lc->scaling_factor_num, lc->scaling_factor_den, &width,
&height);
- // For low resolutions: set phase of the filter = 8 (for symmetric averaging
- // filter), use bilinear for now.
- if (width <= 320 && height <= 240) {
+ // For resolutions <= QVGA: set phase of the filter = 8 (for symmetric
+ // averaging filter), use bilinear for now.
+ if (width * height <= 320 * 240) {
cpi->svc.downsample_filter_type[cpi->svc.spatial_layer_id] = BILINEAR;
cpi->svc.downsample_filter_phase[cpi->svc.spatial_layer_id] = 8;
}
DSP_SRCS-yes += quantize.c
DSP_SRCS-yes += quantize.h
+DSP_SRCS-$(HAVE_SSE2) += x86/quantize_x86.h
DSP_SRCS-$(HAVE_SSE2) += x86/quantize_sse2.c
DSP_SRCS-$(HAVE_SSSE3) += x86/quantize_ssse3.c
DSP_SRCS-$(HAVE_AVX) += x86/quantize_avx.c
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
+#include "vpx_dsp/x86/quantize_x86.h"
void vpx_quantize_b_avx(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
__m128i all_zero;
- __m128i qtmp0, qtmp1;
- __m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
- __m128i eob = zero, eob0, eob1;
+ __m128i eob = zero, eob0;
(void)scan_ptr;
(void)skip_block;
*eob_ptr = 0;
- // Setup global values.
- zbin = _mm_load_si128((const __m128i *)zbin_ptr);
- // x86 has no "greater *or equal*" comparison. Subtract 1 from zbin so
- // it is a strict "greater" comparison.
- zbin = _mm_sub_epi16(zbin, _mm_set1_epi16(1));
- round = _mm_load_si128((const __m128i *)round_ptr);
- quant = _mm_load_si128((const __m128i *)quant_ptr);
- dequant = _mm_load_si128((const __m128i *)dequant_ptr);
- shift = _mm_load_si128((const __m128i *)quant_shift_ptr);
+ load_b_values(zbin_ptr, &zbin, round_ptr, &round, quant_ptr, &quant,
+ dequant_ptr, &dequant, quant_shift_ptr, &shift);
// Do DC and first 15 AC.
coeff0 = load_tran_low(coeff_ptr);
shift = _mm_unpackhi_epi64(shift, shift);
dequant = _mm_unpackhi_epi64(dequant, dequant);
} else {
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
round = _mm_unpackhi_epi64(round, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
quant = _mm_unpackhi_epi64(quant, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
// Reinsert signs
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
store_tran_low(qcoeff0, qcoeff_ptr);
store_tran_low(qcoeff1, qcoeff_ptr + 8);
- coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
+ coeff0 = calculate_dqcoeff(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
- coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
+ coeff1 = calculate_dqcoeff(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
- // Scan for eob.
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
- // Add one to convert from indices to counts
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob = _mm_max_epi16(eob, eob1);
+ eob = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, 0,
+ zero);
}
// AC only loop.
continue;
}
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
store_tran_low(qcoeff0, qcoeff_ptr + index);
store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
- coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
- coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
+ coeff0 = calculate_dqcoeff(qcoeff0, dequant);
+ coeff1 = calculate_dqcoeff(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob0 = _mm_max_epi16(eob0, eob1);
+ eob0 = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr,
+ index, zero);
eob = _mm_max_epi16(eob, eob0);
}
- // Accumulate eob.
- {
- __m128i eob_shuffled;
- eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
- eob = _mm_max_epi16(eob, eob_shuffled);
- *eob_ptr = _mm_extract_epi16(eob, 1);
- }
+ *eob_ptr = accumulate_eob(eob);
}
void vpx_quantize_b_32x32_avx(
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
__m128i all_zero;
- __m128i qtmp0, qtmp1;
- __m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
- __m128i eob = zero, eob0, eob1;
+ __m128i eob = zero, eob0;
(void)scan_ptr;
(void)n_coeffs;
(void)skip_block;
assert(!skip_block);
- *eob_ptr = 0;
-
// Setup global values.
// The 32x32 halves zbin and round.
zbin = _mm_load_si128((const __m128i *)zbin_ptr);
qcoeff1 = _mm_abs_epi16(coeff1);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
- zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC
+ zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC.
