summaryrefslogtreecommitdiffstats
path: root/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h
diff options
context:
space:
mode:
Diffstat (limited to 'third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h')
-rw-r--r--third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h207
1 files changed, 0 insertions, 207 deletions
diff --git a/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h b/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h
index 58e8971dd..12ccf7f26 100644
--- a/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h
+++ b/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.h
@@ -12,15 +12,10 @@
#ifndef AOM_DSP_X86_FWD_TXFM_SSE2_H_
#define AOM_DSP_X86_FWD_TXFM_SSE2_H_
-#include "aom_dsp/x86/txfm_common_intrin.h"
-
#ifdef __cplusplus
extern "C" {
#endif
-#define pair_set_epi32(a, b) \
- _mm_set_epi32((int)(b), (int)(a), (int)(b), (int)(a))
-
static INLINE __m128i k_madd_epi32(__m128i a, __m128i b) {
__m128i buf0, buf1;
buf0 = _mm_mul_epu32(a, b);
@@ -140,112 +135,6 @@ static INLINE int check_epi16_overflow_x32(
return res0 + res1;
}
-static INLINE int k_check_epi32_overflow_4(const __m128i *preg0,
- const __m128i *preg1,
- const __m128i *preg2,
- const __m128i *preg3,
- const __m128i *zero) {
- __m128i minus_one = _mm_set1_epi32(-1);
- // Check for overflows
- __m128i reg0_shifted = _mm_slli_epi64(*preg0, 1);
- __m128i reg1_shifted = _mm_slli_epi64(*preg1, 1);
- __m128i reg2_shifted = _mm_slli_epi64(*preg2, 1);
- __m128i reg3_shifted = _mm_slli_epi64(*preg3, 1);
- __m128i reg0_top_dwords =
- _mm_shuffle_epi32(reg0_shifted, _MM_SHUFFLE(0, 0, 3, 1));
- __m128i reg1_top_dwords =
- _mm_shuffle_epi32(reg1_shifted, _MM_SHUFFLE(0, 0, 3, 1));
- __m128i reg2_top_dwords =
- _mm_shuffle_epi32(reg2_shifted, _MM_SHUFFLE(0, 0, 3, 1));
- __m128i reg3_top_dwords =
- _mm_shuffle_epi32(reg3_shifted, _MM_SHUFFLE(0, 0, 3, 1));
- __m128i top_dwords_01 = _mm_unpacklo_epi64(reg0_top_dwords, reg1_top_dwords);
- __m128i top_dwords_23 = _mm_unpacklo_epi64(reg2_top_dwords, reg3_top_dwords);
- __m128i valid_positve_01 = _mm_cmpeq_epi32(top_dwords_01, *zero);
- __m128i valid_positve_23 = _mm_cmpeq_epi32(top_dwords_23, *zero);
- __m128i valid_negative_01 = _mm_cmpeq_epi32(top_dwords_01, minus_one);
- __m128i valid_negative_23 = _mm_cmpeq_epi32(top_dwords_23, minus_one);
- int overflow_01 =
- _mm_movemask_epi8(_mm_cmpeq_epi32(valid_positve_01, valid_negative_01));
- int overflow_23 =
- _mm_movemask_epi8(_mm_cmpeq_epi32(valid_positve_23, valid_negative_23));
- return (overflow_01 + overflow_23);
-}
-
-static INLINE int k_check_epi32_overflow_8(
- const __m128i *preg0, const __m128i *preg1, const __m128i *preg2,
- const __m128i *preg3, const __m128i *preg4, const __m128i *preg5,
- const __m128i *preg6, const __m128i *preg7, const __m128i *zero) {
- int overflow = k_check_epi32_overflow_4(preg0, preg1, preg2, preg3, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg4, preg5, preg6, preg7, zero);
- }
- return overflow;
-}
-
-static INLINE int k_check_epi32_overflow_16(
- const __m128i *preg0, const __m128i *preg1, const __m128i *preg2,
- const __m128i *preg3, const __m128i *preg4, const __m128i *preg5,
- const __m128i *preg6, const __m128i *preg7, const __m128i *preg8,
- const __m128i *preg9, const __m128i *preg10, const __m128i *preg11,
