diff --git a/internal/cgen/base/fundamental-public.h b/internal/cgen/base/fundamental-public.h index 990a7541..6952a770 100644 --- a/internal/cgen/base/fundamental-public.h +++ b/internal/cgen/base/fundamental-public.h @@ -44,10 +44,19 @@ // Define WUFFS_CONFIG__AVOID_CPU_ARCH to avoid any code tied to a specific CPU // architecture, such as SSE SIMD for the x86 CPU family. -#if defined(WUFFS_CONFIG__AVOID_CPU_ARCH) +#if defined(WUFFS_CONFIG__AVOID_CPU_ARCH) // (#if-chain ref AVOID_CPU_ARCH_0) // No-op. +#else // (#if-chain ref AVOID_CPU_ARCH_0) + +// The "defined(__clang__)" isn't redundant. While vanilla clang defines +// __GNUC__, clang-cl (which mimics MSVC's cl.exe) does not. +#if defined(__GNUC__) || defined(__clang__) +#define WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(arg) __attribute__((target(arg))) #else -#if defined(__GNUC__) +#define WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(arg) +#endif // defined(__GNUC__) || defined(__clang__) + +#if defined(__GNUC__) // (#if-chain ref AVOID_CPU_ARCH_1) // To simplify Wuffs code, "cpu_arch >= arm_xxx" requires xxx but also // unaligned little-endian load/stores. @@ -73,15 +82,24 @@ #define WUFFS_BASE__CPU_ARCH__X86_64 #endif // defined(__x86_64__) -#elif defined(_MSC_VER) // defined(__GNUC__) +#elif defined(_MSC_VER) // (#if-chain ref AVOID_CPU_ARCH_1) #if defined(_M_X64) +#if defined(__clang__) +// No-op. clang-cl (which defines both __clang__ and _MSC_VER) supports +// "__attribute__((target(arg)))". +#elif !defined(__AVX__) +// For MSVC's cl.exe (unlike clang or gcc), SIMD capability is a compile-time +// property of the source file (e.g. a /arch:AVX or -mavx compiler flag), not +// of individual functions (that can be conditionally selected at runtime). +#error "Wuffs with MSVC+X64 needs /arch:AVX or /DWUFFS_CONFIG__AVOID_CPU_ARCH" +#endif // defined(__clang__); !defined(__AVX__) #include #define WUFFS_BASE__CPU_ARCH__X86_64 -#endif // defined(__x86_64__) +#endif // defined(_M_X64) -#endif // defined(__GNUC__); defined(_MSC_VER) -#endif // defined(WUFFS_CONFIG__AVOID_CPU_ARCH) +#endif // (#if-chain ref AVOID_CPU_ARCH_1) +#endif // (#if-chain ref AVOID_CPU_ARCH_0) // -------- @@ -123,6 +141,8 @@ wuffs_base__cpu_arch__have_x86_sse42() { // - bit_POPCNT = (1 << 23) // - bit_SSE4_2 = (1 << 20) const unsigned int sse42_ecx1 = 0x00900002; + + // clang defines __GNUC__ and clang-cl defines _MSC_VER (but not __GNUC__). #if defined(__GNUC__) unsigned int eax1 = 0; unsigned int ebx1 = 0; diff --git a/internal/cgen/base/pixconv-submodule.c b/internal/cgen/base/pixconv-submodule.c index 30d1792c..9a3e7a81 100644 --- a/internal/cgen/base/pixconv-submodule.c +++ b/internal/cgen/base/pixconv-submodule.c @@ -757,9 +757,7 @@ wuffs_base__pixel_swizzler__swap_rgb_bgr(uint8_t* dst_ptr, } #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("sse4.2") static uint64_t // wuffs_base__pixel_swizzler__swap_rgbx_bgrx__sse42(uint8_t* dst_ptr, size_t dst_len, @@ -2102,9 +2100,7 @@ wuffs_base__pixel_swizzler__bgrw__bgrx(uint8_t* dst_ptr, } #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("sse4.2") static uint64_t // wuffs_base__pixel_swizzler__bgrw__rgb__sse42(uint8_t* dst_ptr, size_t dst_len, @@ -2545,9 +2541,7 @@ wuffs_base__pixel_swizzler__xxxx__index_binary_alpha__src_over( } #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("sse4.2") static uint64_t // wuffs_base__pixel_swizzler__xxxx__y__sse42(uint8_t* dst_ptr, size_t dst_len, diff --git a/internal/cgen/data/data.go b/internal/cgen/data/data.go index 1143893e..ee6cfaf1 100644 --- a/internal/cgen/data/data.go +++ b/internal/cgen/data/data.go @@ -54,13 +54,14 @@ const BaseFundamentalPublicH = "" + "// ---------------- Version\n\n// WUFFS_VERSION is the major.minor.patch version, as per https://semver.org/,\n// as a uint64_t. The major number is the high 32 bits. The minor number is the\n// middle 16 bits. The patch number is the low 16 bits. The pre-release label\n// and build metadata are part of the string representation (such as\n// \"1.2.3-beta+456.20181231\") but not the uint64_t representation.