@img/sharp-libvips-dev 1.0.0 → 1.0.2

package/include/hwy/cache_control.h

@@ -25,11 +25,15 @@
 #define HWY_DISABLE_CACHE_CONTROL
 #endif
 
+#ifndef HWY_DISABLE_CACHE_CONTROL
 // intrin.h is sufficient on MSVC and already included by base.h.
-#if HWY_ARCH_X86 && !defined(HWY_DISABLE_CACHE_CONTROL) && !HWY_COMPILER_MSVC
+#if HWY_ARCH_X86 && !HWY_COMPILER_MSVC
 #include <emmintrin.h>  // SSE2
 #include <xmmintrin.h>  // _mm_prefetch
+#elif HWY_ARCH_ARM_A64
+#include <arm_acle.h>
 #endif
+#endif  // HWY_DISABLE_CACHE_CONTROL
 
 namespace hwy {
 
@@ -76,15 +80,16 @@ HWY_INLINE HWY_ATTR_CACHE void FlushStream() {
 // subsequent actual loads.
 template <typename T>
 HWY_INLINE HWY_ATTR_CACHE void Prefetch(const T* p) {
-#if HWY_ARCH_X86 && !defined(HWY_DISABLE_CACHE_CONTROL)
+  (void)p;
+#ifndef HWY_DISABLE_CACHE_CONTROL
+#if HWY_ARCH_X86
   _mm_prefetch(reinterpret_cast<const char*>(p), _MM_HINT_T0);
 #elif HWY_COMPILER_GCC  // includes clang
   // Hint=0 (NTA) behavior differs, but skipping outer caches is probably not
   // desirable, so use the default 3 (keep in caches).
   __builtin_prefetch(p, /*write=*/0, /*hint=*/3);
-#else
-  (void)p;
 #endif
+#endif  //  HWY_DISABLE_CACHE_CONTROL
 }
 
 // Invalidates and flushes the cache line containing "p", if possible.
@@ -96,11 +101,24 @@ HWY_INLINE HWY_ATTR_CACHE void FlushCacheline(const void* p) {
 #endif
 }
 
-// When called inside a spin-loop, may reduce power consumption.
+// Hints that we are inside a spin loop and potentially reduces power
+// consumption and coherency traffic. For example, x86 avoids multiple
+// outstanding load requests, which reduces the memory order violation penalty
+// when exiting the loop.
 HWY_INLINE HWY_ATTR_CACHE void Pause() {
-#if HWY_ARCH_X86 && !defined(HWY_DISABLE_CACHE_CONTROL)
+#ifndef HWY_DISABLE_CACHE_CONTROL
+#if HWY_ARCH_X86
   _mm_pause();
+#elif HWY_ARCH_ARM_A64 && HWY_COMPILER_CLANG
+  // This is documented in ACLE and the YIELD instruction is also available in
+  // Armv7, but the intrinsic is broken for Armv7 clang, hence A64 only.
+  __yield();
+#elif HWY_ARCH_ARM && HWY_COMPILER_GCC  // includes clang
+  __asm__ volatile("yield" ::: "memory");
+#elif HWY_ARCH_PPC && HWY_COMPILER_GCC  // includes clang
+  __asm__ volatile("or 27,27,27" ::: "memory");
 #endif
+#endif  // HWY_DISABLE_CACHE_CONTROL
 }
 
 }  // namespace hwy

package/include/hwy/detect_compiler_arch.h

@@ -73,7 +73,9 @@
 // https://github.com/simd-everywhere/simde/blob/47d6e603de9d04ee05cdfbc57cf282a02be1bf2a/simde/simde-detect-clang.h#L59.
 // Please send updates below to them as well, thanks!
 #if defined(__apple_build_version__) || __clang_major__ >= 999
-#if __has_attribute(nouwtable)  // no new warnings in 16.0
+#if __has_attribute(unsafe_buffer_usage)  // no new warnings in 17.0
+#define HWY_COMPILER_CLANG 1700
+#elif __has_attribute(nouwtable)  // no new warnings in 16.0
 #define HWY_COMPILER_CLANG 1600
 #elif __has_warning("-Warray-parameter")
 #define HWY_COMPILER_CLANG 1500
@@ -187,6 +189,12 @@
 #define HWY_ARCH_PPC 0
 #endif
 
+#if defined(__powerpc64__) || (HWY_ARCH_PPC && defined(__64BIT__))
+#define HWY_ARCH_PPC_64 1
+#else
+#define HWY_ARCH_PPC_64 0
+#endif
+
 // aarch32 is currently not supported; please raise an issue if you want it.
 #if defined(__ARM_ARCH_ISA_A64) || defined(__aarch64__) || defined(_M_ARM64)
 #define HWY_ARCH_ARM_A64 1
@@ -230,10 +238,16 @@
 #define HWY_ARCH_RVV 0
 #endif
 