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
all_zero = _mm_or_si128(cmp_mask0, cmp_mask1);
shift = _mm_unpackhi_epi64(shift, shift);
dequant = _mm_unpackhi_epi64(dequant, dequant);
} else {
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
round = _mm_unpackhi_epi64(round, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
quant = _mm_unpackhi_epi64(quant, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
- // Reinsert signs
+ // Reinsert signs.
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
- // Mask out zbin threshold coeffs
+ // Mask out zbin threshold coeffs.
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
coeff0 = _mm_abs_epi16(qcoeff0);
coeff1 = _mm_abs_epi16(qcoeff1);
- coeff0 = _mm_mullo_epi16(coeff0, dequant);
+ coeff0 = calculate_dqcoeff(coeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
- coeff1 = _mm_mullo_epi16(coeff1, dequant);
+ coeff1 = calculate_dqcoeff(coeff1, dequant);
// "Divide" by 2.
coeff0 = _mm_srli_epi16(coeff0, 1);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
- // Scan for eob.
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
- // Add one to convert from indices to counts
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob = _mm_max_epi16(eob, eob1);
+ eob = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, 0,
+ zero);
}
// AC only loop.
continue;
}
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
coeff0 = _mm_abs_epi16(qcoeff0);
coeff1 = _mm_abs_epi16(qcoeff1);
- coeff0 = _mm_mullo_epi16(coeff0, dequant);
- coeff1 = _mm_mullo_epi16(coeff1, dequant);
+ coeff0 = calculate_dqcoeff(coeff0, dequant);
+ coeff1 = calculate_dqcoeff(coeff1, dequant);
coeff0 = _mm_srli_epi16(coeff0, 1);
coeff1 = _mm_srli_epi16(coeff1, 1);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob0 = _mm_max_epi16(eob0, eob1);
+ eob0 = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr,
+ index, zero);
eob = _mm_max_epi16(eob, eob0);
}
- // Accumulate eob.
- {
- __m128i eob_shuffled;
- eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
- eob = _mm_max_epi16(eob, eob_shuffled);
- *eob_ptr = _mm_extract_epi16(eob, 1);
- }
+ *eob_ptr = accumulate_eob(eob);
}
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
+#include "vpx_dsp/x86/quantize_x86.h"
void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
__m128i coeff0, coeff1, coeff0_sign, coeff1_sign;
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
- __m128i qtmp0, qtmp1;
- __m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
- __m128i eob, eob0, eob1;
+ __m128i eob, eob0;
(void)scan_ptr;
(void)skip_block;
assert(!skip_block);
// Setup global values.
- zbin = _mm_load_si128((const __m128i *)zbin_ptr);
- round = _mm_load_si128((const __m128i *)round_ptr);
- quant = _mm_load_si128((const __m128i *)quant_ptr);
- zbin = _mm_sub_epi16(zbin, _mm_set1_epi16(1));
- dequant = _mm_load_si128((const __m128i *)dequant_ptr);
- shift = _mm_load_si128((const __m128i *)quant_shift_ptr);
+ load_b_values(zbin_ptr, &zbin, round_ptr, &round, quant_ptr, &quant,
+ dequant_ptr, &dequant, quant_shift_ptr, &shift);
// Do DC and first 15 AC.
coeff0 = load_tran_low(coeff_ptr);
// Poor man's abs().
coeff0_sign = _mm_srai_epi16(coeff0, 15);
coeff1_sign = _mm_srai_epi16(coeff1, 15);
- qcoeff0 = _mm_xor_si128(coeff0, coeff0_sign);
- qcoeff1 = _mm_xor_si128(coeff1, coeff1_sign);
- qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
- qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
+ qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign);
+ qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
- round = _mm_unpackhi_epi64(round, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
+ round = _mm_unpackhi_epi64(round, round);
quant = _mm_unpackhi_epi64(quant, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
// Reinsert signs
- qcoeff0 = _mm_xor_si128(qcoeff0, coeff0_sign);
- qcoeff1 = _mm_xor_si128(qcoeff1, coeff1_sign);
- qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
- qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
+ qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign);
+ qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign);
// Mask out zbin threshold coeffs
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
store_tran_low(qcoeff0, qcoeff_ptr);
store_tran_low(qcoeff1, qcoeff_ptr + 8);
- coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
+ coeff0 = calculate_dqcoeff(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
- coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
+ coeff1 = calculate_dqcoeff(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
- // Scan for eob.
