| 1 |
/* { dg-options "-O2 -std=gnu17" } */ |
| 2 |
|
| 3 |
#pragma GCC aarch64 "arm_neon.h" |
| 4 |
typedef __Uint8x8_t uint8x8_t; |
| 5 |
typedef __Uint16x4_t uint16x4_t; |
| 6 |
typedef __Int16x8_t int16x8_t; |
| 7 |
typedef __Uint16x8_t uint16x8_t; |
| 8 |
int jsimd_extbgrx_ycc_convert_neon_image_width, |
| 9 |
jsimd_extbgrx_ycc_convert_neon___trans_tmp_1; |
| 10 |
uint16x4_t jsimd_extbgrx_ycc_convert_neon___trans_tmp_2; |
| 11 |
uint16x8_t vcombine_u16(); |
| 12 |
uint16x8_t vmovl_u8(uint8x8_t __a) { |
| 13 |
return __builtin_aarch64_uxtlv8hi_uu(__a); |
| 14 |
} |
| 15 |
__inline int __attribute__((__gnu_inline__)) vmull_laneq_u16(); |
| 16 |
uint8x8x4_t vld4_u8(); |
| 17 |
void jsimd_extbgrx_ycc_convert_neon() { |
| 18 |
int scaled_128_5 = jsimd_extbgrx_ycc_convert_neon___trans_tmp_1, |
| 19 |
cols_remaining = jsimd_extbgrx_ycc_convert_neon_image_width; |
| 20 |
for (;;) |
| 21 |
if (cols_remaining) { |
| 22 |
uint8x8x4_t input_pixels = vld4_u8(); |
| 23 |
uint16x8_t r = vmovl_u8(input_pixels.val[2]); |
| 24 |
uint16x8_t g = vmovl_u8(input_pixels.val[1]); |
| 25 |
uint16x8_t b = vmovl_u8(input_pixels.val[0]); |
| 26 |
int y_l = vmull_laneq_u16(r); |
| 27 |
uint16x8_t __a = g; |
| 28 |
jsimd_extbgrx_ycc_convert_neon___trans_tmp_2 = |
| 29 |
(uint16x4_t)vget_low_s16((int16x8_t)__a); |
| 30 |
__a = b; |
| 31 |
int cb_l = scaled_128_5; |
| 32 |
int cb_h = scaled_128_5; |
| 33 |
int cr_l = scaled_128_5; |
| 34 |
int cr_h = scaled_128_5; |
| 35 |
uint16x8_t y_u16 = vcombine_u16(y_l); |
| 36 |
uint16x8_t cb_u16 = vcombine_u16(cb_l, cb_h); |
| 37 |
uint16x8_t cr_u16 = vcombine_u16(cr_l, cr_h); |
| 38 |
__a = y_u16 = cb_u16 = cr_u16; |
| 39 |
} |
| 40 |
} |