lixian
/
doubao_chat
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
doubao_chat/apps/common/example/video/image/pixel_flow/sad8x8.c

311 lines
17 KiB
C
Raw Blame History

#include "typedef.h"
static void avg8x8_aligned(u8 *block, const u8 *s0, const u8 *s1,
int dest_size, int line_size, int h)
{
__asm__ volatile(
" %3 = %3-4\n\t"
" %4 = %4-4\n\t"
" 1: \n\t"
" %5 -= 1 # r4 = [%1++=4]\n\t"
" r6 = [%2++=4]\n\t"
" r5 = [%1++=%4]\n\t"
" r7 = [%2++=%4]\n\t"
" r4 = r4 +|+|+|+ r6 (uavg,rnd) \n\t"
" r5 = r5 +|+|+|+ r7 (uavg,rnd) \n\t"
" [%0++=4] = r4\n\t"
" [%0++=%3] = r5\n\t"
"if(%5!=0) goto 1b \n\t"
:"=&r"(block), "=&r"(s0), "=&r"(s1), "=&r"(dest_size), "=&r"(line_size), "=&r"(h)
:"0"(block), "1"(s0), "2"(s1), "3"(dest_size), "4"(line_size), "5"(h)
:"r5", "r4", "r7", "r6"
);
}
#define AVG8x8_UNALIGNED_S0(S0_N,S0_A) \
static void avg8x8_unaligned_##S0_N##_0 (u8 *block, const u8 *s0, const u8 *s1, \
int dest_size, int line_size, int h) \
{ \
__asm__ volatile( \
" %3 = %3-4\n\t" \
" %4 = %4-4\n\t" \
" r9 = %4-4\n\t" \
" r10 = %2-"#S0_N"\n\t" \
" 1: \n\t" \
" %5 -= 1 # r4 = [%1++=4]\n\t" \
" r5 = [%1++=%4]\n\t" \
" r6 = [r10++=4]\n\t" \
" r7 = [r10++=4]\n\t" \
" r8 = [r10++=r9]\n\t" \
" r6 = "#S0_A"(r7,r6)\n\t" \
" r7 = "#S0_A"(r8,r7)\n\t" \
" r4 = r4 +|+|+|+ r6 (uavg,rnd) \n\t" \
" r5 = r5 +|+|+|+ r7 (uavg,rnd) \n\t" \
" [%0++=4] = r4\n\t" \
" [%0++=%3] = r5\n\t" \
"if(%5!=0) goto 1b \n\t" \
:"=&r"(block),"=&r"(s1), "=&r"(s0), "=&r"(dest_size), "=&r"(line_size), "=&r"(h) \
:"0"(block),"1"(s1), "2"(s0), "3"(dest_size), "4"(line_size), "5"(h) \
:"r5", "r4", "r7", "r6", "r9","r8","r10" \
); \
} \
#define AVG8x8_UNALIGNED_S1(S0_N,S0_A) \
static void avg8x8_unaligned_0_##S0_N (u8 *block, const u8 *s0, const u8 *s1, \
int dest_size, int line_size, int h) \
{ \
__asm__ volatile( \
" %3 = %3-4\n\t" \
" %4 = %4-4\n\t" \
" r9 = %4-4\n\t" \
" r10 = %2-"#S0_N"\n\t" \
" 1: \n\t" \
" %5 -= 1 # r4 = [%1++=4]\n\t" \
" r5 = [%1++=%4]\n\t" \
" r6 = [r10++=4]\n\t" \
" r7 = [r10++=4]\n\t" \
" r8 = [r10++=r9]\n\t" \
" r6 = "#S0_A"(r7,r6)\n\t" \
" r7 = "#S0_A"(r8,r7)\n\t" \
" r4 = r4 +|+|+|+ r6 (uavg,rnd) \n\t" \
" r5 = r5 +|+|+|+ r7 (uavg,rnd) \n\t" \
" [%0++=4] = r4\n\t" \
" [%0++=%3] = r5\n\t" \
"if(%5!=0) goto 1b \n\t" \
:"=&r"(block),"=&r"(s0), "=&r"(s1), "=&r"(dest_size), "=&r"(line_size), "=&r"(h) \
:"0"(block),"1"(s0), "2"(s1), "3"(dest_size), "4"(line_size), "5"(h) \