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
- // Add one to convert from indices to counts
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob = _mm_max_epi16(eob, eob1);
+ eob =
+ scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, 0, zero);
// AC only loop.
while (index < n_coeffs) {
coeff0_sign = _mm_srai_epi16(coeff0, 15);
coeff1_sign = _mm_srai_epi16(coeff1, 15);
- qcoeff0 = _mm_xor_si128(coeff0, coeff0_sign);
- qcoeff1 = _mm_xor_si128(coeff1, coeff1_sign);
- qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
- qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
+ qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign);
+ qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
-
- qcoeff0 = _mm_xor_si128(qcoeff0, coeff0_sign);
- qcoeff1 = _mm_xor_si128(qcoeff1, coeff1_sign);
- qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
- qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
+ qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign);
+ qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign);
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
store_tran_low(qcoeff0, qcoeff_ptr + index);
store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
- coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
- coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
+ coeff0 = calculate_dqcoeff(qcoeff0, dequant);
+ coeff1 = calculate_dqcoeff(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob0 = _mm_max_epi16(eob0, eob1);
+ eob0 = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr,
+ index, zero);
eob = _mm_max_epi16(eob, eob0);
index += 16;
}
- // Accumulate eob.
- {
- __m128i eob_shuffled;
- eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
- eob = _mm_max_epi16(eob, eob_shuffled);
- *eob_ptr = _mm_extract_epi16(eob, 1);
- }
+ *eob_ptr = accumulate_eob(eob);
}
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
+#include "vpx_dsp/x86/quantize_x86.h"
void vpx_quantize_b_ssse3(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan_ptr, const int16_t *iscan_ptr) {
const __m128i zero = _mm_setzero_si128();
- intptr_t index = 16;
+ int index = 16;
__m128i zbin, round, quant, dequant, shift;
__m128i coeff0, coeff1;
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
- __m128i qtmp0, qtmp1;
- __m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
- __m128i eob, eob0, eob1;
+ __m128i eob, eob0;
(void)scan_ptr;
(void)skip_block;
assert(!skip_block);
- // Setup global values.
- zbin = _mm_load_si128((const __m128i *)zbin_ptr);
- // x86 has no "greater *or equal*" comparison. Subtract 1 from zbin so
- // it is a strict "greater" comparison.
- zbin = _mm_sub_epi16(zbin, _mm_set1_epi16(1));
- round = _mm_load_si128((const __m128i *)round_ptr);
- quant = _mm_load_si128((const __m128i *)quant_ptr);
- dequant = _mm_load_si128((const __m128i *)dequant_ptr);
- shift = _mm_load_si128((const __m128i *)quant_shift_ptr);
+ load_b_values(zbin_ptr, &zbin, round_ptr, &round, quant_ptr, &quant,
+ dequant_ptr, &dequant, quant_shift_ptr, &shift);
// Do DC and first 15 AC.
coeff0 = load_tran_low(coeff_ptr);
zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
round = _mm_unpackhi_epi64(round, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
quant = _mm_unpackhi_epi64(quant, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
// Reinsert signs
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
store_tran_low(qcoeff0, qcoeff_ptr);
store_tran_low(qcoeff1, qcoeff_ptr + 8);
- coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
+ coeff0 = calculate_dqcoeff(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
- coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
+ coeff1 = calculate_dqcoeff(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
- // Scan for eob.
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
- // Add one to convert from indices to counts
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob = _mm_max_epi16(eob, eob1);
+ eob =
+ scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, 0, zero);
// AC only loop.
while (index < n_coeffs) {
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
store_tran_low(qcoeff0, qcoeff_ptr + index);
store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
- coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
- coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
+ coeff0 = calculate_dqcoeff(qcoeff0, dequant);
+ coeff1 = calculate_dqcoeff(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob0 = _mm_max_epi16(eob0, eob1);
+ eob0 = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr,
+ index, zero);
eob = _mm_max_epi16(eob, eob0);
index += 16;
}
- // Accumulate eob.
