diff --git a/internal/cgen/base/pixconv-submodule.c b/internal/cgen/base/pixconv-submodule.c index ef7c0fa5..0af33930 100644 --- a/internal/cgen/base/pixconv-submodule.c +++ b/internal/cgen/base/pixconv-submodule.c @@ -553,7 +553,7 @@ wuffs_base__pixel_swizzler__swap_rgbx_bgrx__sse128(uint8_t* dst_ptr, while (n >= 4) { __m128i x; - x = _mm_loadu_si128((const __m128i*)(const void*)s); + x = _mm_lddqu_si128((const __m128i*)(const void*)s); x = _mm_shuffle_epi8(x, shuffle); _mm_storeu_si128((__m128i*)(void*)d, x); @@ -1768,7 +1768,7 @@ wuffs_base__pixel_swizzler__bgrw__rgb__sse128(uint8_t* dst_ptr, while (n >= 6) { __m128i x; - x = _mm_loadu_si128((const __m128i*)(const void*)s); + x = _mm_lddqu_si128((const __m128i*)(const void*)s); x = _mm_shuffle_epi8(x, shuffle); x = _mm_or_si128(x, or_ff); _mm_storeu_si128((__m128i*)(void*)d, x); diff --git a/internal/cgen/data/data.go b/internal/cgen/data/data.go index 13d09f01..0c872ab4 100644 --- a/internal/cgen/data/data.go +++ b/internal/cgen/data/data.go @@ -558,7 +558,7 @@ const BasePixConvSubmoduleC = "" + "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__squash_align4_bgr_565_888(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t len = (dst_len < src_len ? dst_len : src_len) / 4;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n\n size_t n = len;\n while (n--) {\n uint32_t argb = wuffs_base__peek_u32le__no_bounds_check(s);\n uint32_t b5 = 0x1F & (argb >> (8 - 5));\n uint32_t g6 = 0x3F & (argb >> (16 - 6));\n uint32_t r5 = 0x1F & (argb >> (24 - 5));\n uint32_t alpha = argb & 0xFF000000;\n wuffs_base__poke_u32le__no_bounds_check(\n d, alpha | (r5 << 11) | (g6 << 5) | (b5 << 0));\n s += 4;\n d += 4;\n }\n return len;\n}\n\nstatic u" + "int64_t //\nwuffs_base__pixel_swizzler__swap_rgb_bgr(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t len = (dst_len < src_len ? dst_len : src_len) / 3;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n\n size_t n = len;\n while (n--) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n d[0] = b2;\n d[1] = b1;\n d[2] = b0;\n s += 3;\n d += 3;\n }\n return len;\n}\n\n#if defined(WUFFS_BASE__CPU_ARCH__X86_64)\n#if defined(__GNUC__)\n__attribute__((target(\"sse4.2\")))\n#endif\nstatic uint64_t //\nwuffs_base__pixel_swizzler__swap_rgbx_bgrx__sse128(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette" + - "_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t len = (dst_len < src_len ? dst_len : src_len) / 4;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n __m128i shuffle = _mm_set_epi8(+0x0F, +0x0C, +0x0D, +0x0E, //\n +0x0B, +0x08, +0x09, +0x0A, //\n +0x07, +0x04, +0x05, +0x06, //\n +0x03, +0x00, +0x01, +0x02);\n\n while (n >= 4) {\n __m128i x;\n x = _mm_loadu_si128((const __m128i*)(const void*)s);\n x = _mm_shuffle_epi8(x, shuffle);\n _mm_storeu_si128((__m128i*)(void*)d, x);\n\n s += 4 * 4;\n d += 4 * 4;\n n -= 4;\n }\n\n while (n--) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n uint8_t b3 = s[3];\n d[0] = b2;\n d[1] = b1;\n d[2] = b0;\n d[3] = b3;\n s += 4;\n " + + "_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t len = (dst_len < src_len ? dst_len : src_len) / 4;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n __m128i shuffle = _mm_set_epi8(+0x0F, +0x0C, +0x0D, +0x0E, //\n +0x0B, +0x08, +0x09, +0x0A, //\n +0x07, +0x04, +0x05, +0x06, //\n +0x03, +0x00, +0x01, +0x02);\n\n while (n >= 4) {\n __m128i x;\n x = _mm_lddqu_si128((const __m128i*)(const void*)s);\n x = _mm_shuffle_epi8(x, shuffle);\n _mm_storeu_si128((__m128i*)(void*)d, x);\n\n s += 4 * 4;\n d += 4 * 4;\n n -= 4;\n }\n\n while (n--) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n uint8_t b3 = s[3];\n d[0] = b2;\n d[1] = b1;\n d[2] = b0;\n d[3] = b3;\n s += 4;\n " + "d += 4;\n }\n return len;\n}\n#endif // defined(WUFFS_BASE__CPU_ARCH__X86_64)\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__swap_rgbx_bgrx(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t len = (dst_len < src_len ? dst_len : src_len) / 4;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n\n size_t n = len;\n while (n--) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n uint8_t b3 = s[3];\n d[0] = b2;\n d[1] = b1;\n d[2] = b0;\n d[3] = b3;\n s += 4;\n d += 4;\n }\n return len;\n}\n\n" + "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__squash_tight_4x8_4x16le(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t src_len8 = src_len / 8;\n size_t len = (dst_len4 < src_len8) ? dst_len4 : src_len8;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n\n size_t n = len;\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), wuffs_base__color_u64__as__color_u32(\n wuffs_base__peek_u64le__no_bounds_check(s + (0 * 8))));\n\n s += 1 * 8;\n d += 1 * 4;\n n -= 1;\n }\n return len;\n}\n\n" + @@ -601,7 +601,7 @@ const BasePixConvSubmoduleC = "" + "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgrw__bgr(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t src_len3 = src_len / 3;\n size_t len = (dst_len4 < src_len3) ? dst_len4 : src_len3;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n // TODO: unroll.