diff options
Diffstat (limited to 'third_party/aom/av1/common/idct.c')
-rw-r--r-- | third_party/aom/av1/common/idct.c | 274 |
1 files changed, 107 insertions, 167 deletions
diff --git a/third_party/aom/av1/common/idct.c b/third_party/aom/av1/common/idct.c index bc758eb57..2c1cb9827 100644 --- a/third_party/aom/av1/common/idct.c +++ b/third_party/aom/av1/common/idct.c @@ -31,21 +31,16 @@ int av1_get_tx_scale(const TX_SIZE tx_size) { // that input and output could be the same buffer. // idct -static void highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, - int stride, int eob, int bd) { +void av1_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride, + int eob, int bd) { if (eob > 1) av1_highbd_iwht4x4_16_add(input, dest, stride, bd); else av1_highbd_iwht4x4_1_add(input, dest, stride, bd); } -static const int32_t *cast_to_int32(const tran_low_t *input) { - assert(sizeof(int32_t) == sizeof(tran_low_t)); - return (const int32_t *)input; -} - -void av1_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]); int eob = txfm_param->eob; int bd = txfm_param->bd; @@ -54,206 +49,150 @@ void av1_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest, const TX_TYPE tx_type = txfm_param->tx_type; if (lossless) { assert(tx_type == DCT_DCT); - highbd_iwht4x4_add(input, dest, stride, eob, bd); + av1_highbd_iwht4x4_add(input, dest, stride, eob, bd); return; } - switch (tx_type) { - // Assembly version doesn't support some transform types, so use C version - // for those. - case V_DCT: - case H_DCT: - case V_ADST: - case H_ADST: - case V_FLIPADST: - case H_FLIPADST: - case IDTX: - av1_inv_txfm2d_add_4x4_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, - bd); - break; - default: - av1_inv_txfm2d_add_4x4(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, - bd); - break; - } + + av1_inv_txfm2d_add_4x4_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, bd); } -static void highbd_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]); const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_4x8(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_4x8_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]); const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_8x4(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_8x4_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_8x16(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_16x32(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_8x16(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_16x32_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_16x8(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_32x16(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_16x8(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_32x16_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_16x32(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_16x4(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_16x32(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_16x4_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_32x16(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_4x16(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_32x16(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_4x16_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_16x4(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_32x8(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_16x4(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_32x8_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_4x16(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_8x32(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_4x16(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_8x32_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_32x8(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_32x64(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_32x8(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_32x64_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_8x32(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_64x32(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_8x32(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_64x32_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_32x64(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_16x64(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_32x64(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_16x64_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_64x32(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_64x16(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_64x32(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_64x16_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_16x64(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { + int bd = txfm_param->bd; + const TX_TYPE tx_type = txfm_param->tx_type; const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_16x64(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); -} -static void highbd_inv_txfm_add_64x16(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { - const int32_t *src = cast_to_int32(input); - av1_inv_txfm2d_add_64x16(src, CONVERT_TO_SHORTPTR(dest), stride, - txfm_param->tx_type, txfm_param->bd); + av1_inv_txfm2d_add_8x8_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, bd); } -static void highbd_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { int bd = txfm_param->bd; const TX_TYPE tx_type = txfm_param->tx_type; const int32_t *src = cast_to_int32(input); - switch (tx_type) { - // Assembly version doesn't support some transform types, so use C version - // for those. - case V_DCT: - case H_DCT: - case V_ADST: - case H_ADST: - case V_FLIPADST: - case H_FLIPADST: - case IDTX: - av1_inv_txfm2d_add_8x8_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, - bd); - break; - default: - av1_inv_txfm2d_add_8x8(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, + + av1_inv_txfm2d_add_16x16_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, bd); - break; - } } -static void highbd_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { - int bd = txfm_param->bd; - const TX_TYPE tx_type = txfm_param->tx_type; +void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int32_t *src = cast_to_int32(input); - switch (tx_type) { - // Assembly version doesn't support some transform types, so use C version - // for those. - case V_DCT: - case H_DCT: - case V_ADST: - case H_ADST: - case V_FLIPADST: - case H_FLIPADST: - case IDTX: - av1_inv_txfm2d_add_16x16_c(src, CONVERT_TO_SHORTPTR(dest), stride, - tx_type, bd); - break; - default: - av1_inv_txfm2d_add_16x16(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, - bd); - break; - } + av1_inv_txfm2d_add_8x16_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); } -static void highbd_inv_txfm_add_32x32(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { + const int32_t *src = cast_to_int32(input); + av1_inv_txfm2d_add_16x8_c(src, CONVERT_TO_SHORTPTR(dest), stride, + txfm_param->tx_type, txfm_param->bd); +} + +void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int bd = txfm_param->bd; const TX_TYPE tx_type = txfm_param->tx_type; const int32_t *src = cast_to_int32(input); - switch (tx_type) { - case DCT_DCT: - av1_inv_txfm2d_add_32x32(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, - bd); - break; - // Assembly version doesn't support IDTX, so use C version for it. - case IDTX: - av1_inv_txfm2d_add_32x32_c(src, CONVERT_TO_SHORTPTR(dest), stride, - tx_type, bd); - break; - default: assert(0); - } + av1_inv_txfm2d_add_32x32_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, + bd); } -static void highbd_inv_txfm_add_64x64(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_64x64_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { const int bd = txfm_param->bd; const TX_TYPE tx_type = txfm_param->tx_type; const int32_t *src = cast_to_int32(input); assert(tx_type == DCT_DCT); - av1_inv_txfm2d_add_64x64(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, bd); + av1_inv_txfm2d_add_64x64_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, + bd); } static void init_txfm_param(const MACROBLOCKD *xd, int plane, TX_SIZE tx_size, @@ -270,70 +209,70 @@ static void init_txfm_param(const MACROBLOCKD *xd, int plane, TX_SIZE tx_size, txfm_param->tx_size, is_inter_block(xd->mi[0]), reduced_tx_set); } -static void highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, - int stride, const TxfmParam *txfm_param) { +void av1_highbd_inv_txfm_add_c(const tran_low_t *input, uint8_t *dest, + int stride, const TxfmParam *txfm_param) { assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]); const TX_SIZE tx_size = txfm_param->tx_size; switch (tx_size) { case TX_32X32: - highbd_inv_txfm_add_32x32(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_32x32_c(input, dest, stride, txfm_param); break; case TX_16X16: - highbd_inv_txfm_add_16x16(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_16x16_c(input, dest, stride, txfm_param); break; case TX_8X8: - highbd_inv_txfm_add_8x8(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_8x8_c(input, dest, stride, txfm_param); break; case TX_4X8: - highbd_inv_txfm_add_4x8(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_4x8(input, dest, stride, txfm_param); break; case TX_8X4: - highbd_inv_txfm_add_8x4(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_8x4(input, dest, stride, txfm_param); break; case TX_8X16: - highbd_inv_txfm_add_8x16(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_8x16_c(input, dest, stride, txfm_param); break; case TX_16X8: - highbd_inv_txfm_add_16x8(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_16x8_c(input, dest, stride, txfm_param); break; case TX_16X32: - highbd_inv_txfm_add_16x32(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_16x32(input, dest, stride, txfm_param); break; case TX_32X16: - highbd_inv_txfm_add_32x16(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_32x16(input, dest, stride, txfm_param); break; case TX_64X64: - highbd_inv_txfm_add_64x64(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_64x64_c(input, dest, stride, txfm_param); break; case TX_32X64: - highbd_inv_txfm_add_32x64(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_32x64(input, dest, stride, txfm_param); break; case TX_64X32: - highbd_inv_txfm_add_64x32(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_64x32(input, dest, stride, txfm_param); break; case TX_16X64: - highbd_inv_txfm_add_16x64(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_16x64(input, dest, stride, txfm_param); break; case TX_64X16: - highbd_inv_txfm_add_64x16(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_64x16(input, dest, stride, txfm_param); break; case TX_4X4: // this is like av1_short_idct4x4 but has a special case around eob<=1 // which is significant (not just an optimization) for the lossless // case. - av1_highbd_inv_txfm_add_4x4(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_4x4_c(input, dest, stride, txfm_param); break; case TX_16X4: - highbd_inv_txfm_add_16x4(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_16x4(input, dest, stride, txfm_param); break; case TX_4X16: - highbd_inv_txfm_add_4x16(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_4x16(input, dest, stride, txfm_param); break; case TX_8X32: - highbd_inv_txfm_add_8x32(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_8x32(input, dest, stride, txfm_param); break; case TX_32X8: - highbd_inv_txfm_add_32x8(input, dest, stride, txfm_param); + av1_highbd_inv_txfm_add_32x8(input, dest, stride, txfm_param); break; default: assert(0 && "Invalid transform size"); break; } @@ -352,7 +291,8 @@ void av1_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, } } - highbd_inv_txfm_add(dqcoeff, CONVERT_TO_BYTEPTR(tmp), tmp_stride, txfm_param); + av1_highbd_inv_txfm_add(dqcoeff, CONVERT_TO_BYTEPTR(tmp), tmp_stride, + txfm_param); for (int r = 0; r < h; ++r) { for (int c = 0; c < w; ++c) { @@ -375,7 +315,7 @@ void av1_inverse_transform_block(const MACROBLOCKD *xd, assert(av1_ext_tx_used[txfm_param.tx_set_type][txfm_param.tx_type]); if (txfm_param.is_hbd) { - highbd_inv_txfm_add(dqcoeff, dst, stride, &txfm_param); + av1_highbd_inv_txfm_add(dqcoeff, dst, stride, &txfm_param); } else { av1_inv_txfm_add(dqcoeff, dst, stride, &txfm_param); } |