+#if defined(__s390x__)
+#define HWY_ARCH_S390X 1
+#else
+#define HWY_ARCH_S390X 0
+#endif
+
 // It is an error to detect multiple architectures at the same time, but OK to
 // detect none of the above.
 #if (HWY_ARCH_X86 + HWY_ARCH_PPC + HWY_ARCH_ARM + HWY_ARCH_ARM_OLD + \
-     HWY_ARCH_WASM + HWY_ARCH_RVV) > 1
+     HWY_ARCH_WASM + HWY_ARCH_RVV + HWY_ARCH_S390X) > 1
 #error "Must not detect more than one architecture"
 #endif
 
@@ -249,6 +263,13 @@
 #define HWY_OS_LINUX 0
 #endif
 
+// iOS or Mac
+#if defined(__APPLE__)
+#define HWY_OS_APPLE 1
+#else
+#define HWY_OS_APPLE 0
+#endif
+
 //------------------------------------------------------------------------------
 // Endianness
 

package/include/hwy/detect_targets.h

@@ -102,12 +102,13 @@
 // --------------------------- Future expansion: 4 targets
 // Bits 39..42 reserved
 
-// --------------------------- IBM Power: 9 targets (+ one fallback)
+// --------------------------- IBM Power/ZSeries: 9 targets (+ one fallback)
 // Bits 43..46 reserved (4 targets)
 #define HWY_PPC10 (1LL << 47)  // v3.1
 #define HWY_PPC9 (1LL << 48)   // v3.0
 #define HWY_PPC8 (1LL << 49)   // v2.07
-// Bits 50..51 reserved for prior VSX/AltiVec (2 targets)
+#define HWY_Z15 (1LL << 50)    // Z15
+#define HWY_Z14 (1LL << 51)    // Z14
 #define HWY_HIGHEST_TARGET_BIT_PPC 51
 
 // --------------------------- WebAssembly: 9 targets (+ one fallback)
@@ -316,6 +317,18 @@
 #define HWY_BASELINE_PPC10 0
 #endif
 
+#if HWY_ARCH_S390X && defined(__VEC__) && defined(__ARCH__) && __ARCH__ >= 12
+#define HWY_BASELINE_Z14 HWY_Z14
+#else
+#define HWY_BASELINE_Z14 0
+#endif
+
+#if HWY_BASELINE_Z14 && __ARCH__ >= 13
+#define HWY_BASELINE_Z15 HWY_Z15
+#else
+#define HWY_BASELINE_Z15 0
+#endif
+
 #define HWY_BASELINE_SVE2 0
 #define HWY_BASELINE_SVE 0
 #define HWY_BASELINE_NEON 0
@@ -498,13 +511,14 @@
 
 // Allow the user to override this without any guarantee of success.
 #ifndef HWY_BASELINE_TARGETS
-#define HWY_BASELINE_TARGETS                                           \
-  (HWY_BASELINE_SCALAR | HWY_BASELINE_WASM | HWY_BASELINE_PPC8 |       \
-   HWY_BASELINE_PPC9 | HWY_BASELINE_PPC10 | HWY_BASELINE_SVE2 |        \
-   HWY_BASELINE_SVE | HWY_BASELINE_NEON | HWY_BASELINE_SSE2 |          \
-   HWY_BASELINE_SSSE3 | HWY_BASELINE_SSE4 | HWY_BASELINE_AVX2 |        \
-   HWY_BASELINE_AVX3 | HWY_BASELINE_AVX3_DL | HWY_BASELINE_AVX3_ZEN4 | \
-   HWY_BASELINE_AVX3_SPR | HWY_BASELINE_RVV)
+#define HWY_BASELINE_TARGETS                                               \
+  (HWY_BASELINE_SCALAR | HWY_BASELINE_WASM | HWY_BASELINE_PPC8 |           \
+   HWY_BASELINE_PPC9 | HWY_BASELINE_PPC10 | HWY_BASELINE_Z14 |             \
+   HWY_BASELINE_Z15 | HWY_BASELINE_SVE2 | HWY_BASELINE_SVE |               \
+   HWY_BASELINE_NEON | HWY_BASELINE_SSE2 | HWY_BASELINE_SSSE3 |            \
+   HWY_BASELINE_SSE4 | HWY_BASELINE_AVX2 | HWY_BASELINE_AVX3 |             \
+   HWY_BASELINE_AVX3_DL | HWY_BASELINE_AVX3_ZEN4 | HWY_BASELINE_AVX3_SPR | \
+   HWY_BASELINE_RVV)
 #endif  // HWY_BASELINE_TARGETS
 
