From 64653fa1330c258a52c52a4f32b940424439167c Mon Sep 17 00:00:00 2001 From: Linfeng Zhang Date: Tue, 12 Sep 2017 17:13:17 -0700 Subject: [PATCH] Change cospi_{1...31}_64 from tran_high_t to tran_coef_t The unnecessary upcast to (int) will be cleaned later. BUG=webm:1450 Change-Id: Ia234575206d5a74540526924b06ed3939322d063 --- vpx_dsp/arm/fdct16x16_neon.c | 4 +- vpx_dsp/arm/fdct32x32_neon.c | 47 ++- vpx_dsp/arm/highbd_idct16x16_add_neon.c | 8 +- vpx_dsp/arm/highbd_idct32x32_add_neon.c | 8 +- vpx_dsp/arm/highbd_idct4x4_add_neon.c | 8 +- vpx_dsp/arm/highbd_idct8x8_add_neon.c | 8 +- vpx_dsp/inv_txfm.c | 418 ++++++++++++++---------- vpx_dsp/txfm_common.h | 62 ++-- vpx_dsp/x86/highbd_idct4x4_add_sse2.c | 6 +- vpx_dsp/x86/highbd_inv_txfm_sse2.h | 6 +- 10 files changed, 325 insertions(+), 250 deletions(-) diff --git a/vpx_dsp/arm/fdct16x16_neon.c b/vpx_dsp/arm/fdct16x16_neon.c index fc250fd0c..6b2bebd09 100644 --- a/vpx_dsp/arm/fdct16x16_neon.c +++ b/vpx_dsp/arm/fdct16x16_neon.c @@ -169,8 +169,8 @@ static INLINE void butterfly_one_coeff(const int16x8_t a, const int16x8_t b, // fdct_round_shift(a * c0 +/- b * c1) static INLINE void butterfly_two_coeff(const int16x8_t a, const int16x8_t b, - const tran_high_t c0, - const tran_high_t c1, int16x8_t *add, + const tran_coef_t c0, + const tran_coef_t c1, int16x8_t *add, int16x8_t *sub) { const int32x4_t a0 = vmull_n_s16(vget_low_s16(a), c0); const int32x4_t a1 = vmull_n_s16(vget_high_s16(a), c0); diff --git a/vpx_dsp/arm/fdct32x32_neon.c b/vpx_dsp/arm/fdct32x32_neon.c index 31e5810a0..e9cd34904 100644 --- a/vpx_dsp/arm/fdct32x32_neon.c +++ b/vpx_dsp/arm/fdct32x32_neon.c @@ -214,8 +214,8 @@ static INLINE void butterfly_one_coeff(const int16x8_t a, const int16x8_t b, // fdct_round_shift(a * c0 +/- b * c1) static INLINE void butterfly_two_coeff(const int16x8_t a, const int16x8_t b, - const tran_high_t constant0, - const tran_high_t constant1, + const tran_coef_t constant0, + const tran_coef_t constant1, int16x8_t *add, int16x8_t *sub) { const int32x4_t a0 = vmull_n_s16(vget_low_s16(a), constant0); const int32x4_t a1 = vmull_n_s16(vget_high_s16(a), constant0); @@ -590,19 +590,14 @@ static INLINE void butterfly_one_coeff_s16_s32( // Like butterfly_one_coeff, but with s32. static INLINE void butterfly_one_coeff_s32( const int32x4_t a_lo, const int32x4_t a_hi, const int32x4_t b_lo, - const int32x4_t b_hi, const tran_high_t constant, int32x4_t *add_lo, + const int32x4_t b_hi, const int32_t constant, int32x4_t *add_lo, int32x4_t *add_hi, int32x4_t *sub_lo, int32x4_t *sub_hi) { - // TODO(johannkoenig): Strangely there is only a conversion warning on int64_t - // to int32_t (const tran_high_t (aka const long long)) but not for int64_t to - // int16_t. The constants fit in int16_t. Investigate using int16_t for the - // constants to avoid bouncing between types. - const int32_t constant_s32 = (int32_t)constant; - const int32x4_t a_lo_0 = vmulq_n_s32(a_lo, constant_s32); - const int32x4_t a_hi_0 = vmulq_n_s32(a_hi, constant_s32); - const int32x4_t sum0 = vmlaq_n_s32(a_lo_0, b_lo, constant_s32); - const int32x4_t sum1 = vmlaq_n_s32(a_hi_0, b_hi, constant_s32); - const int32x4_t diff0 = vmlsq_n_s32(a_lo_0, b_lo, constant_s32); - const int32x4_t diff1 = vmlsq_n_s32(a_hi_0, b_hi, constant_s32); + const int32x4_t a_lo_0 = vmulq_n_s32(a_lo, constant); + const int32x4_t a_hi_0 = vmulq_n_s32(a_hi, constant); + const int32x4_t sum0 = vmlaq_n_s32(a_lo_0, b_lo, constant); + const int32x4_t sum1 = vmlaq_n_s32(a_hi_0, b_hi, constant); + const int32x4_t diff0 = vmlsq_n_s32(a_lo_0, b_lo, constant); + const int32x4_t diff1 = vmlsq_n_s32(a_hi_0, b_hi, constant); *add_lo = vrshrq_n_s32(sum0, DCT_CONST_BITS); *add_hi = vrshrq_n_s32(sum1, DCT_CONST_BITS); *sub_lo = vrshrq_n_s32(diff0, DCT_CONST_BITS); @@ -621,19 +616,17 @@ static INLINE void butterfly_one_coeff_s32( // Like butterfly_two_coeff, but with s32. static INLINE void butterfly_two_coeff_s32( const int32x4_t a_lo, const int32x4_t a_hi, const int32x4_t b_lo, - const int32x4_t b_hi, const tran_high_t constant0, - const tran_high_t constant1, int32x4_t *add_lo, int32x4_t *add_hi, - int32x4_t *sub_lo, int32x4_t *sub_hi) { - const int32_t constant0_s32 = (int32_t)constant0; - const int32_t constant1_s32 = (int32_t)constant1; - const int32x4_t a0 = vmulq_n_s32(a_lo, constant0_s32); - const int32x4_t a1 = vmulq_n_s32(a_hi, constant0_s32); - const int32x4_t a2 = vmulq_n_s32(a_lo, constant1_s32); - const int32x4_t a3 = vmulq_n_s32(a_hi, constant1_s32); - const int32x4_t sum0 = vmlaq_n_s32(a2, b_lo, constant0_s32); - const int32x4_t sum1 = vmlaq_n_s32(a3, b_hi, constant0_s32); - const int32x4_t diff0 = vmlsq_n_s32(a0, b_lo, constant1_s32); - const int32x4_t diff1 = vmlsq_n_s32(a1, b_hi, constant1_s32); + const int32x4_t b_hi, const int32_t constant0, const int32_t constant1, + int32x4_t *add_lo, int32x4_t *add_hi, int32x4_t *sub_lo, + int32x4_t *sub_hi) { + const int32x4_t a0 = vmulq_n_s32(a_lo, constant0); + const int32x4_t a1 = vmulq_n_s32(a_hi, constant0); + const int32x4_t a2 = vmulq_n_s32(a_lo, constant1); + const int32x4_t a3 = vmulq_n_s32(a_hi, constant1); + const int32x4_t sum0 = vmlaq_n_s32(a2, b_lo, constant0); + const int32x4_t sum1 = vmlaq_n_s32(a3, b_hi, constant0); + const int32x4_t diff0 = vmlsq_n_s32(a0, b_lo, constant1); + const int32x4_t diff1 = vmlsq_n_s32(a1, b_hi, constant1); *add_lo = vrshrq_n_s32(sum0, DCT_CONST_BITS); *add_hi = vrshrq_n_s32(sum1, DCT_CONST_BITS); *sub_lo = vrshrq_n_s32(diff0, DCT_CONST_BITS); diff --git a/vpx_dsp/arm/highbd_idct16x16_add_neon.c b/vpx_dsp/arm/highbd_idct16x16_add_neon.c index 98e42cd25..5358839b5 100644 --- a/vpx_dsp/arm/highbd_idct16x16_add_neon.c +++ b/vpx_dsp/arm/highbd_idct16x16_add_neon.c @@ -1410,10 +1410,10 @@ static INLINE void highbd_idct16x16_1_add_neg_kernel(uint16_t **dest, void vpx_highbd_idct16x16_1_add_neon(const tran_low_t *input, uint16_t *dest, int stride, int bd) { - const tran_low_t out0 = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); - const tran_low_t out1 = - HIGHBD_WRAPLOW(dct_const_round_shift(out0 * cospi_16_64), bd); + const tran_low_t out0 = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); + const tran_low_t out1 = HIGHBD_WRAPLOW( + dct_const_round_shift(out0 * (tran_high_t)cospi_16_64), bd); const int16_t a1 = ROUND_POWER_OF_TWO(out1, 6); const int16x8_t dc = vdupq_n_s16(a1); int i; diff --git a/vpx_dsp/arm/highbd_idct32x32_add_neon.c b/vpx_dsp/arm/highbd_idct32x32_add_neon.c index 63eb49678..c1354c0c1 100644 --- a/vpx_dsp/arm/highbd_idct32x32_add_neon.c +++ b/vpx_dsp/arm/highbd_idct32x32_add_neon.c @@ -61,10 +61,10 @@ static INLINE void highbd_idct32x32_1_add_neg_kernel(uint16_t **dest, void vpx_highbd_idct32x32_1_add_neon(const tran_low_t *input, uint16_t *dest, int stride, int bd) { - const tran_low_t out0 = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); - const tran_low_t out1 = - HIGHBD_WRAPLOW(dct_const_round_shift(out0 * cospi_16_64), bd); + const tran_low_t out0 = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); + const tran_low_t out1 = HIGHBD_WRAPLOW( + dct_const_round_shift(out0 * (tran_high_t)cospi_16_64), bd); const int16_t a1 = ROUND_POWER_OF_TWO(out1, 6); const int16x8_t dc = vdupq_n_s16(a1); int i; diff --git a/vpx_dsp/arm/highbd_idct4x4_add_neon.c b/vpx_dsp/arm/highbd_idct4x4_add_neon.c index 20b09f683..1418a75a1 100644 --- a/vpx_dsp/arm/highbd_idct4x4_add_neon.c +++ b/vpx_dsp/arm/highbd_idct4x4_add_neon.c @@ -54,10 +54,10 @@ static INLINE void highbd_idct4x4_1_add_kernel2(uint16_t **dest, void vpx_highbd_idct4x4_1_add_neon(const tran_low_t *input, uint16_t *dest, int stride, int bd) { const int16x8_t max = vdupq_n_s16((1 << bd) - 1); - const tran_low_t out0 = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); - const tran_low_t out1 = - HIGHBD_WRAPLOW(dct_const_round_shift(out0 * cospi_16_64), bd); + const tran_low_t out0 = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); + const tran_low_t out1 = HIGHBD_WRAPLOW( + dct_const_round_shift(out0 * (tran_high_t)cospi_16_64), bd); const int16_t a1 = ROUND_POWER_OF_TWO(out1, 4); const int16x8_t dc = vdupq_n_s16(a1); diff --git a/vpx_dsp/arm/highbd_idct8x8_add_neon.c b/vpx_dsp/arm/highbd_idct8x8_add_neon.c index 6687e7649..dd90134a6 100644 --- a/vpx_dsp/arm/highbd_idct8x8_add_neon.c +++ b/vpx_dsp/arm/highbd_idct8x8_add_neon.c @@ -38,10 +38,10 @@ static INLINE void highbd_idct8x8_1_add_neg_kernel(uint16_t **dest, void vpx_highbd_idct8x8_1_add_neon(const tran_low_t *input, uint16_t *dest, int stride, int bd) { - const tran_low_t out0 = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); - const tran_low_t out1 = - HIGHBD_WRAPLOW(dct_const_round_shift(out0 * cospi_16_64), bd); + const tran_low_t out0 = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); + const tran_low_t out1 = HIGHBD_WRAPLOW( + dct_const_round_shift(out0 * (tran_high_t)cospi_16_64), bd); const int16_t a1 = ROUND_POWER_OF_TWO(out1, 5); const int16x8_t dc = vdupq_n_s16(a1); diff --git a/vpx_dsp/inv_txfm.c b/vpx_dsp/inv_txfm.c index e0c2a67df..2320261ad 100644 --- a/vpx_dsp/inv_txfm.c +++ b/vpx_dsp/inv_txfm.c @@ -1429,12 +1429,14 @@ void vpx_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd) { } // stage 1 - temp1 = (input[0] + input[2]) * cospi_16_64; - temp2 = (input[0] - input[2]) * cospi_16_64; + temp1 = (input[0] + input[2]) * (tran_high_t)cospi_16_64; + temp2 = (input[0] - input[2]) * (tran_high_t)cospi_16_64; step[0] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step[1] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64; - temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64; + temp1 = + input[1] * (tran_high_t)cospi_24_64 - input[3] * (tran_high_t)cospi_8_64; + temp2 = + input[1] * (tran_high_t)cospi_8_64 + input[3] * (tran_high_t)cospi_24_64; step[2] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step[3] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -1474,10 +1476,11 @@ void vpx_highbd_idct4x4_1_add_c(const tran_low_t *input, uint16_t *dest, int stride, int bd) { int i; tran_high_t a1; - tran_low_t out = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); + tran_low_t out = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); - out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd); + out = + HIGHBD_WRAPLOW(dct_const_round_shift(out * (tran_high_t)cospi_16_64), bd); a1 = ROUND_POWER_OF_TWO(out, 4); for (i = 0; i < 4; i++) { @@ -1515,14 +1518,14 @@ void vpx_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd) { } // stage 1 - s0 = cospi_2_64 * x0 + cospi_30_64 * x1; - s1 = cospi_30_64 * x0 - cospi_2_64 * x1; - s2 = cospi_10_64 * x2 + cospi_22_64 * x3; - s3 = cospi_22_64 * x2 - cospi_10_64 * x3; - s4 = cospi_18_64 * x4 + cospi_14_64 * x5; - s5 = cospi_14_64 * x4 - cospi_18_64 * x5; - s6 = cospi_26_64 * x6 + cospi_6_64 * x7; - s7 = cospi_6_64 * x6 - cospi_26_64 * x7; + s0 = (tran_high_t)cospi_2_64 * x0 + (tran_high_t)cospi_30_64 * x1; + s1 = (tran_high_t)cospi_30_64 * x0 - (tran_high_t)cospi_2_64 * x1; + s2 = (tran_high_t)cospi_10_64 * x2 + (tran_high_t)cospi_22_64 * x3; + s3 = (tran_high_t)cospi_22_64 * x2 - (tran_high_t)cospi_10_64 * x3; + s4 = (tran_high_t)cospi_18_64 * x4 + (tran_high_t)cospi_14_64 * x5; + s5 = (tran_high_t)cospi_14_64 * x4 - (tran_high_t)cospi_18_64 * x5; + s6 = (tran_high_t)cospi_26_64 * x6 + (tran_high_t)cospi_6_64 * x7; + s7 = (tran_high_t)cospi_6_64 * x6 - (tran_high_t)cospi_26_64 * x7; x0 = HIGHBD_WRAPLOW(dct_const_round_shift(s0 + s4), bd); x1 = HIGHBD_WRAPLOW(dct_const_round_shift(s1 + s5), bd); @@ -1538,10 +1541,10 @@ void vpx_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd) { s1 = x1; s2 = x2; s3 = x3; - s4 = cospi_8_64 * x4 + cospi_24_64 * x5; - s5 = cospi_24_64 * x4 - cospi_8_64 * x5; - s6 = -cospi_24_64 * x6 + cospi_8_64 * x7; - s7 = cospi_8_64 * x6 + cospi_24_64 * x7; + s4 = (tran_high_t)cospi_8_64 * x4 + (tran_high_t)cospi_24_64 * x5; + s5 = (tran_high_t)cospi_24_64 * x4 - (tran_high_t)cospi_8_64 * x5; + s6 = (tran_high_t)(-cospi_24_64) * x6 + (tran_high_t)cospi_8_64 * x7; + s7 = (tran_high_t)cospi_8_64 * x6 + (tran_high_t)cospi_24_64 * x7; x0 = HIGHBD_WRAPLOW(s0 + s2, bd); x1 = HIGHBD_WRAPLOW(s1 + s3, bd); @@ -1553,10 +1556,10 @@ void vpx_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd) { x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 - s7), bd); // stage 3 - s2 = cospi_16_64 * (x2 + x3); - s3 = cospi_16_64 * (x2 - x3); - s6 = cospi_16_64 * (x6 + x7); - s7 = cospi_16_64 * (x6 - x7); + s2 = (tran_high_t)cospi_16_64 * (x2 + x3); + s3 = (tran_high_t)cospi_16_64 * (x2 - x3); + s6 = (tran_high_t)cospi_16_64 * (x6 + x7); + s7 = (tran_high_t)cospi_16_64 * (x6 - x7); x2 = HIGHBD_WRAPLOW(dct_const_round_shift(s2), bd); x3 = HIGHBD_WRAPLOW(dct_const_round_shift(s3), bd); @@ -1590,12 +1593,16 @@ void vpx_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd) { step1[2] = input[4]; step1[1] = input[2]; step1[3] = input[6]; - temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64; - temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64; + temp1 = + input[1] * (tran_high_t)cospi_28_64 - input[7] * (tran_high_t)cospi_4_64; + temp2 = + input[1] * (tran_high_t)cospi_4_64 + input[7] * (tran_high_t)cospi_28_64; step1[4] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[7] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64; - temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64; + temp1 = + input[5] * (tran_high_t)cospi_12_64 - input[3] * (tran_high_t)cospi_20_64; + temp2 = + input[5] * (tran_high_t)cospi_20_64 + input[3] * (tran_high_t)cospi_12_64; step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -1610,8 +1617,8 @@ void vpx_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd) { // stage 3 - odd half step1[4] = step2[4]; - temp1 = (step2[6] - step2[5]) * cospi_16_64; - temp2 = (step2[5] + step2[6]) * cospi_16_64; + temp1 = (step2[6] - step2[5]) * (tran_high_t)cospi_16_64; + temp2 = (step2[5] + step2[6]) * (tran_high_t)cospi_16_64; step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[7] = step2[7]; @@ -1682,10 +1689,11 @@ void vpx_highbd_idct8x8_1_add_c(const tran_low_t *input, uint16_t *dest, int stride, int bd) { int i, j; tran_high_t a1; - tran_low_t out = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); + tran_low_t out = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); - out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd); + out = + HIGHBD_WRAPLOW(dct_const_round_shift(out * (tran_high_t)cospi_16_64), bd); a1 = ROUND_POWER_OF_TWO(out, 5); for (j = 0; j < 8; ++j) { for (i = 0; i < 8; ++i) dest[i] = highbd_clip_pixel_add(dest[i], a1, bd); @@ -1729,22 +1737,22 @@ void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd) { } // stage 1 - s0 = x0 * cospi_1_64 + x1 * cospi_31_64; - s1 = x0 * cospi_31_64 - x1 * cospi_1_64; - s2 = x2 * cospi_5_64 + x3 * cospi_27_64; - s3 = x2 * cospi_27_64 - x3 * cospi_5_64; - s4 = x4 * cospi_9_64 + x5 * cospi_23_64; - s5 = x4 * cospi_23_64 - x5 * cospi_9_64; - s6 = x6 * cospi_13_64 + x7 * cospi_19_64; - s7 = x6 * cospi_19_64 - x7 * cospi_13_64; - s8 = x8 * cospi_17_64 + x9 * cospi_15_64; - s9 = x8 * cospi_15_64 - x9 * cospi_17_64; - s10 = x10 * cospi_21_64 + x11 * cospi_11_64; - s11 = x10 * cospi_11_64 - x11 * cospi_21_64; - s12 = x12 * cospi_25_64 + x13 * cospi_7_64; - s13 = x12 * cospi_7_64 - x13 * cospi_25_64; - s14 = x14 * cospi_29_64 + x15 * cospi_3_64; - s15 = x14 * cospi_3_64 - x15 * cospi_29_64; + s0 = x0 * (tran_high_t)cospi_1_64 + x1 * (tran_high_t)cospi_31_64; + s1 = x0 * (tran_high_t)cospi_31_64 - x1 * (tran_high_t)cospi_1_64; + s2 = x2 * (tran_high_t)cospi_5_64 + x3 * (tran_high_t)cospi_27_64; + s3 = x2 * (tran_high_t)cospi_27_64 - x3 * (tran_high_t)cospi_5_64; + s4 = x4 * (tran_high_t)cospi_9_64 + x5 * (tran_high_t)cospi_23_64; + s5 = x4 * (tran_high_t)cospi_23_64 - x5 * (tran_high_t)cospi_9_64; + s6 = x6 * (tran_high_t)cospi_13_64 + x7 * (tran_high_t)cospi_19_64; + s7 = x6 * (tran_high_t)cospi_19_64 - x7 * (tran_high_t)cospi_13_64; + s8 = x8 * (tran_high_t)cospi_17_64 + x9 * (tran_high_t)cospi_15_64; + s9 = x8 * (tran_high_t)cospi_15_64 - x9 * (tran_high_t)cospi_17_64; + s10 = x10 * (tran_high_t)cospi_21_64 + x11 * (tran_high_t)cospi_11_64; + s11 = x10 * (tran_high_t)cospi_11_64 - x11 * (tran_high_t)cospi_21_64; + s12 = x12 * (tran_high_t)cospi_25_64 + x13 * (tran_high_t)cospi_7_64; + s13 = x12 * (tran_high_t)cospi_7_64 - x13 * (tran_high_t)cospi_25_64; + s14 = x14 * (tran_high_t)cospi_29_64 + x15 * (tran_high_t)cospi_3_64; + s15 = x14 * (tran_high_t)cospi_3_64 - x15 * (tran_high_t)cospi_29_64; x0 = HIGHBD_WRAPLOW(dct_const_round_shift(s0 + s8), bd); x1 = HIGHBD_WRAPLOW(dct_const_round_shift(s1 + s9), bd); @@ -1772,14 +1780,14 @@ void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd) { s5 = x5; s6 = x6; s7 = x7; - s8 = x8 * cospi_4_64 + x9 * cospi_28_64; - s9 = x8 * cospi_28_64 - x9 * cospi_4_64; - s10 = x10 * cospi_20_64 + x11 * cospi_12_64; - s11 = x10 * cospi_12_64 - x11 * cospi_20_64; - s12 = -x12 * cospi_28_64 + x13 * cospi_4_64; - s13 = x12 * cospi_4_64 + x13 * cospi_28_64; - s14 = -x14 * cospi_12_64 + x15 * cospi_20_64; - s15 = x14 * cospi_20_64 + x15 * cospi_12_64; + s8 = x8 * (tran_high_t)cospi_4_64 + x9 * (tran_high_t)cospi_28_64; + s9 = x8 * (tran_high_t)cospi_28_64 - x9 * (tran_high_t)cospi_4_64; + s10 = x10 * (tran_high_t)cospi_20_64 + x11 * (tran_high_t)cospi_12_64; + s11 = x10 * (tran_high_t)cospi_12_64 - x11 * (tran_high_t)cospi_20_64; + s12 = -x12 * (tran_high_t)cospi_28_64 + x13 * (tran_high_t)cospi_4_64; + s13 = x12 * (tran_high_t)cospi_4_64 + x13 * (tran_high_t)cospi_28_64; + s14 = -x14 * (tran_high_t)cospi_12_64 + x15 * (tran_high_t)cospi_20_64; + s15 = x14 * (tran_high_t)cospi_20_64 + x15 * (tran_high_t)cospi_12_64; x0 = HIGHBD_WRAPLOW(s0 + s4, bd); x1 = HIGHBD_WRAPLOW(s1 + s5, bd); @@ -1803,18 +1811,18 @@ void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd) { s1 = x1; s2 = x2; s3 = x3; - s4 = x4 * cospi_8_64 + x5 * cospi_24_64; - s5 = x4 * cospi_24_64 - x5 * cospi_8_64; - s6 = -x6 * cospi_24_64 + x7 * cospi_8_64; - s7 = x6 * cospi_8_64 + x7 * cospi_24_64; + s4 = x4 * (tran_high_t)cospi_8_64 + x5 * (tran_high_t)cospi_24_64; + s5 = x4 * (tran_high_t)cospi_24_64 - x5 * (tran_high_t)cospi_8_64; + s6 = -x6 * (tran_high_t)cospi_24_64 + x7 * (tran_high_t)cospi_8_64; + s7 = x6 * (tran_high_t)cospi_8_64 + x7 * (tran_high_t)cospi_24_64; s8 = x8; s9 = x9; s10 = x10; s11 = x11; - s12 = x12 * cospi_8_64 + x13 * cospi_24_64; - s13 = x12 * cospi_24_64 - x13 * cospi_8_64; - s14 = -x14 * cospi_24_64 + x15 * cospi_8_64; - s15 = x14 * cospi_8_64 + x15 * cospi_24_64; + s12 = x12 * (tran_high_t)cospi_8_64 + x13 * (tran_high_t)cospi_24_64; + s13 = x12 * (tran_high_t)cospi_24_64 - x13 * (tran_high_t)cospi_8_64; + s14 = -x14 * (tran_high_t)cospi_24_64 + x15 * (tran_high_t)cospi_8_64; + s15 = x14 * (tran_high_t)cospi_8_64 + x15 * (tran_high_t)cospi_24_64; x0 = HIGHBD_WRAPLOW(s0 + s2, bd); x1 = HIGHBD_WRAPLOW(s1 + s3, bd); @@ -1834,14 +1842,14 @@ void vpx_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd) { x15 = HIGHBD_WRAPLOW(dct_const_round_shift(s13 - s15), bd); // stage 4 - s2 = (-cospi_16_64) * (x2 + x3); - s3 = cospi_16_64 * (x2 - x3); - s6 = cospi_16_64 * (x6 + x7); - s7 = cospi_16_64 * (-x6 + x7); - s10 = cospi_16_64 * (x10 + x11); - s11 = cospi_16_64 * (-x10 + x11); - s14 = (-cospi_16_64) * (x14 + x15); - s15 = cospi_16_64 * (x14 - x15); + s2 = (tran_high_t)(-cospi_16_64) * (x2 + x3); + s3 = (tran_high_t)cospi_16_64 * (x2 - x3); + s6 = (tran_high_t)cospi_16_64 * (x6 + x7); + s7 = (tran_high_t)cospi_16_64 * (-x6 + x7); + s10 = (tran_high_t)cospi_16_64 * (x10 + x11); + s11 = (tran_high_t)cospi_16_64 * (-x10 + x11); + s14 = (tran_high_t)(-cospi_16_64) * (x14 + x15); + s15 = (tran_high_t)cospi_16_64 * (x14 - x15); x2 = HIGHBD_WRAPLOW(dct_const_round_shift(s2), bd); x3 = HIGHBD_WRAPLOW(dct_const_round_shift(s3), bd); @@ -1911,23 +1919,31 @@ void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { step2[6] = step1[6]; step2[7] = step1[7]; - temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64; - temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64; + temp1 = + step1[8] * (tran_high_t)cospi_30_64 - step1[15] * (tran_high_t)cospi_2_64; + temp2 = + step1[8] * (tran_high_t)cospi_2_64 + step1[15] * (tran_high_t)cospi_30_64; step2[8] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[15] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64; - temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64; + temp1 = step1[9] * (tran_high_t)cospi_14_64 - + step1[14] * (tran_high_t)cospi_18_64; + temp2 = step1[9] * (tran_high_t)cospi_18_64 + + step1[14] * (tran_high_t)cospi_14_64; step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64; - temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64; + temp1 = step1[10] * (tran_high_t)cospi_22_64 - + step1[13] * (tran_high_t)cospi_10_64; + temp2 = step1[10] * (tran_high_t)cospi_10_64 + + step1[13] * (tran_high_t)cospi_22_64; step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64; - temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64; + temp1 = step1[11] * (tran_high_t)cospi_6_64 - + step1[12] * (tran_high_t)cospi_26_64; + temp2 = step1[11] * (tran_high_t)cospi_26_64 + + step1[12] * (tran_high_t)cospi_6_64; step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -1937,12 +1953,16 @@ void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { step1[2] = step2[2]; step1[3] = step2[3]; - temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64; - temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64; + temp1 = + step2[4] * (tran_high_t)cospi_28_64 - step2[7] * (tran_high_t)cospi_4_64; + temp2 = + step2[4] * (tran_high_t)cospi_4_64 + step2[7] * (tran_high_t)cospi_28_64; step1[4] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[7] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64; - temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64; + temp1 = + step2[5] * (tran_high_t)cospi_12_64 - step2[6] * (tran_high_t)cospi_20_64; + temp2 = + step2[5] * (tran_high_t)cospi_20_64 + step2[6] * (tran_high_t)cospi_12_64; step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -1956,12 +1976,14 @@ void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { step1[15] = HIGHBD_WRAPLOW(step2[14] + step2[15], bd); // stage 4 - temp1 = (step1[0] + step1[1]) * cospi_16_64; - temp2 = (step1[0] - step1[1]) * cospi_16_64; + temp1 = (step1[0] + step1[1]) * (tran_high_t)cospi_16_64; + temp2 = (step1[0] - step1[1]) * (tran_high_t)cospi_16_64; step2[0] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[1] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64; - temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64; + temp1 = + step1[2] * (tran_high_t)cospi_24_64 - step1[3] * (tran_high_t)cospi_8_64; + temp2 = + step1[2] * (tran_high_t)cospi_8_64 + step1[3] * (tran_high_t)cospi_24_64; step2[2] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[3] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd); @@ -1971,12 +1993,16 @@ void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { step2[8] = step1[8]; step2[15] = step1[15]; - temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64; - temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64; + temp1 = -step1[9] * (tran_high_t)cospi_8_64 + + step1[14] * (tran_high_t)cospi_24_64; + temp2 = + step1[9] * (tran_high_t)cospi_24_64 + step1[14] * (tran_high_t)cospi_8_64; step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64; - temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64; + temp1 = -step1[10] * (tran_high_t)cospi_24_64 - + step1[13] * (tran_high_t)cospi_8_64; + temp2 = -step1[10] * (tran_high_t)cospi_8_64 + + step1[13] * (tran_high_t)cospi_24_64; step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step2[11] = step1[11]; @@ -1988,8 +2014,8 @@ void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { step1[2] = HIGHBD_WRAPLOW(step2[1] - step2[2], bd); step1[3] = HIGHBD_WRAPLOW(step2[0] - step2[3], bd); step1[4] = step2[4]; - temp1 = (step2[6] - step2[5]) * cospi_16_64; - temp2 = (step2[5] + step2[6]) * cospi_16_64; + temp1 = (step2[6] - step2[5]) * (tran_high_t)cospi_16_64; + temp2 = (step2[5] + step2[6]) * (tran_high_t)cospi_16_64; step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[7] = step2[7]; @@ -2014,12 +2040,12 @@ void vpx_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) { step2[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd); step2[8] = step1[8]; step2[9] = step1[9]; - temp1 = (-step1[10] + step1[13]) * cospi_16_64; - temp2 = (step1[10] + step1[13]) * cospi_16_64; + temp1 = (-step1[10] + step1[13]) * (tran_high_t)cospi_16_64; + temp2 = (step1[10] + step1[13]) * (tran_high_t)cospi_16_64; step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = (-step1[11] + step1[12]) * cospi_16_64; - temp2 = (step1[11] + step1[12]) * cospi_16_64; + temp1 = (-step1[11] + step1[12]) * (tran_high_t)cospi_16_64; + temp2 = (step1[11] + step1[12]) * (tran_high_t)cospi_16_64; step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step2[14] = step1[14]; @@ -2127,10 +2153,11 @@ void vpx_highbd_idct16x16_1_add_c(const tran_low_t *input, uint16_t *dest, int stride, int bd) { int i, j; tran_high_t a1; - tran_low_t out = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); + tran_low_t out = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); - out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd); + out = + HIGHBD_WRAPLOW(dct_const_round_shift(out * (tran_high_t)cospi_16_64), bd); a1 = ROUND_POWER_OF_TWO(out, 6); for (j = 0; j < 16; ++j) { for (i = 0; i < 16; ++i) dest[i] = highbd_clip_pixel_add(dest[i], a1, bd); @@ -2170,43 +2197,59 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[14] = input[14]; step1[15] = input[30]; - temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64; - temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64; + temp1 = + input[1] * (tran_high_t)cospi_31_64 - input[31] * (tran_high_t)cospi_1_64; + temp2 = + input[1] * (tran_high_t)cospi_1_64 + input[31] * (tran_high_t)cospi_31_64; step1[16] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[31] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64; - temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64; + temp1 = input[17] * (tran_high_t)cospi_15_64 - + input[15] * (tran_high_t)cospi_17_64; + temp2 = input[17] * (tran_high_t)cospi_17_64 + + input[15] * (tran_high_t)cospi_15_64; step1[17] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[30] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64; - temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64; + temp1 = + input[9] * (tran_high_t)cospi_23_64 - input[23] * (tran_high_t)cospi_9_64; + temp2 = + input[9] * (tran_high_t)cospi_9_64 + input[23] * (tran_high_t)cospi_23_64; step1[18] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[29] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64; - temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64; + temp1 = + input[25] * (tran_high_t)cospi_7_64 - input[7] * (tran_high_t)cospi_25_64; + temp2 = + input[25] * (tran_high_t)cospi_25_64 + input[7] * (tran_high_t)cospi_7_64; step1[19] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[28] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64; - temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64; + temp1 = + input[5] * (tran_high_t)cospi_27_64 - input[27] * (tran_high_t)cospi_5_64; + temp2 = + input[5] * (tran_high_t)cospi_5_64 + input[27] * (tran_high_t)cospi_27_64; step1[20] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[27] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64; - temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64; + temp1 = input[21] * (tran_high_t)cospi_11_64 - + input[11] * (tran_high_t)cospi_21_64; + temp2 = input[21] * (tran_high_t)cospi_21_64 + + input[11] * (tran_high_t)cospi_11_64; step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64; - temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64; + temp1 = input[13] * (tran_high_t)cospi_19_64 - + input[19] * (tran_high_t)cospi_13_64; + temp2 = input[13] * (tran_high_t)cospi_13_64 + + input[19] * (tran_high_t)cospi_19_64; step1[22] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[25] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64; - temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64; + temp1 = + input[29] * (tran_high_t)cospi_3_64 - input[3] * (tran_high_t)cospi_29_64; + temp2 = + input[29] * (tran_high_t)cospi_29_64 + input[3] * (tran_high_t)cospi_3_64; step1[23] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[24] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -2220,23 +2263,31 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step2[6] = step1[6]; step2[7] = step1[7]; - temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64; - temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64; + temp1 = + step1[8] * (tran_high_t)cospi_30_64 - step1[15] * (tran_high_t)cospi_2_64; + temp2 = + step1[8] * (tran_high_t)cospi_2_64 + step1[15] * (tran_high_t)cospi_30_64; step2[8] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[15] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64; - temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64; + temp1 = step1[9] * (tran_high_t)cospi_14_64 - + step1[14] * (tran_high_t)cospi_18_64; + temp2 = step1[9] * (tran_high_t)cospi_18_64 + + step1[14] * (tran_high_t)cospi_14_64; step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64; - temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64; + temp1 = step1[10] * (tran_high_t)cospi_22_64 - + step1[13] * (tran_high_t)cospi_10_64; + temp2 = step1[10] * (tran_high_t)cospi_10_64 + + step1[13] * (tran_high_t)cospi_22_64; step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64; - temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64; + temp1 = step1[11] * (tran_high_t)cospi_6_64 - + step1[12] * (tran_high_t)cospi_26_64; + temp2 = step1[11] * (tran_high_t)cospi_26_64 + + step1[12] * (tran_high_t)cospi_6_64; step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -2263,12 +2314,16 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[2] = step2[2]; step1[3] = step2[3]; - temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64; - temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64; + temp1 = + step2[4] * (tran_high_t)cospi_28_64 - step2[7] * (tran_high_t)cospi_4_64; + temp2 = + step2[4] * (tran_high_t)cospi_4_64 + step2[7] * (tran_high_t)cospi_28_64; step1[4] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[7] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64; - temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64; + temp1 = + step2[5] * (tran_high_t)cospi_12_64 - step2[6] * (tran_high_t)cospi_20_64; + temp2 = + step2[5] * (tran_high_t)cospi_20_64 + step2[6] * (tran_high_t)cospi_12_64; step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); @@ -2283,22 +2338,30 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[16] = step2[16]; step1[31] = step2[31]; - temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64; - temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64; + temp1 = -step2[17] * (tran_high_t)cospi_4_64 + + step2[30] * (tran_high_t)cospi_28_64; + temp2 = step2[17] * (tran_high_t)cospi_28_64 + + step2[30] * (tran_high_t)cospi_4_64; step1[17] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[30] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64; - temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64; + temp1 = -step2[18] * (tran_high_t)cospi_28_64 - + step2[29] * (tran_high_t)cospi_4_64; + temp2 = -step2[18] * (tran_high_t)cospi_4_64 + + step2[29] * (tran_high_t)cospi_28_64; step1[18] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[29] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[19] = step2[19]; step1[20] = step2[20]; - temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64; - temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64; + temp1 = -step2[21] * (tran_high_t)cospi_20_64 + + step2[26] * (tran_high_t)cospi_12_64; + temp2 = step2[21] * (tran_high_t)cospi_12_64 + + step2[26] * (tran_high_t)cospi_20_64; step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64; - temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64; + temp1 = -step2[22] * (tran_high_t)cospi_12_64 - + step2[25] * (tran_high_t)cospi_20_64; + temp2 = -step2[22] * (tran_high_t)cospi_20_64 + + step2[25] * (tran_high_t)cospi_12_64; step1[22] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[25] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[23] = step2[23]; @@ -2307,12 +2370,14 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[28] = step2[28]; // stage 4 - temp1 = (step1[0] + step1[1]) * cospi_16_64; - temp2 = (step1[0] - step1[1]) * cospi_16_64; + temp1 = (step1[0] + step1[1]) * (tran_high_t)cospi_16_64; + temp2 = (step1[0] - step1[1]) * (tran_high_t)cospi_16_64; step2[0] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[1] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64; - temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64; + temp1 = + step1[2] * (tran_high_t)cospi_24_64 - step1[3] * (tran_high_t)cospi_8_64; + temp2 = + step1[2] * (tran_high_t)cospi_8_64 + step1[3] * (tran_high_t)cospi_24_64; step2[2] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[3] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd); @@ -2322,12 +2387,16 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step2[8] = step1[8]; step2[15] = step1[15]; - temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64; - temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64; + temp1 = -step1[9] * (tran_high_t)cospi_8_64 + + step1[14] * (tran_high_t)cospi_24_64; + temp2 = + step1[9] * (tran_high_t)cospi_24_64 + step1[14] * (tran_high_t)cospi_8_64; step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64; - temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64; + temp1 = -step1[10] * (tran_high_t)cospi_24_64 - + step1[13] * (tran_high_t)cospi_8_64; + temp2 = -step1[10] * (tran_high_t)cospi_8_64 + + step1[13] * (tran_high_t)cospi_24_64; step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step2[11] = step1[11]; @@ -2357,8 +2426,8 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[2] = HIGHBD_WRAPLOW(step2[1] - step2[2], bd); step1[3] = HIGHBD_WRAPLOW(step2[0] - step2[3], bd); step1[4] = step2[4]; - temp1 = (step2[6] - step2[5]) * cospi_16_64; - temp2 = (step2[5] + step2[6]) * cospi_16_64; + temp1 = (step2[6] - step2[5]) * (tran_high_t)cospi_16_64; + temp2 = (step2[5] + step2[6]) * (tran_high_t)cospi_16_64; step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[7] = step2[7]; @@ -2374,20 +2443,28 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[16] = step2[16]; step1[17] = step2[17]; - temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64; - temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64; + temp1 = -step2[18] * (tran_high_t)cospi_8_64 + + step2[29] * (tran_high_t)cospi_24_64; + temp2 = step2[18] * (tran_high_t)cospi_24_64 + + step2[29] * (tran_high_t)cospi_8_64; step1[18] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[29] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64; - temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64; + temp1 = -step2[19] * (tran_high_t)cospi_8_64 + + step2[28] * (tran_high_t)cospi_24_64; + temp2 = step2[19] * (tran_high_t)cospi_24_64 + + step2[28] * (tran_high_t)cospi_8_64; step1[19] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[28] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64; - temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64; + temp1 = -step2[20] * (tran_high_t)cospi_24_64 - + step2[27] * (tran_high_t)cospi_8_64; + temp2 = -step2[20] * (tran_high_t)cospi_8_64 + + step2[27] * (tran_high_t)cospi_24_64; step1[20] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[27] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64; - temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64; + temp1 = -step2[21] * (tran_high_t)cospi_24_64 - + step2[26] * (tran_high_t)cospi_8_64; + temp2 = -step2[21] * (tran_high_t)cospi_8_64 + + step2[26] * (tran_high_t)cospi_24_64; step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[22] = step2[22]; @@ -2408,12 +2485,12 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step2[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd); step2[8] = step1[8]; step2[9] = step1[9]; - temp1 = (-step1[10] + step1[13]) * cospi_16_64; - temp2 = (step1[10] + step1[13]) * cospi_16_64; + temp1 = (-step1[10] + step1[13]) * (tran_high_t)cospi_16_64; + temp2 = (step1[10] + step1[13]) * (tran_high_t)cospi_16_64; step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = (-step1[11] + step1[12]) * cospi_16_64; - temp2 = (step1[11] + step1[12]) * cospi_16_64; + temp1 = (-step1[11] + step1[12]) * (tran_high_t)cospi_16_64; + temp2 = (step1[11] + step1[12]) * (tran_high_t)cospi_16_64; step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step2[14] = step1[14]; @@ -2459,20 +2536,20 @@ static void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, step1[17] = step2[17]; step1[18] = step2[18]; step1[19] = step2[19]; - temp1 = (-step2[20] + step2[27]) * cospi_16_64; - temp2 = (step2[20] + step2[27]) * cospi_16_64; + temp1 = (-step2[20] + step2[27]) * (tran_high_t)cospi_16_64; + temp2 = (step2[20] + step2[27]) * (tran_high_t)cospi_16_64; step1[20] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[27] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = (-step2[21] + step2[26]) * cospi_16_64; - temp2 = (step2[21] + step2[26]) * cospi_16_64; + temp1 = (-step2[21] + step2[26]) * (tran_high_t)cospi_16_64; + temp2 = (step2[21] + step2[26]) * (tran_high_t)cospi_16_64; step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = (-step2[22] + step2[25]) * cospi_16_64; - temp2 = (step2[22] + step2[25]) * cospi_16_64; + temp1 = (-step2[22] + step2[25]) * (tran_high_t)cospi_16_64; + temp2 = (step2[22] + step2[25]) * (tran_high_t)cospi_16_64; step1[22] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[25] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); - temp1 = (-step2[23] + step2[24]) * cospi_16_64; - temp2 = (step2[23] + step2[24]) * cospi_16_64; + temp1 = (-step2[23] + step2[24]) * (tran_high_t)cospi_16_64; + temp2 = (step2[23] + step2[24]) * (tran_high_t)cospi_16_64; step1[23] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd); step1[24] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd); step1[28] = step2[28]; @@ -2604,10 +2681,11 @@ void vpx_highbd_idct32x32_1_add_c(const tran_low_t *input, uint16_t *dest, int stride, int bd) { int i, j; int a1; - tran_low_t out = - HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); + tran_low_t out = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); - out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd); + out = + HIGHBD_WRAPLOW(dct_const_round_shift(out * (tran_high_t)cospi_16_64), bd); a1 = ROUND_POWER_OF_TWO(out, 6); for (j = 0; j < 32; ++j) { diff --git a/vpx_dsp/txfm_common.h b/vpx_dsp/txfm_common.h index 95e4e4bdc..d01d7085a 100644 --- a/vpx_dsp/txfm_common.h +++ b/vpx_dsp/txfm_common.h @@ -25,37 +25,37 @@ // printf("static const int cospi_%d_64 = %.0f;\n", i, // round(16384 * cos(i*M_PI/64))); // Note: sin(k*Pi/64) = cos((32-k)*Pi/64) -static const tran_high_t cospi_1_64 = 16364; -static const tran_high_t cospi_2_64 = 16305; -static const tran_high_t cospi_3_64 = 16207; -static const tran_high_t cospi_4_64 = 16069; -static const tran_high_t cospi_5_64 = 15893; -static const tran_high_t cospi_6_64 = 15679; -static const tran_high_t cospi_7_64 = 15426; -static const tran_high_t cospi_8_64 = 15137; -static const tran_high_t cospi_9_64 = 14811; -static const tran_high_t cospi_10_64 = 14449; -static const tran_high_t cospi_11_64 = 14053; -static const tran_high_t cospi_12_64 = 13623; -static const tran_high_t cospi_13_64 = 13160; -static const tran_high_t cospi_14_64 = 12665; -static const tran_high_t cospi_15_64 = 12140; -static const tran_high_t cospi_16_64 = 11585; -static const tran_high_t cospi_17_64 = 11003; -static const tran_high_t cospi_18_64 = 10394; -static const tran_high_t cospi_19_64 = 9760; -static const tran_high_t cospi_20_64 = 9102; -static const tran_high_t cospi_21_64 = 8423; -static const tran_high_t cospi_22_64 = 7723; -static const tran_high_t cospi_23_64 = 7005; -static const tran_high_t cospi_24_64 = 6270; -static const tran_high_t cospi_25_64 = 5520; -static const tran_high_t cospi_26_64 = 4756; -static const tran_high_t cospi_27_64 = 3981; -static const tran_high_t cospi_28_64 = 3196; -static const tran_high_t cospi_29_64 = 2404; -static const tran_high_t cospi_30_64 = 1606; -static const tran_high_t cospi_31_64 = 804; +static const tran_coef_t cospi_1_64 = 16364; +static const tran_coef_t cospi_2_64 = 16305; +static const tran_coef_t cospi_3_64 = 16207; +static const tran_coef_t cospi_4_64 = 16069; +static const tran_coef_t cospi_5_64 = 15893; +static const tran_coef_t cospi_6_64 = 15679; +static const tran_coef_t cospi_7_64 = 15426; +static const tran_coef_t cospi_8_64 = 15137; +static const tran_coef_t cospi_9_64 = 14811; +static const tran_coef_t cospi_10_64 = 14449; +static const tran_coef_t cospi_11_64 = 14053; +static const tran_coef_t cospi_12_64 = 13623; +static const tran_coef_t cospi_13_64 = 13160; +static const tran_coef_t cospi_14_64 = 12665; +static const tran_coef_t cospi_15_64 = 12140; +static const tran_coef_t cospi_16_64 = 11585; +static const tran_coef_t cospi_17_64 = 11003; +static const tran_coef_t cospi_18_64 = 10394; +static const tran_coef_t cospi_19_64 = 9760; +static const tran_coef_t cospi_20_64 = 9102; +static const tran_coef_t cospi_21_64 = 8423; +static const tran_coef_t cospi_22_64 = 7723; +static const tran_coef_t cospi_23_64 = 7005; +static const tran_coef_t cospi_24_64 = 6270; +static const tran_coef_t cospi_25_64 = 5520; +static const tran_coef_t cospi_26_64 = 4756; +static const tran_coef_t cospi_27_64 = 3981; +static const tran_coef_t cospi_28_64 = 3196; +static const tran_coef_t cospi_29_64 = 2404; +static const tran_coef_t cospi_30_64 = 1606; +static const tran_coef_t cospi_31_64 = 804; // 16384 * sqrt(2) * sin(kPi/9) * 2 / 3 static const tran_coef_t sinpi_1_9 = 5283; diff --git a/vpx_dsp/x86/highbd_idct4x4_add_sse2.c b/vpx_dsp/x86/highbd_idct4x4_add_sse2.c index 417c85a04..d54a036bd 100644 --- a/vpx_dsp/x86/highbd_idct4x4_add_sse2.c +++ b/vpx_dsp/x86/highbd_idct4x4_add_sse2.c @@ -147,8 +147,10 @@ void vpx_highbd_idct4x4_1_add_sse2(const tran_low_t *input, uint16_t *dest, tran_low_t out; __m128i dc, d; - out = HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); - out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd); + out = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); + out = + HIGHBD_WRAPLOW(dct_const_round_shift(out * (tran_high_t)cospi_16_64), bd); a1 = ROUND_POWER_OF_TWO(out, 4); dc = _mm_set1_epi16(a1); diff --git a/vpx_dsp/x86/highbd_inv_txfm_sse2.h b/vpx_dsp/x86/highbd_inv_txfm_sse2.h index c58e49c02..cbe49d17a 100644 --- a/vpx_dsp/x86/highbd_inv_txfm_sse2.h +++ b/vpx_dsp/x86/highbd_inv_txfm_sse2.h @@ -265,8 +265,10 @@ static INLINE void highbd_idct_1_add_kernel(const tran_low_t *input, tran_low_t out; __m128i dc, d; - out = HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd); - out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd); + out = HIGHBD_WRAPLOW( + dct_const_round_shift(input[0] * (tran_high_t)cospi_16_64), bd); + out = + HIGHBD_WRAPLOW(dct_const_round_shift(out * (tran_high_t)cospi_16_64), bd); a1 = ROUND_POWER_OF_TWO(out, (size == 8) ? 5 : 6); dc = _mm_set1_epi16(a1); -- 2.40.0