diff options
author | Xin Li <delphij@google.com> | 2022-04-01 20:32:27 +0000 |
---|---|---|
committer | Xin Li <delphij@google.com> | 2022-04-01 20:32:27 +0000 |
commit | aa4f8822088582e1532f2f135a51643c89bd823d (patch) | |
tree | 8bfac2959f81c8dd9362fad804a4c04f8cb718d0 | |
parent | 9a308447a4e42482f63dc540b4fb4fa135986e36 (diff) | |
parent | f5513f57222ba2c14b68b72f350a550481cfe433 (diff) | |
download | libavc-aa4f8822088582e1532f2f135a51643c89bd823d.tar.gz |
Merge Android 12 QPR3 ab/8391262
Bug: 226662282
Merged-In: I8e4f1f4c57fae1510835733173c505326f39e28f
Change-Id: Ie5766e7c1506152070639040b82a9d2ca66766c1
-rw-r--r-- | common/x86/ih264_iquant_itrans_recon_sse42.c | 8 | ||||
-rw-r--r-- | common/x86/ih264_platform_macros.h | 14 | ||||
-rw-r--r-- | common/x86/ih264_resi_trans_quant_sse42.c | 9 |
3 files changed, 23 insertions, 8 deletions
diff --git a/common/x86/ih264_iquant_itrans_recon_sse42.c b/common/x86/ih264_iquant_itrans_recon_sse42.c index a7b9e82..3ae18a5 100644 --- a/common/x86/ih264_iquant_itrans_recon_sse42.c +++ b/common/x86/ih264_iquant_itrans_recon_sse42.c @@ -227,10 +227,10 @@ void ih264_iquant_itrans_recon_4x4_sse42(WORD16 *pi2_src, //Transform ends -- horizontal transform //Load pred buffer - pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0])); //p00 p01 p02 p03 0 0 0 0 0 0 0 0 -- all 8 bits - pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[pred_strd])); //p10 p11 p12 p13 0 0 0 0 0 0 0 0 -- all 8 bits - pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * pred_strd])); //p20 p21 p22 p23 0 0 0 0 0 0 0 0 -- all 8 bits - pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * pred_strd])); //p30 p31 p32 p33 0 0 0 0 0 0 0 0 -- all 8 bits + pred_r0 = loadu_32(&pu1_pred[0]); //p00 p01 p02 p03 -- all 8 bits + pred_r1 = loadu_32(&pu1_pred[pred_strd]); //p10 p11 p12 p13 -- all 8 bits + pred_r2 = loadu_32(&pu1_pred[2 * pred_strd]); //p20 p21 p22 p23 -- all 8 bits + pred_r3 = loadu_32(&pu1_pred[3 * pred_strd]); //p30 p31 p32 p33 -- all 8 bits pred_r0 = _mm_cvtepu8_epi32(pred_r0); //p00 p01 p02 p03 -- all 32 bits pred_r1 = _mm_cvtepu8_epi32(pred_r1); //p10 p11 p12 p13 -- all 32 bits diff --git a/common/x86/ih264_platform_macros.h b/common/x86/ih264_platform_macros.h index 54af325..a733617 100644 --- a/common/x86/ih264_platform_macros.h +++ b/common/x86/ih264_platform_macros.h @@ -41,6 +41,20 @@ #include <stdint.h> #include <immintrin.h> +#ifndef __ANDROID__ +static __inline__ __m128i +loadu_32(void const *__a) +{ + struct __loadu_si32 { + int __v; + } __attribute__((__packed__, __may_alias__)); + int __u = ((struct __loadu_si32*)__a)->__v; + return __extension__ (__m128i)(__v4si){__u, 0, 0, 0}; +} +#else +static __inline__ __m128i loadu_32(void const *__a) { return _mm_loadu_si32(__a); }; +#endif + #define CLIP_U8(x) CLIP3(0, UINT8_MAX, (x)) #define CLIP_S8(x) CLIP3(INT8_MIN, INT8_MAX, (x)) diff --git a/common/x86/ih264_resi_trans_quant_sse42.c b/common/x86/ih264_resi_trans_quant_sse42.c index f4f5cbf..232ab1a 100644 --- a/common/x86/ih264_resi_trans_quant_sse42.c +++ b/common/x86/ih264_resi_trans_quant_sse42.c @@ -46,6 +46,7 @@ #include "ih264_defs.h" #include "ih264_size_defs.h" #include "ih264_macros.h" +#include "ih264_platform_macros.h" #include "ih264_trans_macros.h" #include "ih264_trans_data.h" #include "ih264_structs.h" @@ -136,10 +137,10 @@ void ih264_resi_trans_quant_4x4_sse42(UWORD8 *pu1_src, UWORD8 *pu1_pred, src_r2 = _mm_cvtepu8_epi16(src_r2); src_r3 = _mm_cvtepu8_epi16(src_r3); - pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0])); //p00 p01 p02 p03 0 0 0 0 0 0 0 0 -- all 8 bits - pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[pred_strd])); //p10 p11 p12 p13 0 0 0 0 0 0 0 0 -- all 8 bits - pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * pred_strd])); //p20 p21 p22 p23 0 0 0 0 0 0 0 0 -- all 8 bits - pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * pred_strd])); //p30 p31 p32 p33 0 0 0 0 0 0 0 0 -- all 8 bits + pred_r0 = loadu_32(&pu1_pred[0]); //p00 p01 p02 p03 -- all 8 bits + pred_r1 = loadu_32(&pu1_pred[pred_strd]); //p10 p11 p12 p13 -- all 8 bits + pred_r2 = loadu_32(&pu1_pred[2 * pred_strd]); //p20 p21 p22 p23 -- all 8 bits + pred_r3 = loadu_32(&pu1_pred[3 * pred_strd]); //p30 p31 p32 p33 -- all 8 bits pred_r0 = _mm_cvtepu8_epi16(pred_r0); //p00 p01 p02 p03 -- all 16 bits pred_r1 = _mm_cvtepu8_epi16(pred_r1); //p10 p11 p12 p13 -- all 16 bits |