ref: 7760836ddbcc9250ccb717d13918a37e3722604a
dir: /hantest.c/
#include <u.h>
#include <libc.h>
#include "g1regs.h"
enum {
DecodeH264,
Baseline = 66,
High = 100,
};
typedef struct VPUctx VPUctx;
#define Default_4x4_Intra {6, 13, 13, 20, 20, 20, 28, 28, 28, 28, 32, 32, 32, 37, 37, 42}
#define Default_4x4_Inter {10, 14, 14, 20, 20, 20, 24, 24, 24, 24, 27, 27, 27, 30, 30, 34}
#define Default_8x8_Intra \
{6, 10, 10, 13, 11, 13, 16, 16, 16, 16, 18, 18, 18, 18, 18, 23, \
23, 23, 23, 23, 23, 25, 25, 25, 25, 25, 25, 25, 27, 27, 27, 27, \
27, 27, 27, 27, 29, 29, 29, 29, 29, 29, 29, 31, 31, 31, 31, 31, \
31, 33, 33, 33, 33, 33, 36, 36, 36, 36, 38, 38, 38, 40, 40, 42}
#define Default_8x8_Inter \
{9, 13, 13, 15, 13, 15, 17, 17, 17, 17, 19, 19, 19, 19, 19, 21, \
21, 21, 21, 21, 21, 22, 22, 22, 22, 22, 22, 22, 24, 24, 24, 24, \
24, 24, 24, 24, 25, 25, 25, 25, 25, 25, 25, 27, 27, 27, 27, 27, \
27, 28, 28, 28, 28, 28, 30, 30, 30, 30, 32, 32, 32, 33, 33, 35}
static u8int defsl4x4[6][16] = {
Default_4x4_Intra, /* Sl_4x4_Intra_Y */
Default_4x4_Intra, /* Sl_4x4_Intra_Cb */
Default_4x4_Intra, /* Sl_4x4_Intra_Cr */
Default_4x4_Inter, /* Sl_4x4_Inter_Y */
Default_4x4_Inter, /* Sl_4x4_Inter_Cb */
Default_4x4_Inter, /* Sl_4x4_Inter_Cr */
};
static u8int defs8x8[6][64] = {
Default_8x8_Intra, /* Sl_8x8_Intra_Y */
Default_8x8_Inter, /* Sl_8x8_Inter_Y */
Default_8x8_Intra, /* Sl_8x8_Intra_Cb */
Default_8x8_Inter, /* Sl_8x8_Inter_Cb */
Default_8x8_Intra, /* Sl_8x8_Intra_Cr */
Default_8x8_Inter, /* Sl_8x8_Inter_Cr */
};
struct VPUctx {
int process;
int slen;
union {
/* h.264 decoding */
struct {
struct {
u16int pic_height_in_map_units_minus1;
u16int pic_width_in_mbs_minus1;
u8int chroma_format_idc;
u8int frame_mbs_only_flag;
u8int log2_max_frame_num_minus4;
u8int max_num_ref_frames;
u8int nal_ref_idc;
u8int profile_idc;
u8int direct_8x8_inference_flag;
u8int mb_adaptive_frame_field_flag;
u8int seq_scaling_matrix_present_flag;
}sps;
struct {
s8int chroma_qp_index_offset;
s8int pic_init_qp_minus26;
s8int second_chroma_qp_index_offset;
u8int weighted_bipred_idc;
u8int pic_parameter_set_id;
u8int num_ref_idx_l0_default_active_minus1;
u8int num_ref_idx_l1_default_active_minus1;
u8int constrained_intra_pred_flag;
u8int deblocking_filter_control_present_flag;
u8int entropy_coding_mode_flag;
u8int pic_scaling_matrix_present_flag;
u8int redundant_pic_cnt_present_flag;
u8int transform_8x8_mode_flag;
u8int weighted_pred_flag;
}pps;
struct {
u16int frame_num;
u16int idr_pic_id;
u32int size_dec_ref_pic_marking;
u8int size_pic_order_cnt_lsb;
u8int size_delta_pic_order_cnt_bottom;
u8int size_delta_pic_order_cnt0;
u8int size_delta_pic_order_cnt1;
u8int nal_unit_type;
u8int field_pic_flag;
u8int bottom_field_flag;
}sh;
};
};
};
static u32int *g1, *g2;
static void
g1load(VPUctx *c)
{
g1[2] = 16 << SW_DEC_MAX_BURST | /* AXI burst length = 16 */
1 << SW_DEC_STRSWAP32_E | 1 << SW_DEC_STRENDIAN_E | /* 64bit / little endian */
1 << SW_DEC_INSWAP32_E | 1 << SW_DEC_OUTSWAP32_E |
1 << SW_DEC_IN_ENDIAN | 1 << SW_DEC_OUT_ENDIAN |