\n//\n// WUFFS_VERSION_PRE_RELEASE_LABEL (such as \"\", \"beta\" or \"rc.1\") being\n// non-empty denotes a developer preview, not a release version, and has no\n// backwards or forwards compatibility guarantees.\n//\n// WUFFS_VERSION_BUILD_METADATA_XXX, if non-zero, are the number of commits and\n// the last commit date in the repository used to build this library. Within\n// each major.minor branch, the commit count should increase monotonically.\n//\n// !! Some code generation programs can override WUFFS_VERSION.\n#define WUFFS_VERSION 0\n#define WUFFS_VERSION_MAJOR 0\n#define WUFFS_VERSION_MINOR 0\n#define WUFFS_VERSION_PATCH 0\n#de" + "fine WUFFS_VERSION_PRE_RELEASE_LABEL \"work.in.progress\"\n#define WUFFS_VERSION_BUILD_METADATA_COMMIT_COUNT 0\n#define WUFFS_VERSION_BUILD_METADATA_COMMIT_DATE 0\n#define WUFFS_VERSION_STRING \"0.0.0+0.00000000\"\n\n" + "" + - "// ---------------- Configuration\n\n// Define WUFFS_CONFIG__AVOID_CPU_ARCH to avoid any code tied to a specific CPU\n// architecture, such as SSE SIMD for the x86 CPU family.\n#if defined(WUFFS_CONFIG__AVOID_CPU_ARCH)\n// No-op.\n#else\n#if defined(__GNUC__)\n\n// To simplify Wuffs code, \"cpu_arch >= arm_xxx\" requires xxx but also\n// unaligned little-endian load/stores.\n#if defined(__ARM_FEATURE_UNALIGNED) && defined(__BYTE_ORDER__) && \\\n (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)\n// Not all gcc versions define __ARM_ACLE, even if they support crc32\n// intrinsics. Look for __ARM_FEATURE_CRC32 instead.\n#if defined(__ARM_FEATURE_CRC32)\n#include \n#define WUFFS_BASE__CPU_ARCH__ARM_CRC32\n#endif // defined(__ARM_FEATURE_CRC32)\n#if defined(__ARM_NEON)\n#include \n#define WUFFS_BASE__CPU_ARCH__ARM_NEON\n#endif // defined(__ARM_NEON)\n#endif // defined(__ARM_FEATURE_UNALIGNED) etc\n\n// Similarly, \"cpu_arch >= x86_sse42\" requires SSE4.2 but also PCLMUL and\n// POPCNT. This is checked at runtime via cpu" + - "id, not at compile time.\n#if defined(__x86_64__)\n#include \n#include \n#define WUFFS_BASE__CPU_ARCH__X86_64\n#endif // defined(__x86_64__)\n\n#elif defined(_MSC_VER) // defined(__GNUC__)\n\n#if defined(_M_X64)\n#include \n#define WUFFS_BASE__CPU_ARCH__X86_64\n#endif // defined(__x86_64__)\n\n#endif // defined(__GNUC__); defined(_MSC_VER)\n#endif // defined(WUFFS_CONFIG__AVOID_CPU_ARCH)\n\n" + + "// ---------------- Configuration\n\n// Define WUFFS_CONFIG__AVOID_CPU_ARCH to avoid any code tied to a specific CPU\n// architecture, such as SSE SIMD for the x86 CPU family.\n#if defined(WUFFS_CONFIG__AVOID_CPU_ARCH) // (#if-chain ref AVOID_CPU_ARCH_0)\n// No-op.\n#else // (#if-chain ref AVOID_CPU_ARCH_0)\n\n// The \"defined(__clang__)\" isn't redundant. While vanilla clang defines\n// __GNUC__, clang-cl (which mimics MSVC's cl.exe) does not.\n#if defined(__GNUC__) || defined(__clang__)\n#define WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(arg) __attribute__((target(arg)))\n#else\n#define WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(arg)\n#endif // defined(__GNUC__) || defined(__clang__)\n\n#if defined(__GNUC__) // (#if-chain ref AVOID_CPU_ARCH_1)\n\n// To simplify Wuffs code, \"cpu_arch >= arm_xxx\" requires xxx but also\n// unaligned little-endian load/stores.