ref: b58c99eb71aeca19f27757e10c1392662781b65d
parent: c369daf3ea163d0e1a72e43315d0023b93e40c4e
author: Yaowu Xu <[email protected]>
date: Tue Jul 7 14:29:14 EDT 2015
Remove clamp operations. The clamp calls with INT32_MIN and INT32_MAX have no effect at all on int values passed in, therefore this commit removes those effectless clamps and also adds more const intermediate results to make the code more readable. Change-Id: I66d8811f58bb74ec31cbec9a6c441983a662352e
--- a/vp9/encoder/vp9_quantize.c
+++ b/vp9/encoder/vp9_quantize.c
@@ -57,17 +57,14 @@
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
if (!skip_block) {- const int rc = 0;
- const int coeff = coeff_ptr[rc];
+ const int coeff = coeff_ptr[0];
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-
- const int64_t tmp =
- (clamp(abs_coeff + round_ptr[rc != 0], INT32_MIN, INT32_MAX) *
- quant) >> 16;
- qcoeff_ptr[rc] = (tran_low_t)((tmp ^ coeff_sign) - coeff_sign);
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr;
- if (tmp)
+ const int64_t tmp = abs_coeff + round_ptr[0];
+ const uint32_t abs_qcoeff = (uint32_t)((tmp * quant) >> 16);
+ qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+ dqcoeff_ptr[0] = qcoeff_ptr[0] * dequant_ptr;
+ if (abs_qcoeff)
eob = 0;
}
*eob_ptr = eob + 1;
@@ -89,7 +86,6 @@
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
if (!skip_block) {-
tmp = clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
INT16_MIN, INT16_MAX);
tmp = (tmp * quant) >> 15;
@@ -117,17 +113,14 @@
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
if (!skip_block) {- const int rc = 0;
- const int coeff = coeff_ptr[rc];
+ const int coeff = coeff_ptr[0];
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-
- const int64_t tmp =
- (clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
- INT32_MIN, INT32_MAX) * quant) >> 15;
- qcoeff_ptr[rc] = (tran_low_t)((tmp ^ coeff_sign) - coeff_sign);
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr / 2;
- if (tmp)
+ const int64_t tmp = abs_coeff + ROUND_POWER_OF_TWO(round_ptr[0], 1);
+ const uint32_t abs_qcoeff = (uint32_t)((tmp * quant) >> 15);
+ qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+ dqcoeff_ptr[0] = qcoeff_ptr[0] * dequant_ptr / 2;
+ if (abs_qcoeff)
eob = 0;
}
*eob_ptr = eob + 1;
@@ -207,15 +200,11 @@
const int coeff = coeff_ptr[rc];
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-
- const int64_t tmp =
- (clamp(abs_coeff + round_ptr[rc != 0], INT32_MIN, INT32_MAX) *
- quant_ptr[rc != 0]) >> 16;
-
- qcoeff_ptr[rc] = (tran_low_t)((tmp ^ coeff_sign) - coeff_sign);
+ const int64_t tmp = abs_coeff + round_ptr[rc != 0];
+ const uint32_t abs_qcoeff = (uint32_t)((tmp * quant_ptr[rc != 0]) >> 16);
+ qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
-
- if (tmp)
+ if (abs_qcoeff)
eob = i;
}
}
@@ -287,21 +276,21 @@
if (!skip_block) { for (i = 0; i < n_coeffs; i++) {+ uint32_t abs_qcoeff = 0;
const int rc = scan[i];
const int coeff = coeff_ptr[rc];
const int coeff_sign = (coeff >> 31);
- int64_t tmp = 0;
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
if (abs_coeff >= (dequant_ptr[rc != 0] >> 2)) {- tmp = clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
- INT32_MIN, INT32_MAX);
- tmp = (tmp * quant_ptr[rc != 0]) >> 15;
- qcoeff_ptr[rc] = (tran_low_t)((tmp ^ coeff_sign) - coeff_sign);
+ const int64_t tmp = abs_coeff
+ + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