1 << SW_DEC_TIMEOUT_E; /* timeout interrupt */
g1[3] = (c->sps.frame_mbs_only_flag == 0 &&
(c->sps.mb_adaptive_frame_field_flag || c->sh.field_pic_flag)) << SW_PIC_INTERLACE_E |
(c->sps.frame_mbs_only_flag == 0 && c->sh.field_pic_flag) << SW_PIC_FIELDMODE_E |
(c->sps.frame_mbs_only_flag == 0 && c->sh.bottom_field_flag == 0) << SW_PIC_TOPFIELD_E |
(c->sps.profile_idc > Baseline && c->sps.nal_ref_idc != 0) << SW_WRITE_MVS_E |
c->sps.mb_adaptive_frame_field_flag << SW_SEQ_MBAFF_E |
(c->sps.profile_idc > Baseline) << SW_PICORD_COUNT_E |
0xff << SW_DEC_AXI_WR_ID;
g1[4] = (c->sps.pic_width_in_mbs_minus1 + 1) << SW_PIC_MB_WIDTH |
(c->sps.pic_height_in_map_units_minus1 + 1) << SW_PIC_MB_HEIGHT_P |
c->sps.max_num_ref_frames << SW_REF_FRAMES;
g1[5] = 0 << SW_STRM_START_BIT |
c->pps.pic_scaling_matrix_present_flag << SW_SCALING_LIST_E |
(c->pps.chroma_qp_index_offset & ((1<<5)-1)) << SW_CH_QP_OFFSET |
(c->pps.second_chroma_qp_index_offset & ((1<<5)-1)) << SW_CH_QP_OFFSET2 |
(c->sps.frame_mbs_only_flag == 0) << SW_FIELDPIC_FLAG_E;
g1[6] = 1 << SW_START_CODE_E |
(c->pps.pic_init_qp_minus26 + 26) << SW_INIT_QP |
c->slen << SW_STREAM_LEN;
g1[7] = c->pps.entropy_coding_mode_flag << SW_CABAC_E |
(c->sps.chroma_format_idc == 0 && c->sps.profile_idc >= High) << SW_BLACKWHITE_E |
c->sps.direct_8x8_inference_flag << SW_DIR_8X8_INFER_E |
c->pps.weighted_pred_flag << SW_WEIGHT_PRED_E |
c->pps.weighted_bipred_idc << SW_WEIGHT_BIPR_IDC |
(c->sps.log2_max_frame_num_minus4 + 4) << SW_FRAMENUM_LEN |
c->sh.frame_num << SW_FRAMENUM;
g1[8] = c->pps.constrained_intra_pred_flag << SW_CONST_INTRA_E |
c->pps.deblocking_filter_control_present_flag << SW_FILT_CTRL_PRES |
c->pps.redundant_pic_cnt_present_flag << SW_RDPIC_CNT_PRES |
c->pps.transform_8x8_mode_flag << SW_8X8TRANS_FLAG_E |
(c->sps.nal_ref_idc != 0 ? c->sh.size_dec_ref_pic_marking : 0) << SW_REFPIC_MK_LEN |
(c->sh.nal_unit_type == 5) << SW_IDR_PIC_E |
c->sh.idr_pic_id << SW_IDR_PIC_ID;
g1[9] = c->pps.pic_parameter_set_id << SW_PPS_ID |
(c->pps.num_ref_idx_l1_default_active_minus1 + 1) << SW_REFIDX1_ACTIVE |
(c->pps.num_ref_idx_l0_default_active_minus1 + 1) << SW_REFIDX0_ACTIVE |
(c->sh.size_pic_order_cnt_lsb + c->sh.size_delta_pic_order_cnt_bottom +
c->sh.size_delta_pic_order_cnt0 + c->sh.size_delta_pic_order_cnt1) << SW_POC_LENGTH;
/* fixed 6-tap filter values (1 -5 20) */
g1[49] = 1 << SW_PRED_BC_TAP_0_0 |
(-5 & ((1<<(SW_PRED_BC_TAP_0_0-SW_PRED_BC_TAP_0_1))-1)) << SW_PRED_BC_TAP_0_1 |
20 << SW_PRED_BC_TAP_0_2;
}
static int
parsecmd(VPUctx *c, char *s)
{
char *a[4];
int n, x;
if((n = tokenize(s, a, nelem(a))) < 2)
return -1;
#define value(name, field, min, max) \
if(strcmp(a[1], name) == 0){ \
if(x < (min) || x > (max)){ \
werrstr("%s: out of range [%d, %d]", a[1], (min), (max)); \
return -1; \
} \
field = x; \
return 0; \
}
if(strcmp(a[0], "decode") == 0){
if(n == 2 && strcmp(a[1], "h.264") == 0){
c->process = DecodeH264;
return 0;
}
}else if(c->process == DecodeH264 && n == 3){
x = atoi(a[2]);
if(strcmp(a[0], "sps") == 0){
