Merge "Merge QQ3A.200605.002 into master"
diff --git a/common/x86/ihevc_intra_pred_filters_sse42_intr.c b/common/x86/ihevc_intra_pred_filters_sse42_intr.c
index 6488de6..e72f042 100644
--- a/common/x86/ihevc_intra_pred_filters_sse42_intr.c
+++ b/common/x86/ihevc_intra_pred_filters_sse42_intr.c
@@ -1969,7 +1969,7 @@
__m128i row_4x32b, two_nt_4x32b, ref_main_idx_4x32b, res_temp5_4x32b, sm3;
- UWORD8 ref_tmp[2 * MAX_CU_SIZE + 2];
+ UWORD8 ref_tmp[2 * MAX_CU_SIZE + 2] = {0};
UWORD8 *ref_main;
UWORD8 *ref_temp;
UNUSED(src_strd);
@@ -2799,7 +2799,7 @@
WORD32 inv_ang, inv_ang_sum;
//WORD32 ref_main_idx, pos, fract, idx;
WORD32 ref_idx;
- UWORD8 ref_tmp[(2 * MAX_CU_SIZE) + 2];
+ UWORD8 ref_tmp[(2 * MAX_CU_SIZE) + 2] = {0};
UWORD8 *ref_main, *ref_temp;
__m128i /*fract_8x16b,*/ const_temp_8x16b, sm3;
diff --git a/decoder/ihevcd_decode.c b/decoder/ihevcd_decode.c
index 16ed36b..9f634b7 100644
--- a/decoder/ihevcd_decode.c
+++ b/decoder/ihevcd_decode.c
@@ -626,6 +626,13 @@
nal_ofst = ihevcd_nal_search_start_code(ps_codec->pu1_inp_bitsbuf,
ps_codec->i4_bytes_remaining);
+ /* If there is no start code found, consume the data and break */
+ if(nal_ofst == ps_codec->i4_bytes_remaining)
+ {
+ ps_codec->pu1_inp_bitsbuf += nal_ofst;
+ ps_codec->i4_bytes_remaining -= nal_ofst;
+ break;
+ }
ps_codec->i4_nal_ofst = nal_ofst;
{
diff --git a/decoder/ihevcd_fmt_conv.c b/decoder/ihevcd_fmt_conv.c
index 4e0e4f7..4637fe8 100644
--- a/decoder/ihevcd_fmt_conv.c
+++ b/decoder/ihevcd_fmt_conv.c
@@ -65,6 +65,8 @@
#include "ihevcd_fmt_conv.h"
#include "ihevcd_profile.h"
+/* SIMD variants of format conversion modules do not support width less than 32 */
+#define MIN_FMT_CONV_SIMD_WIDTH 32
/**
*******************************************************************************
*
@@ -830,18 +832,35 @@
if((IV_YUV_420SP_UV == ps_codec->e_chroma_fmt) || (IV_YUV_420SP_VU == ps_codec->e_chroma_fmt))
{
-
- ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_420sp_fptr(pu1_y_src, pu1_uv_src,
- pu1_y_dst_tmp, pu1_uv_dst_tmp,
- ps_codec->i4_disp_wd,
- num_rows,
- ps_codec->i4_strd,
- ps_codec->i4_strd,
- ps_codec->i4_disp_strd,
- ps_codec->i4_disp_strd);
+ ihevcd_fmt_conv_420sp_to_420sp_ft *fmt_conv_fptr;
+ if(ps_codec->i4_disp_wd >= MIN_FMT_CONV_SIMD_WIDTH)
+ {
+ fmt_conv_fptr = ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_420sp_fptr;
+ }
+ else
+ {
+ fmt_conv_fptr = ihevcd_fmt_conv_420sp_to_420sp;
+ }
+ fmt_conv_fptr(pu1_y_src, pu1_uv_src,