+ abs_qcoeff = (uint32_t) ((tmp * quant_ptr[rc != 0]) >> 15);
+ qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
}
- if (tmp)
+ if (abs_qcoeff)
eob = i;
}
}
@@ -398,14 +387,13 @@
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
if (abs_coeff >= zbins[rc != 0]) {- int64_t tmp = clamp(abs_coeff + round_ptr[rc != 0],
- INT32_MIN, INT32_MAX);
- tmp = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
- quant_shift_ptr[rc != 0]) >> 16; // quantization
- qcoeff_ptr[rc] = (tran_low_t)((tmp ^ coeff_sign) - coeff_sign);
+ const int64_t tmp1 = abs_coeff + round_ptr[rc != 0];
+ const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
+ const uint32_t abs_qcoeff =
+ (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 16);
+ qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
-
- if (tmp)
+ if (abs_qcoeff)
eob = i;
}
}
@@ -513,16 +501,14 @@
const int coeff = coeff_ptr[rc];
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
- int64_t tmp = clamp(abs_coeff +
- ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
- INT32_MIN, INT32_MAX);
- tmp = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
- quant_shift_ptr[rc != 0]) >> 15;
-
- qcoeff_ptr[rc] = (tran_low_t)((tmp ^ coeff_sign) - coeff_sign);
+ const int64_t tmp1 = abs_coeff
+ + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
+ const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
+ const uint32_t abs_qcoeff =
+ (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 15);
+ qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
-
- if (tmp)
+ if (abs_qcoeff)
eob = idx_arr[i];
}
}
--- a/vp9/encoder/x86/vp9_highbd_quantize_intrin_sse2.c
+++ b/vp9/encoder/x86/vp9_highbd_quantize_intrin_sse2.c
@@ -85,13 +85,13 @@
for (j = 0; j < 4; j++) { if (test & (1 << (4 * j))) {int k = 4 * i + j;
- int64_t tmp = clamp(abs_coeff[j] + round_ptr[k != 0],
- INT32_MIN, INT32_MAX);
- tmp = ((((tmp * quant_ptr[k != 0]) >> 16) + tmp) *
- quant_shift_ptr[k != 0]) >> 16; // quantization
- qcoeff_ptr[k] = (tmp ^ coeff_sign[j]) - coeff_sign[j];
+ const int64_t tmp1 = abs_coeff[j] + round_ptr[k != 0];
+ const int64_t tmp2 = ((tmp1 * quant_ptr[k != 0]) >> 16) + tmp1;
+ const uint32_t abs_qcoeff =
+ (uint32_t)((tmp2 * quant_shift_ptr[k != 0]) >> 16);
+ qcoeff_ptr[k] = (int)(abs_qcoeff ^ coeff_sign[j]) - coeff_sign[j];
dqcoeff_ptr[k] = qcoeff_ptr[k] * dequant_ptr[k != 0];
- if (tmp)
+ if (abs_qcoeff)
eob_i = iscan[k] > eob_i ? iscan[k] : eob_i;
}
}
@@ -162,17 +162,15 @@
const int rc = idx_arr[i];
const int coeff = coeff_ptr[rc];
const int coeff_sign = (coeff >> 31);
- int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
- int64_t tmp = clamp(abs_coeff +
- ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
- INT32_MIN, INT32_MAX);
- tmp = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
- quant_shift_ptr[rc != 0]) >> 15;
-
- qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
+ const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+ const int64_t tmp1 = abs_coeff
+ + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
+ const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
+ const uint32_t abs_qcoeff =
+ (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 15);
+ qcoeff_ptr[rc] = (int)(abs_qcoeff ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
-
- if (tmp)
+ if (abs_qcoeff)
eob = iscan[idx_arr[i]] > eob ? iscan[idx_arr[i]] : eob;
}
}