:"r5", "r4", "r7", "r6", "r9","r8","r10" \
); \
} \
#define AVG8x8_UNALIGNED_BOTH(S0_N,S1_N,S0_A,S1_A) \
static void avg8x8_unaligned_##S0_N##_##S1_N (u8 *block, const u8 *s0, const u8 *s1, \
int dest_size, int line_size, int h) \
{ \
__asm__ volatile( \
" %3 = %3-4\n\t" \
" %4 = %4-8\n\t" \
" r9 = %1-"#S0_N"\n\t" \
" r10 = %2-"#S1_N"\n\t" \
" 1: \n\t" \
" %5 -= 1 \n\t" \
" r4 = [r9++=4]\n\t" \
" r5 = [r9++=4]\n\t" \
" r8 = [r9++=%4]\n\t" \
" r4 = "#S0_A"(r5,r4)\n\t" \
" r5 = "#S0_A"(r8,r5)\n\t" \
" r6 = [r10++=4]\n\t" \
" r7 = [r10++=4]\n\t" \
" r8 = [r10++=%4]\n\t" \
" r6 = "#S1_A"(r7,r6)\n\t" \
" r7 = "#S1_A"(r8,r7)\n\t" \
" r4 = r4 +|+|+|+ r6 (uavg,rnd) \n\t" \
" r5 = r5 +|+|+|+ r7 (uavg,rnd) \n\t" \
" [%0++=4] = r4\n\t" \
" [%0++=%3] = r5\n\t" \
"if(%5!=0) goto 1b \n\t" \
:"=&r"(block),"=&r"(s0), "=&r"(s1), "=&r"(dest_size), "=&r"(line_size), "=&r"(h) \
:"0"(block),"1"(s0), "2"(s1), "3"(dest_size), "4"(line_size), "5"(h) \
:"r5", "r4", "r7", "r6", "r9","r8","r10" \
); \
} \
AVG8x8_UNALIGNED_S0(1, align8)
AVG8x8_UNALIGNED_S0(2, align16)
AVG8x8_UNALIGNED_S0(3, align24)
AVG8x8_UNALIGNED_S1(1, align8)
AVG8x8_UNALIGNED_S1(2, align16)
AVG8x8_UNALIGNED_S1(3, align24)
AVG8x8_UNALIGNED_BOTH(1, 1, align8, align8)
AVG8x8_UNALIGNED_BOTH(1, 2, align8, align16)
AVG8x8_UNALIGNED_BOTH(1, 3, align8, align24)
AVG8x8_UNALIGNED_BOTH(2, 1, align16, align8)
AVG8x8_UNALIGNED_BOTH(2, 2, align16, align16)
AVG8x8_UNALIGNED_BOTH(2, 3, align16, align24)
AVG8x8_UNALIGNED_BOTH(3, 1, align24, align8)
AVG8x8_UNALIGNED_BOTH(3, 2, align24, align16)
AVG8x8_UNALIGNED_BOTH(3, 3, align24, align24)
void avg8x8(u8 *block, const u8 *s0, const u8 *s1,
int dest_size, int line_size, int h)
{
int s0_lower2bit = (int)s0 & 0x3;
int s1_lower2bit = (int)s1 & 0x3;
static void (*avg8x8_tab[4][4])(u8 * block, const u8 * s0, const u8 * s1,
int dest_size, int line_size, int h) = {
{avg8x8_aligned, avg8x8_unaligned_0_1, avg8x8_unaligned_0_2, avg8x8_unaligned_0_3},
{avg8x8_unaligned_1_0, avg8x8_unaligned_1_1, avg8x8_unaligned_1_2, avg8x8_unaligned_1_3},
{avg8x8_unaligned_2_0, avg8x8_unaligned_2_1, avg8x8_unaligned_2_2, avg8x8_unaligned_2_3},
{avg8x8_unaligned_3_0, avg8x8_unaligned_3_1, avg8x8_unaligned_3_2, avg8x8_unaligned_3_3},
};
avg8x8_tab[s0_lower2bit][s1_lower2bit](block, s0, s1, dest_size, line_size, h);
}
static u32 sad8x8_aligned(const u8 *s0, const u8 *s1,