\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4),\n 0xFF000000 | wuffs_base__peek_u24le__no_bounds_check(s + (0 * 3)));\n\n s += 1 * 3;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgrw__bgrx(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palett" + "e_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t src_len4 = src_len / 4;\n size_t len = (dst_len4 < src_len4) ? dst_len4 : src_len4;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n // TODO: unroll.\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4),\n 0xFF000000 | wuffs_base__peek_u32le__no_bounds_check(s + (0 * 4)));\n\n s += 1 * 4;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\n#if defined(WUFFS_BASE__CPU_ARCH__X86_64)\n#if defined(__GNUC__)\n__attribute__((target(\"sse4.2\")))\n#endif\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgrw__rgb__sse128(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette" + - "_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t src_len3 = src_len / 3;\n size_t len = (dst_len4 < src_len3) ? dst_len4 : src_len3;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n __m128i shuffle = _mm_set_epi8(+0x00, +0x09, +0x0A, +0x0B, //\n +0x00, +0x06, +0x07, +0x08, //\n +0x00, +0x03, +0x04, +0x05, //\n +0x00, +0x00, +0x01, +0x02);\n __m128i or_ff = _mm_set_epi8(-0x01, +0x00, +0x00, +0x00, //\n -0x01, +0x00, +0x00, +0x00, //\n -0x01, +0x00, +0x00, +0x00, //\n -0x01, +0x00, +0x00, +0x00);\n\n while (n >= 6) {\n __m128i x;\n x = _mm_loadu_si128((const __m128i*)(const void*)s);\n x = _mm_shuffle_epi8(x, shuffle);\n x = _mm_or_si128(x, or_ff);\n _mm_storeu_si" + + "_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t src_len3 = src_len / 3;\n size_t len = (dst_len4 < src_len3) ? dst_len4 : src_len3;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n __m128i shuffle = _mm_set_epi8(+0x00, +0x09, +0x0A, +0x0B, //\n +0x00, +0x06, +0x07, +0x08, //\n +0x00, +0x03, +0x04, +0x05, //\n +0x00, +0x00, +0x01, +0x02);\n __m128i or_ff = _mm_set_epi8(-0x01, +0x00, +0x00, +0x00, //\n -0x01, +0x00, +0x00, +0x00, //\n -0x01, +0x00, +0x00, +0x00, //\n -0x01, +0x00, +0x00, +0x00);\n\n while (n >= 6) {\n __m128i x;\n x = _mm_lddqu_si128((const __m128i*)(const void*)s);\n x = _mm_shuffle_epi8(x, shuffle);\n x = _mm_or_si128(x, or_ff);\n _mm_storeu_si" + "128((__m128i*)(void*)d, x);\n\n s += 4 * 3;\n d += 4 * 4;\n n -= 4;\n }\n\n while (n >= 1) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n d[0] = b2;\n d[1] = b1;\n d[2] = b0;\n d[3] = 0xFF;\n\n s += 1 * 3;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n#endif // defined(WUFFS_BASE__CPU_ARCH__X86_64)\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgrw__rgb(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const uint8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t src_len3 = src_len / 3;\n size_t len = (dst_len4 < src_len3) ? dst_len4 : src_len3;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n while (n >= 1) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n d[0] = b2;\n " + "d[1] = b1;\n d[2] = b0;\n d[3] = 0xFF;\n\n s += 1 * 3;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\n" + "" + diff --git a/release/c/wuffs-unsupported-snapshot.c b/release/c/wuffs-unsupported-snapshot.c index 482178c7..606fa8fb 100644 --- a/release/c/wuffs-unsupported-snapshot.c +++ b/release/c/wuffs-unsupported-snapshot.c @@ -14658,7 +14658,7 @@ wuffs_base__pixel_swizzler__swap_rgbx_bgrx__sse128(uint8_t* dst_ptr, while (n >= 4) { __m128i x; - x = _mm_loadu_si128((const __m128i*)(const void*)s); + x = _mm_lddqu_si128((const __m128i*)(const void*)s); x = _mm_shuffle_epi8(x, shuffle); _mm_storeu_si128((__m128i*)(void*)d, x); @@ -15873,7 +15873,7 @@ wuffs_base__pixel_swizzler__bgrw__rgb__sse128(uint8_t* dst_ptr, while (n >= 6) { __m128i x; - x = _mm_loadu_si128((const __m128i*)(const void*)s); + x = _mm_lddqu_si128((const __m128i*)(const void*)s); x = _mm_shuffle_epi8(x, shuffle); x = _mm_or_si128(x, or_ff); _mm_storeu_si128((__m128i*)(void*)d, x);