+ pu1_y_dst_tmp, pu1_uv_dst_tmp,
+ ps_codec->i4_disp_wd,
+ num_rows,
+ ps_codec->i4_strd,
+ ps_codec->i4_strd,
+ ps_codec->i4_disp_strd,
+ ps_codec->i4_disp_strd);
}
else if(IV_YUV_420P == ps_codec->e_chroma_fmt)
{
+ ihevcd_fmt_conv_420sp_to_420p_ft *fmt_conv_fptr;
+ if(ps_codec->i4_disp_wd >= MIN_FMT_CONV_SIMD_WIDTH)
+ {
+ fmt_conv_fptr = ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_420p_fptr;
+ }
+ else
+ {
+ fmt_conv_fptr = ihevcd_fmt_conv_420sp_to_420p;
+ }
if(0 == disable_luma_copy)
{
@@ -858,46 +877,60 @@
disable_luma_copy = 1;
}
-
- ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_420p_fptr(pu1_y_src, pu1_uv_src,
- pu1_y_dst_tmp, pu1_u_dst_tmp, pu1_v_dst_tmp,
- ps_codec->i4_disp_wd,
- num_rows,
- ps_codec->i4_strd,
- ps_codec->i4_strd,
- ps_codec->i4_disp_strd,
- (ps_codec->i4_disp_strd / 2),
- is_u_first,
- disable_luma_copy);
-
+ fmt_conv_fptr(pu1_y_src, pu1_uv_src,
+ pu1_y_dst_tmp, pu1_u_dst_tmp, pu1_v_dst_tmp,
+ ps_codec->i4_disp_wd,
+ num_rows,
+ ps_codec->i4_strd,
+ ps_codec->i4_strd,
+ ps_codec->i4_disp_strd,
+ (ps_codec->i4_disp_strd / 2),
+ is_u_first,
+ disable_luma_copy);
}
else if(IV_RGB_565 == ps_codec->e_chroma_fmt)
{
+ ihevcd_fmt_conv_420sp_to_rgb565_ft *fmt_conv_fptr;
+ if(ps_codec->i4_disp_wd >= MIN_FMT_CONV_SIMD_WIDTH)
+ {
+ fmt_conv_fptr = ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_rgb565_fptr;
+ }
+ else
+ {
+ fmt_conv_fptr = ihevcd_fmt_conv_420sp_to_rgb565;
+ }
- ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_rgb565_fptr(pu1_y_src, pu1_uv_src,
- pu2_rgb_dst_tmp,
- ps_codec->i4_disp_wd,
- num_rows,
- ps_codec->i4_strd,
- ps_codec->i4_strd,
- ps_codec->i4_disp_strd,
- is_u_first);
-
+ fmt_conv_fptr(pu1_y_src, pu1_uv_src,
+ pu2_rgb_dst_tmp,
+ ps_codec->i4_disp_wd,
+ num_rows,
+ ps_codec->i4_strd,
+ ps_codec->i4_strd,
+ ps_codec->i4_disp_strd,
+ is_u_first);
}
else if(IV_RGBA_8888 == ps_codec->e_chroma_fmt)
{
+ ihevcd_fmt_conv_420sp_to_rgba8888_ft *fmt_conv_fptr;
+ if(ps_codec->i4_disp_wd >= MIN_FMT_CONV_SIMD_WIDTH)
+ {
+ fmt_conv_fptr = ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_rgba8888_fptr;
+ }
+ else
+ {
+ fmt_conv_fptr = ihevcd_fmt_conv_420sp_to_rgba8888;
+ }
+
ASSERT(is_u_first == 1);
-
- ps_codec->s_func_selector.ihevcd_fmt_conv_420sp_to_rgba8888_fptr(pu1_y_src,
- pu1_uv_src,
- pu4_rgb_dst_tmp,
- ps_codec->i4_disp_wd,
- num_rows,
- ps_codec->i4_strd,
- ps_codec->i4_strd,
- ps_codec->i4_disp_strd,
- is_u_first);
-
+ fmt_conv_fptr(pu1_y_src,