\n#if defined(__ARM_FEATURE_UNALIGNED) && defined(__BYTE_ORDER__) && \\\n (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)\n// Not all gcc versions define __ARM_ACLE, even if they support crc32" + + "\n// intrinsics. Look for __ARM_FEATURE_CRC32 instead.\n#if defined(__ARM_FEATURE_CRC32)\n#include \n#define WUFFS_BASE__CPU_ARCH__ARM_CRC32\n#endif // defined(__ARM_FEATURE_CRC32)\n#if defined(__ARM_NEON)\n#include \n#define WUFFS_BASE__CPU_ARCH__ARM_NEON\n#endif // defined(__ARM_NEON)\n#endif // defined(__ARM_FEATURE_UNALIGNED) etc\n\n// Similarly, \"cpu_arch >= x86_sse42\" requires SSE4.2 but also PCLMUL and\n// POPCNT. This is checked at runtime via cpuid, not at compile time.\n#if defined(__x86_64__)\n#include \n#include \n#define WUFFS_BASE__CPU_ARCH__X86_64\n#endif // defined(__x86_64__)\n\n#elif defined(_MSC_VER) // (#if-chain ref AVOID_CPU_ARCH_1)\n\n#if defined(_M_X64)\n#if defined(__clang__)\n// No-op. clang-cl (which defines both __clang__ and _MSC_VER) supports\n// \"__attribute__((target(arg)))\".\n#elif !defined(__AVX__)\n// For MSVC's cl.exe (unlike clang or gcc), SIMD capability is a compile-time\n// property of the source file (e.g. a /arch:AVX or -mavx compiler flag), not" + + "\n// of individual functions (that can be conditionally selected at runtime).\n#error \"Wuffs with MSVC+X64 needs /arch:AVX or /DWUFFS_CONFIG__AVOID_CPU_ARCH\"\n#endif // defined(__clang__); !defined(__AVX__)\n#include \n#define WUFFS_BASE__CPU_ARCH__X86_64\n#endif // defined(_M_X64)\n\n#endif // (#if-chain ref AVOID_CPU_ARCH_1)\n#endif // (#if-chain ref AVOID_CPU_ARCH_0)\n\n" + "" + "// --------\n\n// Define WUFFS_CONFIG__STATIC_FUNCTIONS to make all of Wuffs' functions have\n// static storage. The motivation is discussed in the \"ALLOW STATIC\n// IMPLEMENTATION\" section of\n// https://raw.githubusercontent.com/nothings/stb/master/docs/stb_howto.txt\n#if defined(WUFFS_CONFIG__STATIC_FUNCTIONS)\n#define WUFFS_BASE__MAYBE_STATIC static\n#else\n#define WUFFS_BASE__MAYBE_STATIC\n#endif // defined(WUFFS_CONFIG__STATIC_FUNCTIONS)\n\n" + "" + - "// ---------------- CPU Architecture\n\nstatic inline bool //\nwuffs_base__cpu_arch__have_arm_crc32() {\n#if defined(WUFFS_BASE__CPU_ARCH__ARM_CRC32)\n return true;\n#else\n return false;\n#endif // defined(WUFFS_BASE__CPU_ARCH__ARM_CRC32)\n}\n\nstatic inline bool //\nwuffs_base__cpu_arch__have_arm_neon() {\n#if defined(WUFFS_BASE__CPU_ARCH__ARM_NEON)\n return true;\n#else\n return false;\n#endif // defined(WUFFS_BASE__CPU_ARCH__ARM_NEON)\n}\n\nstatic inline bool //\nwuffs_base__cpu_arch__have_x86_sse42() {\n#if defined(WUFFS_BASE__CPU_ARCH__X86_64)\n // GCC defines these macros but MSVC does not.\n // - bit_PCLMUL = (1 << 1)\n // - bit_POPCNT = (1 << 23)\n // - bit_SSE4_2 = (1 << 20)\n const unsigned int sse42_ecx1 = 0x00900002;\n#if defined(__GNUC__)\n unsigned int eax1 = 0;\n unsigned int ebx1 = 0;\n unsigned int ecx1 = 0;\n unsigned int edx1 = 0;\n if (__get_cpuid(1, &eax1, &ebx1, &ecx1, &edx1)) {\n return (ecx1 & sse42_ecx1) == sse42_ecx1;\n }\n#elif defined(_MSC_VER) // defined(__GNUC__)\n int x[4];\n __cpuid(x" + - ", 1);\n return (((unsigned int)(x[2])) & sse42_ecx1) == sse42_ecx1;\n#else\n#error \"WUFFS_BASE__CPU_ARCH__ETC combined with an unsupported compiler\"\n#endif // defined(__GNUC__); defined(_MSC_VER)\n#endif // defined(WUFFS_BASE__CPU_ARCH__X86_64)\n return false;\n}\n\n" + + "// ---------------- CPU Architecture\n\nstatic inline bool //\nwuffs_base__cpu_arch__have_arm_crc32() {\n#if defined(WUFFS_BASE__CPU_ARCH__ARM_CRC32)\n return true;\n#else\n return false;\n#endif // defined(WUFFS_BASE__CPU_ARCH__ARM_CRC32)\n}\n\nstatic inline bool //\nwuffs_base__cpu_arch__have_arm_neon() {\n#if defined(WUFFS_BASE__CPU_ARCH__ARM_NEON)\n return true;\n#else\n return false;\n#endif // defined(WUFFS_BASE__CPU_ARCH__ARM_NEON)\n}\n\nstatic inline bool //\nwuffs_base__cpu_arch__have_x86_sse42() {\n#if defined(WUFFS_BASE__CPU_ARCH__X86_64)\n // GCC defines these macros but MSVC does not.