- const __m128i *preg12, const __m128i *preg13, const __m128i *preg14,
- const __m128i *preg15, const __m128i *zero) {
- int overflow = k_check_epi32_overflow_4(preg0, preg1, preg2, preg3, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg4, preg5, preg6, preg7, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg8, preg9, preg10, preg11, zero);
- if (!overflow) {
- overflow =
- k_check_epi32_overflow_4(preg12, preg13, preg14, preg15, zero);
- }
- }
- }
- return overflow;
-}
-
-static INLINE int k_check_epi32_overflow_32(
- const __m128i *preg0, const __m128i *preg1, const __m128i *preg2,
- const __m128i *preg3, const __m128i *preg4, const __m128i *preg5,
- const __m128i *preg6, const __m128i *preg7, const __m128i *preg8,
- const __m128i *preg9, const __m128i *preg10, const __m128i *preg11,
- const __m128i *preg12, const __m128i *preg13, const __m128i *preg14,
- const __m128i *preg15, const __m128i *preg16, const __m128i *preg17,
- const __m128i *preg18, const __m128i *preg19, const __m128i *preg20,
- const __m128i *preg21, const __m128i *preg22, const __m128i *preg23,
- const __m128i *preg24, const __m128i *preg25, const __m128i *preg26,
- const __m128i *preg27, const __m128i *preg28, const __m128i *preg29,
- const __m128i *preg30, const __m128i *preg31, const __m128i *zero) {
- int overflow = k_check_epi32_overflow_4(preg0, preg1, preg2, preg3, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg4, preg5, preg6, preg7, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg8, preg9, preg10, preg11, zero);
- if (!overflow) {
- overflow =
- k_check_epi32_overflow_4(preg12, preg13, preg14, preg15, zero);
- if (!overflow) {
- overflow =
- k_check_epi32_overflow_4(preg16, preg17, preg18, preg19, zero);
- if (!overflow) {
- overflow =
- k_check_epi32_overflow_4(preg20, preg21, preg22, preg23, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg24, preg25, preg26,
- preg27, zero);
- if (!overflow) {
- overflow = k_check_epi32_overflow_4(preg28, preg29, preg30,
- preg31, zero);
- }
- }
- }
- }
- }
- }
- }
- return overflow;
-}
-
static INLINE void store_output(const __m128i *poutput, tran_low_t *dst_ptr) {
if (sizeof(tran_low_t) == 4) {
const __m128i zero = _mm_setzero_si128();
@@ -259,102 +148,6 @@ static INLINE void store_output(const __m128i *poutput, tran_low_t *dst_ptr) {
}
}
-static INLINE __m128i mult_round_shift(const __m128i *pin0, const __m128i *pin1,
- const __m128i *pmultiplier,
- const __m128i *prounding, int shift) {
- const __m128i u0 = _mm_madd_epi16(*pin0, *pmultiplier);
- const __m128i u1 = _mm_madd_epi16(*pin1, *pmultiplier);
- const __m128i v0 = _mm_add_epi32(u0, *prounding);
- const __m128i v1 = _mm_add_epi32(u1, *prounding);
- const __m128i w0 = _mm_srai_epi32(v0, shift);
- const __m128i w1 = _mm_srai_epi32(v1, shift);
- return _mm_packs_epi32(w0, w1);
-}
-
-static INLINE void transpose_and_output8x8(
- const __m128i *pin00, const __m128i *pin01, const __m128i *pin02,
- const __m128i *pin03, const __m128i *pin04, const __m128i *pin05,
- const __m128i *pin06, const __m128i *pin07, int pass, int16_t *out0_ptr,
- tran_low_t *out1_ptr) {
- // 00 01 02 03 04 05 06 07
- // 10 11 12 13 14 15 16 17
- // 20 21 22 23 24 25 26 27
- // 30 31 32 33 34 35 36 37
- // 40 41 42 43 44 45 46 47