+ pu1_uv_src,
+ pu4_rgb_dst_tmp,
+ ps_codec->i4_disp_wd,
+ num_rows,
+ ps_codec->i4_strd,
+ ps_codec->i4_strd,
+ ps_codec->i4_disp_strd,
+ is_u_first);
}
diff --git a/decoder/ihevcd_iquant_itrans_recon_ctb.c b/decoder/ihevcd_iquant_itrans_recon_ctb.c
index 8425506..504271c 100644
--- a/decoder/ihevcd_iquant_itrans_recon_ctb.c
+++ b/decoder/ihevcd_iquant_itrans_recon_ctb.c
@@ -946,7 +946,7 @@
/* While (MAX_TU_SIZE * 2 * 2) + 1 is the actaul size needed,
au1_ref_sub_out size is kept as multiple of 8,
so that SIMD functions can load 64 bits */
- UWORD8 au1_ref_sub_out[(MAX_TU_SIZE * 2 * 2) + 8];
+ UWORD8 au1_ref_sub_out[(MAX_TU_SIZE * 2 * 2) + 8] = {0};
UWORD8 *pu1_top_left, *pu1_top, *pu1_left;
WORD32 luma_pred_func_idx, chroma_pred_func_idx;
diff --git a/decoder/ihevcd_parse_slice_header.c b/decoder/ihevcd_parse_slice_header.c
index c161fc4..23d5030 100644
--- a/decoder/ihevcd_parse_slice_header.c
+++ b/decoder/ihevcd_parse_slice_header.c
@@ -325,6 +325,12 @@
{
BITS_PARSE("dependent_slice_flag", value, ps_bitstrm, 1);
+ /* First slice to be decoded in the current picture can't be dependent slice */
+ if (value && 0 == ps_codec->i4_pic_present)
+ {
+ return IHEVCD_IGNORE_SLICE;
+ }
+
/* If dependendent slice, copy slice header from previous slice */
if(value && (ps_codec->s_parse.i4_cur_slice_idx > 0))
{
@@ -471,7 +477,8 @@
ps_slice_hdr->i1_num_long_term_sps = value;
}
UEV_PARSE("num_long_term_pics", value, ps_bitstrm);
- if((value + ps_slice_hdr->i1_num_long_term_sps + num_neg_pics + num_pos_pics) > (MAX_DPB_SIZE - 1))
+ if(((ULWORD64)value + ps_slice_hdr->i1_num_long_term_sps + num_neg_pics +
+ num_pos_pics) > (MAX_DPB_SIZE - 1))
{
return IHEVCD_INVALID_PARAMETER;
}
@@ -487,6 +494,10 @@
{
WORD32 num_bits = 32 - CLZ(ps_sps->i1_num_long_term_ref_pics_sps - 1);
BITS_PARSE("lt_idx_sps[ i ]", value, ps_bitstrm, num_bits);
+ if(value >= ps_sps->i1_num_long_term_ref_pics_sps)
+ {
+ return IHEVCD_INVALID_PARAMETER;
+ }
}
else
{
diff --git a/encoder/ihevce_enc_loop_pass.c b/encoder/ihevce_enc_loop_pass.c
index 16f2280..c73c7ff 100644
--- a/encoder/ihevce_enc_loop_pass.c
+++ b/encoder/ihevce_enc_loop_pass.c
@@ -141,6 +141,8 @@
/* Constant Macros */
/*****************************************************************************/
#define UPDATE_QP_AT_CTB 6
+#define INTRAPRED_SIMD_LEFT_PADDING 16
+#define INTRAPRED_SIMD_RIGHT_PADDING 8
/*****************************************************************************/