int dest_size, int line_size, int h)
{
int out = 0;
__asm__ volatile(
" %3 = %3-4\n\t"
" %4 = %4-4\n\t"
" r8 = 0\n\t"
" r9 = 0\n\t"
" 1: \n\t"
" rep %5 {\n\t"
" r4 = [%1++=4]\n\t"
" r6 = [%2++=4]\n\t"
" r5 = [%1++=%3]\n\t"
" r7 = [%2++=%4]\n\t"
" r9_r8 += abs(r4 -|-|-|- r6) (usat) \n\t"
" r9_r8 += abs(r5 -|-|-|- r7) (usat) \n\t"
"}\n\t"
"r8 = r8.h,r8.l +|+ r9.h,r9.l (usat) \n\t"
"%0.l = r8.h + r8.l \n\t"
:"=&r"(out), "=&r"(s0), "=&r"(s1), "=&r"(dest_size), "=&r"(line_size), "=&r"(h)
:"0"(out), "1"(s0), "2"(s1), "3"(dest_size), "4"(line_size), "5"(h)
:"r5", "r4", "r7", "r6", "r9", "r8"
);
return out;
}
#define SAD8x8_UNALIGNED_S0(S0_N,S0_A) \
static u32 sad8x8_unaligned_##S0_N##_0 (const u8 *s0, const u8 *s1, \
int dest_size, int line_size, int h) \
{ \
int out = 0; \
__asm__ volatile( \
" %3 = %3-4\n\t" \
" %4 = %4-8\n\t" \
" r8 = 0\n\t" \
" r9 = 0\n\t" \
" r10 = %2-"#S0_N"\n\t" \
" 1: \n\t" \
" %5 -= 1 # r4 = [%1++=4]\n\t" \
" r5 = [%1++=%3]\n\t" \
" r6 = [r10++=4]\n\t" \
" r7 = [r10++=4]\n\t" \
" r12 = [r10++=%4]\n\t" \
" r6 = "#S0_A"(r7,r6)\n\t" \
" r7 = "#S0_A"(r12,r7)\n\t" \
" r9_r8 += abs(r4 -|-|-|- r6) (usat) \n\t" \
" r9_r8 += abs(r5 -|-|-|- r7) (usat) \n\t" \
"if(%5!=0) goto 1b \n\t" \
"r8 = r8.h,r8.l +|+ r9.h,r9.l (usat) \n\t" \
"%0.l = r8.h + r8.l \n\t" \
:"=&r"(out),"=&r"(s1), "=&r"(s0), "=&r"(line_size), "=&r"(dest_size), "=&r"(h) \
:"0"(out),"1"(s1), "2"(s0), "3"(line_size), "4"(dest_size), "5"(h) \
:"r5", "r4", "r7", "r6", "r9","r8","r10", "r12" \
); \
return out; \
}
#define SAD8x8_UNALIGNED_S1(S0_N,S0_A) \
static u32 sad8x8_unaligned_0_##S0_N (const u8 *s0, const u8 *s1, \
int dest_size, int line_size, int h) \
{ \
int out = 0; \
__asm__ volatile( \
" %3 = %3-4\n\t" \
" %4 = %4-8\n\t" \
" r8 = 0\n\t" \
" r9 = 0\n\t" \
" r10 = %2-"#S0_N"\n\t" \
" 1: \n\t" \
" %5 -= 1 # r4 = [%1++=4]\n\t" \
" r5 = [%1++=%3]\n\t" \
" r6 = [r10++=4]\n\t" \
" r7 = [r10++=4]\n\t" \
" r12 = [r10++=%4]\n\t" \
" r6 = "#S0_A"(r7,r6)\n\t" \
" r7 = "#S0_A"(r12,r7)\n\t" \
" r9_r8 += abs(r4 -|-|-|- r6) (usat) \n\t" \
" r9_r8 += abs(r5 -|-|-|- r7) (usat) \n\t" \
"if(%5!=0) goto 1b \n\t" \
"r8 = r8.h,r8.l +|+ r9.h,r9.l (usat) \n\t" \
"%0.l = r8.h + r8.l \n\t" \
:"=&r"(out),"=&r"(s0), "=&r"(s1), "=&r"(dest_size), "=&r"(line_size), "=&r"(h) \
:"0"(out),"1"(s0), "2"(s1), "3"(dest_size), "4"(line_size), "5"(h) \
:"r5", "r4", "r7", "r6", "r9","r8","r10","r12" \
); \
return out; \