\n // - bit_PCLMUL = (1 << 1)\n // - bit_POPCNT = (1 << 23)\n // - bit_SSE4_2 = (1 << 20)\n const unsigned int sse42_ecx1 = 0x00900002;\n\n // clang defines __GNUC__ and clang-cl defines _MSC_VER (but not __GNUC__).\n#if defined(__GNUC__)\n unsigned int eax1 = 0;\n unsigned int ebx1 = 0;\n unsigned int ecx1 = 0;\n unsigned int edx1 = 0;\n if (__get_cpuid(1, &eax1, &ebx1, &ecx1, &edx1)) {\n return (ecx1 & sse42_ecx1) == sse42_" + + "ecx1;\n }\n#elif defined(_MSC_VER) // defined(__GNUC__)\n int x[4];\n __cpuid(x, 1);\n return (((unsigned int)(x[2])) & sse42_ecx1) == sse42_ecx1;\n#else\n#error \"WUFFS_BASE__CPU_ARCH__ETC combined with an unsupported compiler\"\n#endif // defined(__GNUC__); defined(_MSC_VER)\n#endif // defined(WUFFS_BASE__CPU_ARCH__X86_64)\n return false;\n}\n\n" + "" + "// ---------------- Fundamentals\n\n// Wuffs assumes that:\n// - converting a uint32_t to a size_t will never overflow.\n// - converting a size_t to a uint64_t will never overflow.\n#ifdef __WORDSIZE\n#if (__WORDSIZE != 32) && (__WORDSIZE != 64)\n#error \"Wuffs requires a word size of either 32 or 64 bits\"\n#endif\n#endif\n\n#if defined(__clang__)\n#define WUFFS_BASE__POTENTIALLY_UNUSED_FIELD __attribute__((unused))\n#else\n#define WUFFS_BASE__POTENTIALLY_UNUSED_FIELD\n#endif\n\n// Clang also defines \"__GNUC__\".\n#if defined(__GNUC__)\n#define WUFFS_BASE__POTENTIALLY_UNUSED __attribute__((unused))\n#define WUFFS_BASE__WARN_UNUSED_RESULT __attribute__((warn_unused_result))\n#else\n#define WUFFS_BASE__POTENTIALLY_UNUSED\n#define WUFFS_BASE__WARN_UNUSED_RESULT\n#endif\n\n" + "" + @@ -582,9 +583,9 @@ const BasePixConvSubmoduleC = "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__squash_align4_bgr_565_8888(uint8_t* dst_ptr,\n size_t dst_len,\n const uint8_t* src_ptr,\n size_t src_len,\n bool nonpremul) {\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 if (nonpremul) {\n argb =\n wuffs_base__color_u32_argb_nonpremul__as__color_u32_argb_premul(argb);\n }\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\n" + "" + - "// --------\n\nstatic uint64_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__sse42(uint8_t* dst_ptr,\n size_t dst_len,\n ui" + - "nt8_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_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__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)\nWUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(\"sse4.2\")\nstatic uint64_t //\nwuffs_base__pixel_swizzler__swap_rgbx_bgrx__sse42(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_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" + "" + @@ -629,10 +630,10 @@ const BasePixConvSubmoduleC = "" + "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 uint32_t d0 = wuffs_base__peek_u32le__no_bounds_check(d + (0 * 4));\n uint32_t s0 = wuffs_base__swap_u32_argb_abgr(\n wuffs_base__peek_u32le__no_bounds_check(s + (0 * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), wuffs_base__composite_premul_nonpremul_u32_axxx(d0, s0));\n\n s += 1 * 