- {
- __m128i eob_shuffled;
- eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
- eob = _mm_max_epi16(eob, eob_shuffled);
- *eob_ptr = _mm_extract_epi16(eob, 1);
- }
+ *eob_ptr = accumulate_eob(eob);
}
void vpx_quantize_b_32x32_ssse3(
const int16_t *scan_ptr, const int16_t *iscan_ptr) {
const __m128i zero = _mm_setzero_si128();
const __m128i one = _mm_set1_epi16(1);
- intptr_t index = 16;
+ int index;
__m128i zbin, round, quant, dequant, shift;
__m128i coeff0, coeff1;
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
__m128i all_zero;
- __m128i qtmp0, qtmp1;
- __m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
- __m128i eob = zero, eob0, eob1;
+ __m128i eob = zero, eob0;
(void)scan_ptr;
(void)n_coeffs;
_mm_store_si128((__m128i *)(qcoeff_ptr + 12), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + 4), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + 12), zero);
-#endif
+#endif // CONFIG_HIGHBITDEPTH
round = _mm_unpackhi_epi64(round, round);
quant = _mm_unpackhi_epi64(quant, quant);
shift = _mm_unpackhi_epi64(shift, shift);
dequant = _mm_unpackhi_epi64(dequant, dequant);
} else {
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
round = _mm_unpackhi_epi64(round, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
quant = _mm_unpackhi_epi64(quant, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
// Reinsert signs.
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
coeff0 = _mm_abs_epi16(qcoeff0);
coeff1 = _mm_abs_epi16(qcoeff1);
- coeff0 = _mm_mullo_epi16(coeff0, dequant);
+ coeff0 = calculate_dqcoeff(coeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
- coeff1 = _mm_mullo_epi16(coeff1, dequant);
+ coeff1 = calculate_dqcoeff(coeff1, dequant);
// "Divide" by 2.
coeff0 = _mm_srli_epi16(coeff0, 1);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
- // Scan for eob.
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
- // Add one to convert from indices to counts.
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob = _mm_max_epi16(eob, eob1);
+ eob = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, 0,
+ zero);
}
// AC only loop.
_mm_store_si128((__m128i *)(qcoeff_ptr + index + 12), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + index + 4), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + index + 12), zero);
-#endif
+#endif // CONFIG_VP9_HIGHBITDEPTH
continue;
}
- qcoeff0 = _mm_adds_epi16(qcoeff0, round);
- qcoeff1 = _mm_adds_epi16(qcoeff1, round);
-
- qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
- qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
-
- qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
- qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
-
- qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
- qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
+ calculate_qcoeff(&qcoeff0, round, quant, shift);
+ calculate_qcoeff(&qcoeff1, round, quant, shift);
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
coeff0 = _mm_abs_epi16(qcoeff0);
coeff1 = _mm_abs_epi16(qcoeff1);
- coeff0 = _mm_mullo_epi16(coeff0, dequant);
- coeff1 = _mm_mullo_epi16(coeff1, dequant);
+ coeff0 = calculate_dqcoeff(coeff0, dequant);
+ coeff1 = calculate_dqcoeff(coeff1, dequant);
coeff0 = _mm_srli_epi16(coeff0, 1);
coeff1 = _mm_srli_epi16(coeff1, 1);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
- zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
- zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
- iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
- iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
- iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
- iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
- eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
- eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
- eob0 = _mm_max_epi16(eob0, eob1);
+ eob0 = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr,
+ index, zero);
eob = _mm_max_epi16(eob, eob0);
}
- {
- __m128i eob_shuffled;
- eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
- eob = _mm_max_epi16(eob, eob_shuffled);
- eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
- eob = _mm_max_epi16(eob, eob_shuffled);
- *eob_ptr = _mm_extract_epi16(eob, 1);
- }
+ *eob_ptr = accumulate_eob(eob);
}
--- /dev/null
+/*
+ * Copyright (c) 2017 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
+
+static INLINE void load_b_values(const int16_t *zbin_ptr, __m128i *zbin,
+ const int16_t *round_ptr, __m128i *round,
+ const int16_t *quant_ptr, __m128i *quant,
+ const int16_t *dequant_ptr, __m128i *dequant,
+ const int16_t *shift_ptr, __m128i *shift) {
+ *zbin = _mm_load_si128((const __m128i *)zbin_ptr);
+ *round = _mm_load_si128((const __m128i *)round_ptr);
+ *quant = _mm_load_si128((const __m128i *)quant_ptr);
+ *zbin = _mm_sub_epi16(*zbin, _mm_set1_epi16(1));
+ *dequant = _mm_load_si128((const __m128i *)dequant_ptr);
+ *shift = _mm_load_si128((const __m128i *)shift_ptr);
+}
+
+// With ssse3 and later abs() and sign() are preferred.