}
#define SAD8x8_UNALIGNED_BOTH(S0_N,S1_N,S0_A,S1_A) \
static u32 sad8x8_unaligned_##S0_N##_##S1_N (const u8 *s0, const u8 *s1, \
int dest_size, int line_size, int h) \
{ \
int out = 0; \
__asm__ volatile( \
" %3 = %3-8\n\t" \
" %4 = %4-8\n\t" \
" r8 = 0\n\t" \
" r9 = 0\n\t" \
" r11 = %1-"#S0_N"\n\t" \
" r10 = %2-"#S1_N"\n\t" \
" 1: \n\t" \
" %5 -= 1 \n\t" \
" r4 = [r11++=4]\n\t" \
" r5 = [r11++=4]\n\t" \
" r12 = [r11++=%3]\n\t" \
" r4 = "#S0_A"(r5,r4)\n\t" \
" r5 = "#S0_A"(r12,r5)\n\t" \
" r6 = [r10++=4]\n\t" \
" r7 = [r10++=4]\n\t" \
" r12 = [r10++=%4]\n\t" \
" r6 = "#S1_A"(r7,r6)\n\t" \
" r7 = "#S1_A"(r12,r7)\n\t" \
" r9_r8 += abs(r4 -|-|-|- r6) (usat) \n\t" \
" r9_r8 += abs(r5 -|-|-|- r7) (usat) \n\t" \
"if(%5!=0) goto 1b \n\t" \
"r8 = r8.h,r8.l +|+ r9.h,r9.l (usat) \n\t" \
"%0.l = r8.h + r8.l \n\t" \
:"=&r"(out),"=&r"(s0), "=&r"(s1), "=&r"(dest_size), "=&r"(line_size), "=&r"(h) \
:"0"(out),"1"(s0), "2"(s1), "3"(dest_size), "4"(line_size), "5"(h) \
:"r5", "r4", "r7", "r6", "r9","r8","r10","r11","r12" \
); \
return out; \
}
SAD8x8_UNALIGNED_S0(1, align8)
SAD8x8_UNALIGNED_S0(2, align16)
SAD8x8_UNALIGNED_S0(3, align24)
SAD8x8_UNALIGNED_S1(1, align8)
SAD8x8_UNALIGNED_S1(2, align16)
SAD8x8_UNALIGNED_S1(3, align24)
SAD8x8_UNALIGNED_BOTH(1, 1, align8, align8)
SAD8x8_UNALIGNED_BOTH(1, 2, align8, align16)
SAD8x8_UNALIGNED_BOTH(1, 3, align8, align24)
SAD8x8_UNALIGNED_BOTH(2, 1, align16, align8)
SAD8x8_UNALIGNED_BOTH(2, 2, align16, align16)
SAD8x8_UNALIGNED_BOTH(2, 3, align16, align24)
SAD8x8_UNALIGNED_BOTH(3, 1, align24, align8)
SAD8x8_UNALIGNED_BOTH(3, 2, align24, align16)
SAD8x8_UNALIGNED_BOTH(3, 3, align24, align24)
u32 sad8x8(const u8 *s0, const u8 *s1,
int dest_size, int line_size, int h)
{
int s0_lower2bit = (int)s0 & 0x3;
int s1_lower2bit = (int)s1 & 0x3;
static u32(*sad8x8_tab[4][4])(const u8 * s0, const u8 * s1,
int dest_size, int line_size, int h) = {
{sad8x8_aligned, sad8x8_unaligned_0_1, sad8x8_unaligned_0_2, sad8x8_unaligned_0_3},
{sad8x8_unaligned_1_0, sad8x8_unaligned_1_1, sad8x8_unaligned_1_2, sad8x8_unaligned_1_3},
{sad8x8_unaligned_2_0, sad8x8_unaligned_2_1, sad8x8_unaligned_2_2, sad8x8_unaligned_2_3},
{sad8x8_unaligned_3_0, sad8x8_unaligned_3_1, sad8x8_unaligned_3_2, sad8x8_unaligned_3_3},
};
return sad8x8_tab[s0_lower2bit][s1_lower2bit](s0, s1, dest_size, line_size, h);
}
Reference in New Issue View Git Blame Copy Permalink