From 98e88e6ad83a4aacbe954d99b1576ecb95bcd6df Mon Sep 17 00:00:00 2001 From: hkuang Date: Mon, 1 Jun 2015 17:10:33 -0700 Subject: [PATCH] Optimize the idct assembly code. Change-Id: Ia0ff859ff1c813dbe100e2f27b1ef78167483f4e --- vp9/common/x86/vp9_idct_intrin_sse2.c | 68 +++++++-------------------- 1 file changed, 17 insertions(+), 51 deletions(-) diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c index 52d21be4a..ce010df3b 100644 --- a/vp9/common/x86/vp9_idct_intrin_sse2.c +++ b/vp9/common/x86/vp9_idct_intrin_sse2.c @@ -3205,34 +3205,20 @@ void vp9_idct32x32_34_add_sse2(const int16_t *input, uint8_t *dest, // idct constants for each stage const __m128i stg1_0 = pair_set_epi16(cospi_31_64, -cospi_1_64); const __m128i stg1_1 = pair_set_epi16(cospi_1_64, cospi_31_64); - const __m128i stg1_2 = pair_set_epi16(cospi_15_64, -cospi_17_64); - const __m128i stg1_3 = pair_set_epi16(cospi_17_64, cospi_15_64); - const __m128i stg1_4 = pair_set_epi16(cospi_23_64, -cospi_9_64); - const __m128i stg1_5 = pair_set_epi16(cospi_9_64, cospi_23_64); const __m128i stg1_6 = pair_set_epi16(cospi_7_64, -cospi_25_64); const __m128i stg1_7 = pair_set_epi16(cospi_25_64, cospi_7_64); const __m128i stg1_8 = pair_set_epi16(cospi_27_64, -cospi_5_64); const __m128i stg1_9 = pair_set_epi16(cospi_5_64, cospi_27_64); - const __m128i stg1_10 = pair_set_epi16(cospi_11_64, -cospi_21_64); - const __m128i stg1_11 = pair_set_epi16(cospi_21_64, cospi_11_64); - const __m128i stg1_12 = pair_set_epi16(cospi_19_64, -cospi_13_64); - const __m128i stg1_13 = pair_set_epi16(cospi_13_64, cospi_19_64); const __m128i stg1_14 = pair_set_epi16(cospi_3_64, -cospi_29_64); const __m128i stg1_15 = pair_set_epi16(cospi_29_64, cospi_3_64); const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64); const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64); - const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64); - const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64); - const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64); - const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64); const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64); const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64); const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64); - const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64); - const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64); const __m128i stg3_4 = pair_set_epi16(-cospi_4_64, cospi_28_64); const __m128i stg3_5 = pair_set_epi16(cospi_28_64, cospi_4_64); const __m128i stg3_6 = pair_set_epi16(-cospi_28_64, -cospi_4_64); @@ -3242,8 +3228,6 @@ void vp9_idct32x32_34_add_sse2(const int16_t *input, uint8_t *dest, const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64); const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64); - const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64); - const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64); const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64); const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64); const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64); @@ -3263,47 +3247,29 @@ void vp9_idct32x32_34_add_sse2(const int16_t *input, uint8_t *dest, stp2_30, stp2_31; __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; int i; - // Load input data. - LOAD_DQCOEFF(in[0], input); - LOAD_DQCOEFF(in[8], input); - LOAD_DQCOEFF(in[16], input); - LOAD_DQCOEFF(in[24], input); - LOAD_DQCOEFF(in[1], input); - LOAD_DQCOEFF(in[9], input); - LOAD_DQCOEFF(in[17], input); - LOAD_DQCOEFF(in[25], input); - LOAD_DQCOEFF(in[2], input); - LOAD_DQCOEFF(in[10], input); - LOAD_DQCOEFF(in[18], input); - LOAD_DQCOEFF(in[26], input); - LOAD_DQCOEFF(in[3], input); - LOAD_DQCOEFF(in[11], input); - LOAD_DQCOEFF(in[19], input); - LOAD_DQCOEFF(in[27], input); - - LOAD_DQCOEFF(in[4], input); - LOAD_DQCOEFF(in[12], input); - LOAD_DQCOEFF(in[20], input); - LOAD_DQCOEFF(in[28], input); - LOAD_DQCOEFF(in[5], input); - LOAD_DQCOEFF(in[13], input); - LOAD_DQCOEFF(in[21], input); - LOAD_DQCOEFF(in[29], input); - LOAD_DQCOEFF(in[6], input); - LOAD_DQCOEFF(in[14], input); - LOAD_DQCOEFF(in[22], input); - LOAD_DQCOEFF(in[30], input); - LOAD_DQCOEFF(in[7], input); - LOAD_DQCOEFF(in[15], input); - LOAD_DQCOEFF(in[23], input); - LOAD_DQCOEFF(in[31], input); + + // Load input data. Only need to load the top left 8x8 block. + in[0] = _mm_load_si128((const __m128i *)input); + in[1] = _mm_load_si128((const __m128i *)(input + 32)); + in[2] = _mm_load_si128((const __m128i *)(input + 64)); + in[3] = _mm_load_si128((const __m128i *)(input + 96)); + in[4] = _mm_load_si128((const __m128i *)(input + 128)); + in[5] = _mm_load_si128((const __m128i *)(input + 160)); + in[6] = _mm_load_si128((const __m128i *)(input + 192)); + in[7] = _mm_load_si128((const __m128i *)(input + 224)); + + for (i = 8; i < 32; ++i) { + in[i] = _mm_setzero_si128(); + } array_transpose_8x8(in, in); + // TODO(hkuang): Following transposes are unnecessary. But remove them will + // lead to performance drop on some devices. array_transpose_8x8(in + 8, in + 8); array_transpose_8x8(in + 16, in + 16); array_transpose_8x8(in + 24, in + 24); - IDCT32 + IDCT32_34 // 1_D: Store 32 intermediate results for each 8x32 block. col[0] = _mm_add_epi16(stp1_0, stp1_31); -- 2.50.1