mirrors/SDL
1
0
Fork 0
mirror of https://github.com/libsdl-org/SDL.git synced 2026-05-25 06:18:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Replace custom NEON cast macros with vreinterpret_*_*

Browse Source
This commit is contained in:
Cameron Cawley
2026-05-14 14:50:21 +01:00
committed by Sam Lantinga
parent 006959ca87
commit a3376acc2e
1 changed files with 17 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
src/video/SDL_stretch.c
View File

@@ -375,20 +375,6 @@ static bool scale_mat(const Uint32 *src, int src_w, int src_h, int src_pitch, Ui
return true;
}
#ifdef SDL_NEON_INTRINSICS
#define CAST_uint8x8_t (uint8x8_t)
#define CAST_uint32x2_t (uint32x2_t)
#endif
#if defined(_MSC_VER)
#ifdef SDL_NEON_INTRINSICS
#undef CAST_uint8x8_t
#undef CAST_uint32x2_t
#define CAST_uint8x8_t
#define CAST_uint32x2_t
#endif
#endif
#ifdef SDL_SSE2_INTRINSICS
#if 0
@@ -587,8 +573,8 @@ static SDL_INLINE void INTERPOL_BILINEAR_NEON(const Uint32 *s0, const Uint32 *s1
uint16x8_t d0;
uint8x8_t e0;
x_00_01 = CAST_uint8x8_t vld1_u32(s0); // Load 2 pixels
x_10_11 = CAST_uint8x8_t vld1_u32(s1);
x_00_01 = vreinterpret_u8_u32(vld1_u32(s0)); // Load 2 pixels
x_10_11 = vreinterpret_u8_u32(vld1_u32(s1));
/* Interpolated == x0 + frac * (x1 - x0) == x0 * (1 - frac) + x1 * frac */
k0 = vmull_u8(x_00_01, v_frac_h1); /* k0 := x0 * (1 - frac) */
@@ -608,7 +594,7 @@ static SDL_INLINE void INTERPOL_BILINEAR_NEON(const Uint32 *s0, const Uint32 *s1
e0 = vmovn_u16(d0);
// Store 1 pixel
*dst = vget_lane_u32(CAST_uint32x2_t e0, 0);
*dst = vget_lane_u32(vreinterpret_u32_u8(e0), 0);
}
static bool scale_mat_NEON(const Uint32 *src, int src_w, int src_h, int src_pitch, Uint32 *dst, int dst_w, int dst_h, int dst_pitch)
@@ -672,14 +658,14 @@ static bool scale_mat_NEON(const Uint32 *src, int src_w, int src_h, int src_pitc
s_16_17 = (const Uint32 *)((const Uint8 *)src_h1 + index_w_3);
// Interpolation vertical
x_00_01 = CAST_uint8x8_t vld1_u32(s_00_01); // Load 2 pixels
x_02_03 = CAST_uint8x8_t vld1_u32(s_02_03);
x_04_05 = CAST_uint8x8_t vld1_u32(s_04_05);
x_06_07 = CAST_uint8x8_t vld1_u32(s_06_07);
x_10_11 = CAST_uint8x8_t vld1_u32(s_10_11);
x_12_13 = CAST_uint8x8_t vld1_u32(s_12_13);
x_14_15 = CAST_uint8x8_t vld1_u32(s_14_15);
x_16_17 = CAST_uint8x8_t vld1_u32(s_16_17);
x_00_01 = vreinterpret_u8_u32(vld1_u32(s_00_01)); // Load 2 pixels
x_02_03 = vreinterpret_u8_u32(vld1_u32(s_02_03));
x_04_05 = vreinterpret_u8_u32(vld1_u32(s_04_05));
x_06_07 = vreinterpret_u8_u32(vld1_u32(s_06_07));
x_10_11 = vreinterpret_u8_u32(vld1_u32(s_10_11));
x_12_13 = vreinterpret_u8_u32(vld1_u32(s_12_13));
x_14_15 = vreinterpret_u8_u32(vld1_u32(s_14_15));
x_16_17 = vreinterpret_u8_u32(vld1_u32(s_16_17));
/* Interpolated == x0 + frac * (x1 - x0) == x0 * (1 - frac) + x1 * frac */
k0 = vmull_u8(x_00_01, v_frac_h1); /* k0 := x0 * (1 - frac) */
@@ -729,7 +715,7 @@ static bool scale_mat_NEON(const Uint32 *src, int src_w, int src_h, int src_pitc
// Narrow again
e1 = vmovn_u16(d1);
f0 = vcombine_u32(CAST_uint32x2_t e0, CAST_uint32x2_t e1);
f0 = vcombine_u32(vreinterpret_u32_u8(e0), vreinterpret_u32_u8(e1));
// Store 4 pixels
vst1q_u32(dst, f0);
@@ -768,10 +754,10 @@ static bool scale_mat_NEON(const Uint32 *src, int src_w, int src_h, int src_pitc
s_12_13 = (const Uint32 *)((const Uint8 *)src_h1 + index_w_1);
// Interpolation vertical
x_00_01 = CAST_uint8x8_t vld1_u32(s_00_01); // Load 2 pixels
x_02_03 = CAST_uint8x8_t vld1_u32(s_02_03);
x_10_11 = CAST_uint8x8_t vld1_u32(s_10_11);
x_12_13 = CAST_uint8x8_t vld1_u32(s_12_13);
x_00_01 = vreinterpret_u8_u32(vld1_u32(s_00_01)); // Load 2 pixels
x_02_03 = vreinterpret_u8_u32(vld1_u32(s_02_03));
x_10_11 = vreinterpret_u8_u32(vld1_u32(s_10_11));
x_12_13 = vreinterpret_u8_u32(vld1_u32(s_12_13));
/* Interpolated == x0 + frac * (x1 - x0) == x0 * (1 - frac) + x1 * frac */
k0 = vmull_u8(x_00_01, v_frac_h1); /* k0 := x0 * (1 - frac) */
@@ -801,7 +787,7 @@ static bool scale_mat_NEON(const Uint32 *src, int src_w, int src_h, int src_pitc
e0 = vmovn_u16(d0);
// Store 2 pixels
vst1_u32(dst, CAST_uint32x2_t e0);
vst1_u32(dst, vreinterpret_u32_u8(e0));
dst += 2;
}