+static INLINE __m128i invert_sign_sse2(__m128i a, __m128i sign) {
+ a = _mm_xor_si128(a, sign);
+ return _mm_sub_epi16(a, sign);
+}
+
+static INLINE void calculate_qcoeff(__m128i *coeff, const __m128i round,
+ const __m128i quant, const __m128i shift) {
+ __m128i tmp, qcoeff;
+ qcoeff = _mm_adds_epi16(*coeff, round);
+ tmp = _mm_mulhi_epi16(qcoeff, quant);
+ qcoeff = _mm_add_epi16(tmp, qcoeff);
+ *coeff = _mm_mulhi_epi16(qcoeff, shift);
+}
+
+static INLINE __m128i calculate_dqcoeff(__m128i qcoeff, __m128i dequant) {
+ return _mm_mullo_epi16(qcoeff, dequant);
+}
+
+// Scan 16 values for eob reference in scan_ptr. Use masks (-1) from comparing
+// to zbin to add 1 to the index in 'scan'.
+static INLINE __m128i scan_for_eob(__m128i *coeff0, __m128i *coeff1,
+ const __m128i zbin_mask0,
+ const __m128i zbin_mask1,
+ const int16_t *scan_ptr, const int index,
+ const __m128i zero) {
+ const __m128i zero_coeff0 = _mm_cmpeq_epi16(*coeff0, zero);
+ const __m128i zero_coeff1 = _mm_cmpeq_epi16(*coeff1, zero);
+ __m128i scan0 = _mm_load_si128((const __m128i *)(scan_ptr + index));
+ __m128i scan1 = _mm_load_si128((const __m128i *)(scan_ptr + index + 8));
+ __m128i eob0, eob1;
+ // Add one to convert from indices to counts
+ scan0 = _mm_sub_epi16(scan0, zbin_mask0);
+ scan1 = _mm_sub_epi16(scan1, zbin_mask1);
+ eob0 = _mm_andnot_si128(zero_coeff0, scan0);
+ eob1 = _mm_andnot_si128(zero_coeff1, scan1);
+ return _mm_max_epi16(eob0, eob1);
+}
+
+static INLINE int16_t accumulate_eob(__m128i eob) {
+ __m128i eob_shuffled;
+ eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
+ eob = _mm_max_epi16(eob, eob_shuffled);
+ eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
+ eob = _mm_max_epi16(eob, eob_shuffled);
+ eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
+ eob = _mm_max_epi16(eob, eob_shuffled);
+ return _mm_extract_epi16(eob, 1);
+}
struct WebmInputContext *webm_ctx;
};
+static const arg_def_t help =
+ ARG_DEF(NULL, "help", 0, "Show usage options and exit");
static const arg_def_t looparg =
ARG_DEF(NULL, "loops", 1, "Number of times to decode the file");
static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1, "Codec to use");
ARG_DEF(NULL, "framestats", 1, "Output per-frame stats (.csv format)");
static const arg_def_t *all_args[] = {
- &codecarg, &use_yv12, &use_i420,
- &flipuvarg, &rawvideo, &noblitarg,
- &progressarg, &limitarg, &skiparg,
- &postprocarg, &summaryarg, &outputfile,
- &threadsarg, &frameparallelarg, &verbosearg,
- &scalearg, &fb_arg, &md5arg,
- &error_concealment, &continuearg,
+ &help, &codecarg, &use_yv12,
+ &use_i420, &flipuvarg, &rawvideo,
+ &noblitarg, &progressarg, &limitarg,
+ &skiparg, &postprocarg, &summaryarg,
+ &outputfile, &threadsarg, &frameparallelarg,
+ &verbosearg, &scalearg, &fb_arg,
+ &md5arg, &error_concealment, &continuearg,
#if CONFIG_VP9_HIGHBITDEPTH
&outbitdeptharg,
#endif
- &svcdecodingarg, &framestatsarg, NULL
+ &svcdecodingarg, &framestatsarg, NULL
};
#if CONFIG_VP8_DECODER
dst->d_h, mode);
}
#endif
-
-void usage_exit(void) {
+void show_help(FILE *fout, int shorthelp) {
int i;
- fprintf(stderr,