/* Function Definitions */
@@ -3731,8 +3733,12 @@
ps_mem_tab[ENC_LOOP_CHROMA_PRED_INTRA].i4_mem_alignment = 8;
/* Memory required to store pred for reference substitution output */
+ /* While (MAX_TU_SIZE * 2 * 2) + 1 is the actual size needed,
+ allocate 16 bytes to the left and 7 bytes to the right to facilitate
+ SIMD access */
ps_mem_tab[ENC_LOOP_REF_SUB_OUT].i4_mem_size =
- i4_num_proc_thrds * ((MAX_TU_SIZE * 2 * 2) + 4) *
+ i4_num_proc_thrds * (((MAX_TU_SIZE * 2 * 2) + INTRAPRED_SIMD_RIGHT_PADDING)
+ + INTRAPRED_SIMD_LEFT_PADDING)*
((ps_init_prms->s_tgt_lyr_prms.i4_internal_bit_depth > 8) ? 2 : 1) * sizeof(UWORD8);
ps_mem_tab[ENC_LOOP_REF_SUB_OUT].e_mem_type = (IV_MEM_TYPE_T)i4_mem_space;
@@ -3740,8 +3746,12 @@
ps_mem_tab[ENC_LOOP_REF_SUB_OUT].i4_mem_alignment = 8;
/* Memory required to store pred for reference filtering output */
+ /* While (MAX_TU_SIZE * 2 * 2) + 1 is the actual size needed,
+ allocate 16 bytes to the left and 7 bytes to the right to facilitate
+ SIMD access */
ps_mem_tab[ENC_LOOP_REF_FILT_OUT].i4_mem_size =
- i4_num_proc_thrds * ((MAX_TU_SIZE * 2 * 2) + 4) *
+ i4_num_proc_thrds * (((MAX_TU_SIZE * 2 * 2) + INTRAPRED_SIMD_RIGHT_PADDING)
+ + INTRAPRED_SIMD_LEFT_PADDING)*
((ps_init_prms->s_tgt_lyr_prms.i4_internal_bit_depth > 8) ? 2 : 1) * sizeof(UWORD8);
ps_mem_tab[ENC_LOOP_REF_FILT_OUT].e_mem_type = (IV_MEM_TYPE_T)i4_mem_space;
@@ -4670,22 +4680,24 @@
/* Memory assignments for reference substitution output */
{
- WORD32 pred_buf_size = ((MAX_TU_SIZE * 2 * 2) + 4);
+ WORD32 pred_buf_size = ((MAX_TU_SIZE * 2 * 2) + INTRAPRED_SIMD_RIGHT_PADDING
+ + INTRAPRED_SIMD_LEFT_PADDING);
WORD32 pred_buf_size_per_thread = pred_buf_size;
UWORD8 *pu1_base = (UWORD8 *)ps_mem_tab[ENC_LOOP_REF_SUB_OUT].pv_base +
(ctr * pred_buf_size_per_thread);
- ps_ctxt->pv_ref_sub_out = pu1_base;
+ ps_ctxt->pv_ref_sub_out = pu1_base + INTRAPRED_SIMD_LEFT_PADDING;
}
/* Memory assignments for reference filtering output */
{
- WORD32 pred_buf_size = ((MAX_TU_SIZE * 2 * 2) + 4);
+ WORD32 pred_buf_size = ((MAX_TU_SIZE * 2 * 2) + INTRAPRED_SIMD_RIGHT_PADDING
+ + INTRAPRED_SIMD_LEFT_PADDING);
WORD32 pred_buf_size_per_thread = pred_buf_size;
UWORD8 *pu1_base = (UWORD8 *)ps_mem_tab[ENC_LOOP_REF_FILT_OUT].pv_base +
(ctr * pred_buf_size_per_thread);
- ps_ctxt->pv_ref_filt_out = pu1_base;
+ ps_ctxt->pv_ref_filt_out = pu1_base + INTRAPRED_SIMD_LEFT_PADDING;
}
/* Memory assignments for recon storage during CU Recursion */