4;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\n" + "" + "// --------\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__sse42(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_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_si128((_" + - "_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" + + "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)\nWUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(\"sse4.2\")\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgrw__rgb__sse42(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_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_si128((__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" + "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxx__index__src(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 if (dst_palette_len != 1024) {\n return 0;\n }\n size_t dst_len3 = dst_len / 3;\n size_t len = (dst_len3 < src_len) ? dst_len3 : src_len;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n // The comparison in the while condition is \">\", not \">=\", because with\n // \">=\", the last 4-byte store could write past the end of the dst slice.\n //\n // Each 4-byte store writes one too many bytes, but a subsequent store\n // will overwrite that with the correct byte. There is always another\n // store, whether a 4-byte store in this loop" + " or a 1-byte store in the\n // next loop.\n while (n > loop_unroll_count) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 3), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[0] * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (1 * 3), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[1] * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (2 * 3), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[2] * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (3 * 3), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[3] * 4)));\n\n s += loop_unroll_count * 1;\n d += loop_unroll_count * 3;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_ptr +\n ((size_t)" + @@ -646,10 +647,10 @@ const BasePixConvSubmoduleC = "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__index__src(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 if (dst_palette_len != 1024) {\n return 0;\n }\n size_t dst_len4 = dst_len / 4;\n size_t len = (dst_len4 < src_len) ? dst_len4 : src_len;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n while (n >= loop_unroll_count) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[0] * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (1 * 4), wuffs_base__peek_u32le__no_bounds_check(\n ds" + "t_palette_ptr + ((size_t)s[1] * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (2 * 4), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[2] * 4)));\n wuffs_base__poke_u32le__no_bounds_check(\n d + (3 * 4), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[3] * 4)));\n\n s += loop_unroll_count * 1;\n d += loop_unroll_count * 4;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), wuffs_base__peek_u32le__no_bounds_check(\n dst_palette_ptr + ((size_t)s[0] * 4)));\n\n s += 1 * 1;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__index_binary_alpha__src_over(\n 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 if (dst_palette_len != 1024) {\n return 0;\n }\n " + " size_t dst_len4 = dst_len / 4;\n size_t len = (dst_len4 < src_len) ? dst_len4 : src_len;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n while (n >= loop_unroll_count) {\n uint32_t s0 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_ptr +\n ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__poke_u32le__no_bounds_check(d + (0 * 4), s0);\n }\n uint32_t s1 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_ptr +\n ((size_t)s[1] * 4));\n if (s1) {\n wuffs_base__poke_u32le__no_bounds_check(d + (1 * 4), s1);\n }\n uint32_t s2 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_ptr +\n ((size_t)s[2] * 4));\n if (s2) {\n wuffs_base__poke_u32le__no_bounds_check(d + (2 * 4), s2);\n }\n uint32_t s3 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_" + - "ptr +\n ((size_t)s[3] * 4));\n if (s3) {\n wuffs_base__poke_u32le__no_bounds_check(d + (3 * 4), s3);\n }\n\n s += loop_unroll_count * 1;\n d += loop_unroll_count * 4;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_ptr +\n ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__poke_u32le__no_bounds_check(d + (0 * 4), s0);\n }\n\n s += 1 * 1;\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__xxxx__y__sse42(uint8_t* dst_ptr,\n size_t dst_len,\n uint8_t* dst_palette_ptr,\n size_t dst_palette_len,\n const ui" + - "nt8_t* src_ptr,\n size_t src_len) {\n size_t dst_len4 = dst_len / 4;\n size_t len = (dst_len4 < src_len) ? dst_len4 : src_len;\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(+0x03, +0x03, +0x03, +0x03, //\n +0x02, +0x02, +0x02, +0x02, //\n +0x01, +0x01, +0x01, +0x01, //\n +0x00, +0x00, +0x00, +0x00);\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 >= 4) {\n __m128i x;\n x = _mm_cvtsi32_si128((int)(wuffs_base__peek_u32le__no_bounds_check(s)));\n x = _mm_shuffle_epi8(x, shuffle);\n x = _mm_or_si128(x, or_ff);\n _mm_storeu_si128((__m128i*)(void*)d, x);\n\n s += 4 * 1;\n d += 4 * 4;\n n -= 4;\n }\n" + - "\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 1;\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__xxxx__y(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 len = (dst_len4 < src_len) ? dst_len4 : src_len;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 1;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__y_16" + - "be(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_len2 = src_len / 2;\n size_t len = (dst_len4 < src_len2) ? dst_len4 : src_len2;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 2;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\n" + + "ptr +\n ((size_t)s[3] * 4));\n if (s3) {\n wuffs_base__poke_u32le__no_bounds_check(d + (3 * 4), s3);\n }\n\n s += loop_unroll_count * 1;\n d += loop_unroll_count * 4;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__peek_u32le__no_bounds_check(dst_palette_ptr +\n ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__poke_u32le__no_bounds_check(d + (0 * 4), s0);\n }\n\n s += 1 * 1;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\n#if defined(WUFFS_BASE__CPU_ARCH__X86_64)\nWUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(\"sse4.2\")\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__y__sse42(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 len = (dst_len4 < src_len) ? dst_len4 : src_len;\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(+0x03, +0x03, +0x03, +0x03, //\n +0x02, +0x02, +0x02, +0x02, //\n +0x01, +0x01, +0x01, +0x01, //\n +0x00, +0x00, +0x00, +0x00);\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 >= 4) {\n __m128i x;\n x = _mm_cvtsi32_si128((int)(wuffs_base__peek_u32le__no_bounds_check(s)));\n x = _mm_shuffle_epi8(x, shuffle);\n x = _mm_or_si128(x, or_ff);\n _mm_storeu_si128((__m128i*)(void*)d, x);\n\n s += 4 * 1;\n d += 4 * 4;\n n -= 4;\n }\n\n while (n >= 1) " + + "{\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 1;\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__xxxx__y(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 len = (dst_len4 < src_len) ? dst_len4 : src_len;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 1;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__y_16be(uint8_t* dst_pt" + + "r,\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_len2 = src_len / 2;\n size_t len = (dst_len4 < src_len2) ? dst_len4 : src_len2;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n while (n >= 1) {\n wuffs_base__poke_u32le__no_bounds_check(\n d + (0 * 4), 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 2;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\n" + "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__y__y_16be(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 src_len2 = src_len / 2;\n size_t len = (dst_len < src_len2) ? dst_len : src_len2;\n uint8_t* d = dst_ptr;\n const uint8_t* s = src_ptr;\n size_t n = len;\n\n while (n >= 1) {\n d[0] = s[0];\n\n s += 1 * 2;\n d += 1 * 1;\n n -= 1;\n }\n\n return len;\n}\n\n" + "" + diff --git a/internal/cgen/func.go b/internal/cgen/func.go index 45263286..6a2c3388 100644 --- a/internal/cgen/func.go +++ b/internal/cgen/func.go @@ -239,7 +239,7 @@ func (g *gen) writeFuncImpl(b *buffer, n *a.Func) error { b.printf("#if defined(WUFFS_BASE__CPU_ARCH__%s)\n", caMacro) } if caAttribute != "" { - b.printf("#if defined(__GNUC__)\n%s\n#endif\n", caAttribute) + b.printf("%s\n", caAttribute) } if err := g.writeFuncSignature(b, n, wfsCDecl); err != nil { diff --git a/internal/cgen/statement.go b/internal/cgen/statement.go index a60a5787..23ee9146 100644 --- a/internal/cgen/statement.go +++ b/internal/cgen/statement.go @@ -277,7 +277,8 @@ func cpuArchCNames(asserts []*a.Node) (caMacro string, caName string, caAttribut caMacro, caName, caAttribute = "ARM_NEON", "arm_neon", "" case t.IDX86SSE42: caMacro, caName, caAttribute = - "X86_64", "x86_sse42", "__attribute__((target(\"pclmul,popcnt,sse4.2\")))" + "X86_64", "x86_sse42", + "WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(\"pclmul,popcnt,sse4.2\")" } } } diff --git a/release/c/wuffs-unsupported-snapshot.c b/release/c/wuffs-unsupported-snapshot.c index d548daad..fcc55225 100644 --- a/release/c/wuffs-unsupported-snapshot.c +++ b/release/c/wuffs-unsupported-snapshot.c @@ -78,10 +78,19 @@ extern "C" { // Define WUFFS_CONFIG__AVOID_CPU_ARCH to avoid any code tied to a specific CPU // architecture, such as SSE SIMD for the x86 CPU family. -#if defined(WUFFS_CONFIG__AVOID_CPU_ARCH) +#if defined(WUFFS_CONFIG__AVOID_CPU_ARCH) // (#if-chain ref AVOID_CPU_ARCH_0) // No-op. +#else // (#if-chain ref AVOID_CPU_ARCH_0) + +// The "defined(__clang__)" isn't redundant. While vanilla clang defines +// __GNUC__, clang-cl (which mimics MSVC's cl.exe) does not. +#if defined(__GNUC__) || defined(__clang__) +#define WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(arg) __attribute__((target(arg))) #else -#if defined(__GNUC__) +#define WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET(arg) +#endif // defined(__GNUC__) || defined(__clang__) + +#if defined(__GNUC__) // (#if-chain ref AVOID_CPU_ARCH_1) // To simplify Wuffs code, "cpu_arch >= arm_xxx" requires xxx but also // unaligned little-endian load/stores. @@ -107,15 +116,24 @@ extern "C" { #define WUFFS_BASE__CPU_ARCH__X86_64 #endif // defined(__x86_64__) -#elif defined(_MSC_VER) // defined(__GNUC__) +#elif defined(_MSC_VER) // (#if-chain ref AVOID_CPU_ARCH_1) #if defined(_M_X64) +#if defined(__clang__) +// No-op. clang-cl (which defines both __clang__ and _MSC_VER) supports +// "__attribute__((target(arg)))". +#elif !defined(__AVX__) +// For MSVC's cl.exe (unlike clang or gcc), SIMD capability is a compile-time +// property of the source file (e.g. a /arch:AVX or -mavx compiler flag), not +// of individual functions (that can be conditionally selected at runtime). +#error "Wuffs with MSVC+X64 needs /arch:AVX or /DWUFFS_CONFIG__AVOID_CPU_ARCH" +#endif // defined(__clang__); !defined(__AVX__) #include #define WUFFS_BASE__CPU_ARCH__X86_64 -#endif // defined(__x86_64__) +#endif // defined(_M_X64) -#endif // defined(__GNUC__); defined(_MSC_VER) -#endif // defined(WUFFS_CONFIG__AVOID_CPU_ARCH) +#endif // (#if-chain ref AVOID_CPU_ARCH_1) +#endif // (#if-chain ref AVOID_CPU_ARCH_0) // -------- @@ -157,6 +175,8 @@ wuffs_base__cpu_arch__have_x86_sse42() { // - bit_POPCNT = (1 << 23) // - bit_SSE4_2 = (1 << 20) const unsigned int sse42_ecx1 = 0x00900002; + + // clang defines __GNUC__ and clang-cl defines _MSC_VER (but not __GNUC__). #if defined(__GNUC__) unsigned int eax1 = 0; unsigned int ebx1 = 0; @@ -15463,9 +15483,7 @@ wuffs_base__pixel_swizzler__swap_rgb_bgr(uint8_t* dst_ptr, } #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("sse4.2") static uint64_t // wuffs_base__pixel_swizzler__swap_rgbx_bgrx__sse42(uint8_t* dst_ptr, size_t dst_len, @@ -16808,9 +16826,7 @@ wuffs_base__pixel_swizzler__bgrw__bgrx(uint8_t* dst_ptr, } #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("sse4.2") static uint64_t // wuffs_base__pixel_swizzler__bgrw__rgb__sse42(uint8_t* dst_ptr, size_t dst_len, @@ -17251,9 +17267,7 @@ wuffs_base__pixel_swizzler__xxxx__index_binary_alpha__src_over( } #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("sse4.2") static uint64_t // wuffs_base__pixel_swizzler__xxxx__y__sse42(uint8_t* dst_ptr, size_t dst_len, @@ -18700,9 +18714,7 @@ wuffs_adler32__hasher__up_arm_neon( // -------- func adler32.hasher.up_x86_sse42 #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("pclmul,popcnt,sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2") static wuffs_base__empty_struct wuffs_adler32__hasher__up_x86_sse42( wuffs_adler32__hasher* self, @@ -22881,9 +22893,7 @@ wuffs_crc32__ieee_hasher__up_arm_crc32( // -------- func crc32.ieee_hasher.up_x86_sse42 #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("pclmul,popcnt,sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2") static wuffs_base__empty_struct wuffs_crc32__ieee_hasher__up_x86_sse42( wuffs_crc32__ieee_hasher* self, @@ -33533,9 +33543,7 @@ wuffs_png__decoder__filter_4_distance_4_fallback( // -------- func png.decoder.filter_1_distance_4_x86_sse42 #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("pclmul,popcnt,sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2") static wuffs_base__empty_struct wuffs_png__decoder__filter_1_distance_4_x86_sse42( wuffs_png__decoder* self, @@ -33579,9 +33587,7 @@ wuffs_png__decoder__filter_1_distance_4_x86_sse42( // -------- func png.decoder.filter_3_distance_4_x86_sse42 #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("pclmul,popcnt,sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2") static wuffs_base__empty_struct wuffs_png__decoder__filter_3_distance_4_x86_sse42( wuffs_png__decoder* self, @@ -33684,9 +33690,7 @@ wuffs_png__decoder__filter_3_distance_4_x86_sse42( // -------- func png.decoder.filter_4_distance_3_x86_sse42 #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("pclmul,popcnt,sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2") static wuffs_base__empty_struct wuffs_png__decoder__filter_4_distance_3_x86_sse42( wuffs_png__decoder* self, @@ -33810,9 +33814,7 @@ wuffs_png__decoder__filter_4_distance_3_x86_sse42( // -------- func png.decoder.filter_4_distance_4_x86_sse42 #if defined(WUFFS_BASE__CPU_ARCH__X86_64) -#if defined(__GNUC__) -__attribute__((target("pclmul,popcnt,sse4.2"))) -#endif +WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2") static wuffs_base__empty_struct wuffs_png__decoder__filter_4_distance_4_x86_sse42( wuffs_png__decoder* self,