- "Usage: %s <options> filename\n\n"
- "Options:\n",
- exec_name);
- arg_show_usage(stderr, all_args);
+ fprintf(fout, "Usage: %s <options> filename\n\n", exec_name);
+
+ if (shorthelp) {
+ fprintf(fout, "Use --help to see the full list of options.\n");
+ return;
+ }
+
+ fprintf(fout, "Options:\n");
+ arg_show_usage(fout, all_args);
#if CONFIG_VP8_DECODER
- fprintf(stderr, "\nVP8 Postprocessing Options:\n");
- arg_show_usage(stderr, vp8_pp_args);
+ fprintf(fout, "\nVP8 Postprocessing Options:\n");
+ arg_show_usage(fout, vp8_pp_args);
#endif
- fprintf(stderr,
+ fprintf(fout,
"\nOutput File Patterns:\n\n"
" The -o argument specifies the name of the file(s) to "
"write to. If the\n argument does not include any escape "
"characters, the output will be\n written to a single file. "
"Otherwise, the filename will be calculated by\n expanding "
"the following escape characters:\n");
- fprintf(stderr,
+ fprintf(fout,
"\n\t%%w - Frame width"
"\n\t%%h - Frame height"
"\n\t%%<n> - Frame number, zero padded to <n> places (1..9)"
"\n\n Pattern arguments are only supported in conjunction "
"with the --yv12 and\n --i420 options. If the -o option is "
"not specified, the output will be\n directed to stdout.\n");
- fprintf(stderr, "\nIncluded decoders:\n\n");
+ fprintf(fout, "\nIncluded decoders:\n\n");
for (i = 0; i < get_vpx_decoder_count(); ++i) {
const VpxInterface *const decoder = get_vpx_decoder_by_index(i);
- fprintf(stderr, " %-6s - %s\n", decoder->name,
+ fprintf(fout, " %-6s - %s\n", decoder->name,
vpx_codec_iface_name(decoder->codec_interface()));
}
+}
+void usage_exit(void) {
+ show_help(stderr, 1);
exit(EXIT_FAILURE);
}
memset(&arg, 0, sizeof(arg));
arg.argv_step = 1;
- if (arg_match(&arg, &codecarg, argi)) {
+ if (arg_match(&arg, &help, argi)) {
+ show_help(stdout, 0);
+ exit(EXIT_SUCCESS);
+ } else if (arg_match(&arg, &codecarg, argi)) {
interface = get_vpx_decoder_by_name(arg.val);
if (!interface)
die("Error: Unrecognized argument (%s) to --codec\n", arg.val);
if (!fn) {
free(argv);
+ fprintf(stderr, "No input file specified!\n");
usage_exit();
}
/* Open file */
return 0;
}
+static const arg_def_t help =
+ ARG_DEF(NULL, "help", 0, "Show usage options and exit");
static const arg_def_t debugmode =
ARG_DEF("D", "debug", 0, "Debug mode (makes output deterministic)");
static const arg_def_t outputfile =
NULL, "test-16bit-internal", 0, "Force use of 16 bit internal buffer");
#endif
-static const arg_def_t *main_args[] = { &debugmode,
+static const arg_def_t *main_args[] = { &help,
+ &debugmode,
&outputfile,
&codecarg,
&passes,
static const arg_def_t *no_args[] = { NULL };
-void usage_exit(void) {
+void show_help(FILE *fout, int shorthelp) {
int i;
const int num_encoder = get_vpx_encoder_count();
- fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
+ fprintf(fout, "Usage: %s <options> -o dst_filename src_filename \n",
exec_name);
- fprintf(stderr, "\nOptions:\n");
- arg_show_usage(stderr, main_args);
- fprintf(stderr, "\nEncoder Global Options:\n");
- arg_show_usage(stderr, global_args);
- fprintf(stderr, "\nRate Control Options:\n");
- arg_show_usage(stderr, rc_args);