 //------------------------------------------------------------------------------
@@ -537,9 +551,11 @@
 // Clang, GCC and MSVC allow runtime dispatch on x86.
 #if HWY_ARCH_X86
 #define HWY_HAVE_RUNTIME_DISPATCH 1
-// On Arm/PPC, currently only GCC does, and we require Linux to detect CPU
-// capabilities.
-#elif (HWY_ARCH_ARM || HWY_ARCH_PPC) && HWY_COMPILER_GCC_ACTUAL && \
+// On Arm/PPC, GCC and Clang 16+ do, and we require Linux to detect CPU
+// capabilities. Currently require opt-in for Clang because it is experimental.
+#elif (HWY_ARCH_ARM || HWY_ARCH_PPC || HWY_ARCH_S390X) &&                    \
+    (HWY_COMPILER_GCC_ACTUAL || (HWY_COMPILER_CLANG >= 1600 &&               \
+                                 defined(HWY_ENABLE_CLANG_ARM_DISPATCH))) && \
     HWY_OS_LINUX && !defined(TOOLCHAIN_MISS_SYS_AUXV_H)
 #define HWY_HAVE_RUNTIME_DISPATCH 1
 #else
@@ -579,18 +595,41 @@
 
 #if HWY_ARCH_PPC && defined(__ALTIVEC__) && \
     (!HWY_COMPILER_CLANG || HWY_BASELINE_PPC8 != 0)
+
+#if (HWY_BASELINE_PPC9 | HWY_BASELINE_PPC10) && \
+    !defined(HWY_SKIP_NON_BEST_BASELINE)
+// On POWER with -m flags, we get compile errors (#1707) for targets older than
+// the baseline specified via -m, so only generate the static target and better.
+// Note that some Linux distros actually do set POWER9 as the baseline.
+// This works by skipping case 3 below, so case 4 is reached.
+#define HWY_SKIP_NON_BEST_BASELINE
+#endif
+
 #define HWY_ATTAINABLE_PPC (HWY_PPC8 | HWY_PPC9 | HWY_PPC10)
+
 #else
 #define HWY_ATTAINABLE_PPC 0
 #endif
 
+#if HWY_ARCH_S390X && HWY_BASELINE_Z14 != 0
+#define HWY_ATTAINABLE_S390X (HWY_Z14 | HWY_Z15)
+#else
+#define HWY_ATTAINABLE_S390X 0
+#endif
+
 // Attainable means enabled and the compiler allows intrinsics (even when not
 // allowed to autovectorize). Used in 3 and 4.
 #if HWY_ARCH_X86
+#if HWY_COMPILER_MSVC
+// Fewer targets for faster builds.
+#define HWY_ATTAINABLE_TARGETS \
+  HWY_ENABLED(HWY_BASELINE_SCALAR | HWY_STATIC_TARGET | HWY_AVX2)
+#else  // !HWY_COMPILER_MSVC
 #define HWY_ATTAINABLE_TARGETS                                               \
   HWY_ENABLED(HWY_BASELINE_SCALAR | HWY_SSE2 | HWY_SSSE3 | HWY_SSE4 |        \
               HWY_AVX2 | HWY_AVX3 | HWY_ATTAINABLE_AVX3_DL | HWY_AVX3_ZEN4 | \
               HWY_AVX3_SPR)
+#endif  // !HWY_COMPILER_MSVC
 #elif HWY_ARCH_ARM
 #define HWY_ATTAINABLE_TARGETS                                                 \
   HWY_ENABLED(HWY_BASELINE_SCALAR | HWY_ATTAINABLE_NEON | HWY_ATTAINABLE_SVE | \
@@ -598,6 +637,9 @@
 #elif HWY_ARCH_PPC
 #define HWY_ATTAINABLE_TARGETS \
   HWY_ENABLED(HWY_BASELINE_SCALAR | HWY_ATTAINABLE_PPC)
+#elif HWY_ARCH_S390X
+#define HWY_ATTAINABLE_TARGETS \
+  HWY_ENABLED(HWY_BASELINE_SCALAR | HWY_ATTAINABLE_S390X)
 #else
 #define HWY_ATTAINABLE_TARGETS (HWY_ENABLED_BASELINE)
 #endif  // HWY_ARCH_*
@@ -621,7 +663,8 @@
 #define HWY_TARGETS HWY_STATIC_TARGET
 
 // 3) For tests: include all attainable targets (in particular: scalar)
-#elif defined(HWY_COMPILE_ALL_ATTAINABLE) || defined(HWY_IS_TEST)
+#elif (defined(HWY_COMPILE_ALL_ATTAINABLE) || defined(HWY_IS_TEST)) && \
+    !defined(HWY_SKIP_NON_BEST_BASELINE)
 #define HWY_TARGETS HWY_ATTAINABLE_TARGETS
 
