret) dif = dif - vw; clz r5, r4 // clz(rng) - mvn r7, r7 // ~dif eor r5, r5, #16 // d = clz(rng) ^ 16 mov lr, r2 b L(renorm2) @@ -549,7 +563,6 @@ function msac_decode_bool_adapt_neon, export=1 cmp r10, #0 clz r5, r4 // clz(rng) - mvn r7, r7 // ~dif eor r5, r5, #16 // d = clz(rng) ^ 16 mov lr, r2 |