- fprintf(stderr, "\nTwopass Rate Control Options:\n");
- arg_show_usage(stderr, rc_twopass_args);
- fprintf(stderr, "\nKeyframe Placement Options:\n");
- arg_show_usage(stderr, kf_args);
+ if (shorthelp) {
+ fprintf(fout, "Use --help to see the full list of options.\n");
+ return;
+ }
+
+ fprintf(fout, "\nOptions:\n");
+ arg_show_usage(fout, main_args);
+ fprintf(fout, "\nEncoder Global Options:\n");
+ arg_show_usage(fout, global_args);
+ fprintf(fout, "\nRate Control Options:\n");
+ arg_show_usage(fout, rc_args);
+ fprintf(fout, "\nTwopass Rate Control Options:\n");
+ arg_show_usage(fout, rc_twopass_args);
+ fprintf(fout, "\nKeyframe Placement Options:\n");
+ arg_show_usage(fout, kf_args);
#if CONFIG_VP8_ENCODER
- fprintf(stderr, "\nVP8 Specific Options:\n");
- arg_show_usage(stderr, vp8_args);
+ fprintf(fout, "\nVP8 Specific Options:\n");
+ arg_show_usage(fout, vp8_args);
#endif
#if CONFIG_VP9_ENCODER
- fprintf(stderr, "\nVP9 Specific Options:\n");
- arg_show_usage(stderr, vp9_args);
+ fprintf(fout, "\nVP9 Specific Options:\n");
+ arg_show_usage(fout, vp9_args);
#endif
- fprintf(stderr,
+ fprintf(fout,
"\nStream timebase (--timebase):\n"
" The desired precision of timestamps in the output, expressed\n"
" in fractional seconds. Default is 1/1000.\n");
- fprintf(stderr, "\nIncluded encoders:\n\n");
+ fprintf(fout, "\nIncluded encoders:\n\n");
for (i = 0; i < num_encoder; ++i) {
const VpxInterface *const encoder = get_vpx_encoder_by_index(i);
const char *defstr = (i == (num_encoder - 1)) ? "(default)" : "";
- fprintf(stderr, " %-6s - %s %s\n", encoder->name,
+ fprintf(fout, " %-6s - %s %s\n", encoder->name,
vpx_codec_iface_name(encoder->codec_interface()), defstr);
}
- fprintf(stderr, "\n ");
- fprintf(stderr, "Use --codec to switch to a non-default encoder.\n\n");
+ fprintf(fout, "\n ");
+ fprintf(fout, "Use --codec to switch to a non-default encoder.\n\n");
+}
+void usage_exit(void) {
+ show_help(stderr, 1);
exit(EXIT_FAILURE);
}
for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
arg.argv_step = 1;
- if (arg_match(&arg, &codecarg, argi)) {
+ if (arg_match(&arg, &help, argi)) {
+ show_help(stdout, 0);
+ exit(EXIT_SUCCESS);
+ } else if (arg_match(&arg, &codecarg, argi)) {
global->codec = get_vpx_encoder_by_name(arg.val);
if (!global->codec)
die("Error: Unrecognized argument (%s) to --codec\n", arg.val);
memset(&input, 0, sizeof(input));
exec_name = argv_[0];
- if (argc < 3) usage_exit();
-
/* Setup default input stream settings */
input.framerate.numerator = 30;
input.framerate.denominator = 1;
argv = argv_dup(argc - 1, argv_ + 1);
parse_global_config(&global, argv);
+ if (argc < 3) usage_exit();
+
switch (global.color_type) {
case I420: input.fmt = VPX_IMG_FMT_I420; break;
case I422: input.fmt = VPX_IMG_FMT_I422; break;
/* Handle non-option arguments */
input.filename = argv[0];
- if (!input.filename) usage_exit();
+ if (!input.filename) {
+ fprintf(stderr, "No input file specified!\n");
+ usage_exit();
+ }
/* Decide if other chroma subsamplings than 4:2:0 are supported */
if (global.codec->fourcc == VP9_FOURCC) input.only_i420 = 0;