- // 50 51 52 53 54 55 56 57
- // 60 61 62 63 64 65 66 67
- // 70 71 72 73 74 75 76 77
- const __m128i tr0_0 = _mm_unpacklo_epi16(*pin00, *pin01);
- const __m128i tr0_1 = _mm_unpacklo_epi16(*pin02, *pin03);
- const __m128i tr0_2 = _mm_unpackhi_epi16(*pin00, *pin01);
- const __m128i tr0_3 = _mm_unpackhi_epi16(*pin02, *pin03);
- const __m128i tr0_4 = _mm_unpacklo_epi16(*pin04, *pin05);
- const __m128i tr0_5 = _mm_unpacklo_epi16(*pin06, *pin07);
- const __m128i tr0_6 = _mm_unpackhi_epi16(*pin04, *pin05);
- const __m128i tr0_7 = _mm_unpackhi_epi16(*pin06, *pin07);
- // 00 10 01 11 02 12 03 13
- // 20 30 21 31 22 32 23 33
- // 04 14 05 15 06 16 07 17
- // 24 34 25 35 26 36 27 37
- // 40 50 41 51 42 52 43 53
- // 60 70 61 71 62 72 63 73
- // 54 54 55 55 56 56 57 57
- // 64 74 65 75 66 76 67 77
- const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
- const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3);
- const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
- const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3);
- const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
- const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7);
- const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);
- const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7);
- // 00 10 20 30 01 11 21 31
- // 40 50 60 70 41 51 61 71
- // 02 12 22 32 03 13 23 33
- // 42 52 62 72 43 53 63 73
- // 04 14 24 34 05 15 21 36
- // 44 54 64 74 45 55 61 76
- // 06 16 26 36 07 17 27 37
- // 46 56 66 76 47 57 67 77
- const __m128i tr2_0 = _mm_unpacklo_epi64(tr1_0, tr1_4);
- const __m128i tr2_1 = _mm_unpackhi_epi64(tr1_0, tr1_4);
- const __m128i tr2_2 = _mm_unpacklo_epi64(tr1_2, tr1_6);
- const __m128i tr2_3 = _mm_unpackhi_epi64(tr1_2, tr1_6);
- const __m128i tr2_4 = _mm_unpacklo_epi64(tr1_1, tr1_5);
- const __m128i tr2_5 = _mm_unpackhi_epi64(tr1_1, tr1_5);
- const __m128i tr2_6 = _mm_unpacklo_epi64(tr1_3, tr1_7);
- const __m128i tr2_7 = _mm_unpackhi_epi64(tr1_3, tr1_7);
- // 00 10 20 30 40 50 60 70
- // 01 11 21 31 41 51 61 71
- // 02 12 22 32 42 52 62 72
- // 03 13 23 33 43 53 63 73
- // 04 14 24 34 44 54 64 74
- // 05 15 25 35 45 55 65 75
- // 06 16 26 36 46 56 66 76
- // 07 17 27 37 47 57 67 77
- if (pass == 0) {
- _mm_storeu_si128((__m128i *)(out0_ptr + 0 * 16), tr2_0);
- _mm_storeu_si128((__m128i *)(out0_ptr + 1 * 16), tr2_1);
- _mm_storeu_si128((__m128i *)(out0_ptr + 2 * 16), tr2_2);
- _mm_storeu_si128((__m128i *)(out0_ptr + 3 * 16), tr2_3);
- _mm_storeu_si128((__m128i *)(out0_ptr + 4 * 16), tr2_4);
- _mm_storeu_si128((__m128i *)(out0_ptr + 5 * 16), tr2_5);
- _mm_storeu_si128((__m128i *)(out0_ptr + 6 * 16), tr2_6);
- _mm_storeu_si128((__m128i *)(out0_ptr + 7 * 16), tr2_7);
- } else {
- storeu_output(&tr2_0, (out1_ptr + 0 * 16));
- storeu_output(&tr2_1, (out1_ptr + 1 * 16));
- storeu_output(&tr2_2, (out1_ptr + 2 * 16));
- storeu_output(&tr2_3, (out1_ptr + 3 * 16));
- storeu_output(&tr2_4, (out1_ptr + 4 * 16));
- storeu_output(&tr2_5, (out1_ptr + 5 * 16));
- storeu_output(&tr2_6, (out1_ptr + 6 * 16));
- storeu_output(&tr2_7, (out1_ptr + 7 * 16));
- }
-}
-
-void fdct32_8col(__m128i *in0, __m128i *in1);
-
#ifdef __cplusplus
} // extern "C"
#endif
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-26 22:40:24 +0000