value("pic_height_in_map_units_minus1", c->sps.pic_height_in_map_units_minus1, 0, 0xff);
value("pic_width_in_mbs_minus1", c->sps.pic_width_in_mbs_minus1, 0, 0xff);
value("chroma_format_idc", c->sps.chroma_format_idc, 0, 3);
value("log2_max_frame_num_minus4", c->sps.log2_max_frame_num_minus4, 0, 12);
value("max_num_ref_frames", c->sps.max_num_ref_frames, 0, 16);
value("nal_ref_idc", c->sps.nal_ref_idc, 0, 3);
value("profile_idc", c->sps.profile_idc, 0, 0xff);
value("direct_8x8_inference_flag", c->sps.direct_8x8_inference_flag, 0, 1);
value("frame_mbs_only_flag", c->sps.frame_mbs_only_flag, 0, 1);
value("mb_adaptive_frame_field_flag", c->sps.mb_adaptive_frame_field_flag, 0, 1);
value("seq_scaling_matrix_present_flag", c->sps.seq_scaling_matrix_present_flag, 0, 1);
}else if(strcmp(a[0], "pps") == 0){
value("chroma_qp_index_offset", c->pps.chroma_qp_index_offset, -12, 12);
value("pic_init_qp_minus26", c->pps.pic_init_qp_minus26, -26, 25);
value("second_chroma_qp_index_offset", c->pps.second_chroma_qp_index_offset, -12, 12);
value("weighted_bipred_idc", c->pps.weighted_bipred_idc, 0, 2);
value("pic_parameter_set_id", c->pps.pic_parameter_set_id, 0, 0xff);
value("num_ref_idx_l0_default_active_minus1", c->pps.num_ref_idx_l0_default_active_minus1, 0, 31);
value("num_ref_idx_l1_default_active_minus1", c->pps.num_ref_idx_l1_default_active_minus1, 0, 31);
value("constrained_intra_pred_flag", c->pps.constrained_intra_pred_flag, 0, 1);
value("deblocking_filter_control_present_flag", c->pps.deblocking_filter_control_present_flag, 0, 1);
value("entropy_coding_mode_flag", c->pps.entropy_coding_mode_flag, 0, 1);
value("pic_scaling_matrix_present_flag", c->pps.pic_scaling_matrix_present_flag, 0, 1);
value("redundant_pic_cnt_present_flag", c->pps.redundant_pic_cnt_present_flag, 0, 1);
value("transform_8x8_mode_flag", c->pps.transform_8x8_mode_flag, 0, 1);
value("weighted_pred_flag", c->pps.weighted_pred_flag, 0, 1);
}else if(strcmp(a[0], "sh") == 0){
value("frame_num", c->sh.frame_num, 0, 0xffff);
value("idr_pic_id", c->sh.idr_pic_id, 0, 0xffff);
value("nal_unit_type", c->sh.nal_unit_type, 0, 0xff);
value("size_dec_ref_pic_marking", c->sh.size_dec_ref_pic_marking, 0, 0x1ffff);
value("size_pic_order_cnt_lsb", c->sh.size_pic_order_cnt_lsb, 0, 0xff);
value("size_delta_pic_order_cnt_bottom", c->sh.size_delta_pic_order_cnt_bottom, 0, 0xff);
value("size_delta_pic_order_cnt0", c->sh.size_delta_pic_order_cnt0, 0, 0xff);
value("size_delta_pic_order_cnt1", c->sh.size_delta_pic_order_cnt1, 0, 0xff);
value("field_pic_flag", c->sh.field_pic_flag, 0, 1);
value("bottom_field_flag", c->sh.bottom_field_flag, 0, 1);
}
}
#undef value
return -1;
}
void
main(int argc, char **argv)
{
int i;
u32int x;
ARGBEGIN{
}ARGEND
if((g1 = segattach(0, "vpu_hantro_g1", 0, 0x200)) == (void*)-1)
sysfatal("no G1");
/* FIXME will use polling for now */
g1[1] = SW_DEC_IRQ_DIS;
/* run */
g1[2] |= SW_DEC_CLK_GATE_E;
x = (g1[1] |= SW_DEC_E);
for(i = 0; i < 100; i++){
sleep(1);
if(g1[1] != x){
print("irq changed from 0x%ux to 0x%ux\n", x, g1[1]);
break;
}
}
segdetach(g1);
exits(nil);
}