 // 4) Default: attainable WITHOUT non-best baseline. This reduces code size by

package/include/hwy/foreach_target.h

@@ -271,6 +271,30 @@
 #endif
 #endif
 
+// ------------------------------ HWY_ARCH_S390X
+
+#if (HWY_TARGETS & HWY_Z14) && (HWY_STATIC_TARGET != HWY_Z14)
+#undef HWY_TARGET
+#define HWY_TARGET HWY_Z14
+#include HWY_TARGET_INCLUDE
+#ifdef HWY_TARGET_TOGGLE
+#undef HWY_TARGET_TOGGLE
+#else
+#define HWY_TARGET_TOGGLE
+#endif
+#endif
+
+#if (HWY_TARGETS & HWY_Z15) && (HWY_STATIC_TARGET != HWY_Z15)
+#undef HWY_TARGET
+#define HWY_TARGET HWY_Z15
+#include HWY_TARGET_INCLUDE
+#ifdef HWY_TARGET_TOGGLE
+#undef HWY_TARGET_TOGGLE
+#else
+#define HWY_TARGET_TOGGLE
+#endif
+#endif
+
 // ------------------------------ HWY_ARCH_RVV
 
 #if (HWY_TARGETS & HWY_RVV) && (HWY_STATIC_TARGET != HWY_RVV)

package/include/hwy/highway.h

@@ -32,8 +32,8 @@ namespace hwy {
 
 // API version (https://semver.org/); keep in sync with CMakeLists.txt.
 #define HWY_MAJOR 1
-#define HWY_MINOR 0
-#define HWY_PATCH 7
+#define HWY_MINOR 1
+#define HWY_PATCH 0
 
 //------------------------------------------------------------------------------
 // Shorthand for tags (defined in shared-inl.h) used to select overloads.
@@ -98,6 +98,10 @@ namespace hwy {
 #define HWY_STATIC_DISPATCH(FUNC_NAME) N_PPC9::FUNC_NAME
 #elif HWY_STATIC_TARGET == HWY_PPC10
 #define HWY_STATIC_DISPATCH(FUNC_NAME) N_PPC10::FUNC_NAME
+#elif HWY_STATIC_TARGET == HWY_Z14
+#define HWY_STATIC_DISPATCH(FUNC_NAME) N_Z14::FUNC_NAME
+#elif HWY_STATIC_TARGET == HWY_Z15
+#define HWY_STATIC_DISPATCH(FUNC_NAME) N_Z15::FUNC_NAME
 #elif HWY_STATIC_TARGET == HWY_SSE2
 #define HWY_STATIC_DISPATCH(FUNC_NAME) N_SSE2::FUNC_NAME
 #elif HWY_STATIC_TARGET == HWY_SSSE3
@@ -200,6 +204,18 @@ namespace hwy {
 #define HWY_CHOOSE_PPC10(FUNC_NAME) nullptr
 #endif
 
+#if HWY_TARGETS & HWY_Z14
+#define HWY_CHOOSE_Z14(FUNC_NAME) &N_Z14::FUNC_NAME
+#else
+#define HWY_CHOOSE_Z14(FUNC_NAME) nullptr
+#endif
+
+#if HWY_TARGETS & HWY_Z15
+#define HWY_CHOOSE_Z15(FUNC_NAME) &N_Z15::FUNC_NAME
+#else
+#define HWY_CHOOSE_Z15(FUNC_NAME) nullptr
+#endif
+
 #if HWY_TARGETS & HWY_SSE2
 #define HWY_CHOOSE_SSE2(FUNC_NAME) &N_SSE2::FUNC_NAME
 #else
@@ -408,7 +424,8 @@ FunctionCache<RetType, Args...> DeduceFunctionCache(RetType (*)(Args...)) {
 #elif HWY_TARGET == HWY_AVX3 || HWY_TARGET == HWY_AVX3_DL || \
     HWY_TARGET == HWY_AVX3_ZEN4 || HWY_TARGET == HWY_AVX3_SPR
 #include "hwy/ops/x86_512-inl.h"
-#elif HWY_TARGET == HWY_PPC8 || HWY_TARGET == HWY_PPC9 || \
+#elif HWY_TARGET == HWY_Z14 || HWY_TARGET == HWY_Z15 || \
+    HWY_TARGET == HWY_PPC8 || HWY_TARGET == HWY_PPC9 || \
     HWY_TARGET == HWY_PPC10
 #include "hwy/ops/ppc_vsx-inl.h"
 #elif HWY_TARGET == HWY_NEON || HWY_TARGET == HWY_NEON_WITHOUT_AES