@zigc/lib 0.17.0-dev.135 → 0.17.0-dev.215

package/include/immintrin.h

@@ -475,24 +475,12 @@ _storebe_i64(void * __P, long long __D) {
 
 #include <amxfp8intrin.h>
 
-#include <amxtransposeintrin.h>
-
 #include <amxmovrsintrin.h>
 
-#include <amxmovrstransposeintrin.h>
-
 #include <amxavx512intrin.h>
 
 #include <amxtf32intrin.h>
 
-#include <amxtf32transposeintrin.h>
-
-#include <amxbf16transposeintrin.h>
-
-#include <amxfp16transposeintrin.h>
-
-#include <amxcomplextransposeintrin.h>
-
 #include <avx512vp2intersectintrin.h>
 
 #include <avx512vlvp2intersectintrin.h>

package/include/intrin.h

@@ -30,6 +30,10 @@
 #include <arm64intr.h>
 #endif
 
+#if defined(__ARM_ACLE)
+#include <arm_acle.h>
+#endif
+
 /* For the definition of jmp_buf. */
 #if __STDC_HOSTED__
 #include <setjmp.h>

package/include/lasxintrin.h

@@ -10,6 +10,8 @@
 #ifndef _LOONGSON_ASXINTRIN_H
 #define _LOONGSON_ASXINTRIN_H 1
 
+#include <lsxintrin.h>
+
 #if defined(__loongarch_asx)
 
 typedef signed char v32i8 __attribute__((vector_size(32), aligned(32)));
@@ -3882,5 +3884,116 @@ extern __inline
 
 #define __lasx_xvrepli_w(/*si10*/ _1) ((__m256i)__builtin_lasx_xvrepli_w((_1)))
 
+#if defined(__loongarch_asx_sx_conv)
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__,
+                   __artificial__)) __m256 __lasx_cast_128_s(__m128 _1) {
+  return (__m256)__builtin_lasx_cast_128_s((v4f32)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256d
+    __lasx_cast_128_d(__m128d _1) {
+  return (__m256d)__builtin_lasx_cast_128_d((v2f64)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256i
+    __lasx_cast_128(__m128i _1) {
+  return (__m256i)__builtin_lasx_cast_128((v2i64)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256
+    __lasx_concat_128_s(__m128 _1, __m128 _2) {
+  return (__m256)__builtin_lasx_concat_128_s((v4f32)_1, (v4f32)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256d
+    __lasx_concat_128_d(__m128d _1, __m128d _2) {
+  return (__m256d)__builtin_lasx_concat_128_d((v2f64)_1, (v2f64)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256i
+    __lasx_concat_128(__m128i _1, __m128i _2) {
+  return (__m256i)__builtin_lasx_concat_128((v2i64)_1, (v2i64)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m128
+    __lasx_extract_128_lo_s(__m256 _1) {
+  return (__m128)__builtin_lasx_extract_128_lo_s((v8f32)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m128d
+    __lasx_extract_128_lo_d(__m256d _1) {
+  return (__m128d)__builtin_lasx_extract_128_lo_d((v4f64)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m128i
+    __lasx_extract_128_lo(__m256i _1) {
+  return (__m128i)__builtin_lasx_extract_128_lo((v4i64)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m128
+    __lasx_extract_128_hi_s(__m256 _1) {
+  return (__m128)__builtin_lasx_extract_128_hi_s((v8f32)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m128d
+    __lasx_extract_128_hi_d(__m256d _1) {
+  return (__m128d)__builtin_lasx_extract_128_hi_d((v4f64)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m128i
+    __lasx_extract_128_hi(__m256i _1) {
+  return (__m128i)__builtin_lasx_extract_128_hi((v4i64)_1);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256
+    __lasx_insert_128_lo_s(__m256 _1, __m128 _2) {
+  return (__m256)__builtin_lasx_insert_128_lo_s((v8f32)_1, (v4f32)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256d
+    __lasx_insert_128_lo_d(__m256d _1, __m128d _2) {
+  return (__m256d)__builtin_lasx_insert_128_lo_d((v4f64)_1, (v2f64)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256i
+    __lasx_insert_128_lo(__m256i _1, __m128i _2) {
+  return (__m256i)__builtin_lasx_insert_128_lo((v4i64)_1, (v2i64)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256
+    __lasx_insert_128_hi_s(__m256 _1, __m128 _2) {
+  return (__m256)__builtin_lasx_insert_128_hi_s((v8f32)_1, (v4f32)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256d
+    __lasx_insert_128_hi_d(__m256d _1, __m128d _2) {
+  return (__m256d)__builtin_lasx_insert_128_hi_d((v4f64)_1, (v2f64)_2);
+}
+
+extern __inline
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__)) __m256i
+    __lasx_insert_128_hi(__m256i _1, __m128i _2) {
+  return (__m256i)__builtin_lasx_insert_128_hi((v4i64)_1, (v2i64)_2);
+}
+
+#endif /* defined(__loongarch_asx_sx_conv).  */
 #endif /* defined(__loongarch_asx).  */
 #endif /* _LOONGSON_ASXINTRIN_H.  */

package/include/llvm_libc_wrappers/assert.h

@@ -19,13 +19,11 @@
 
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
+#else
+#define __LIBC_ATTRS
 #endif
 
-#pragma omp begin declare target
-
-#include <llvm-libc-decls/assert.h>
-
-#pragma omp end declare target
+// TODO: Define these for CUDA / HIP.
 
 #undef __LIBC_ATTRS
 

package/include/llvm_libc_wrappers/ctype.h

@@ -13,128 +13,16 @@
 #error "This file is for GPU offloading compilation only"
 #endif
 
-// The GNU headers like to define 'toupper' and 'tolower' redundantly. This is
-// necessary to prevent it from doing that and remapping our implementation.
-#if (defined(__NVPTX__) || defined(__AMDGPU__)) && defined(__GLIBC__)
-#pragma push_macro("__USE_EXTERN_INLINES")
-#undef __USE_EXTERN_INLINES
-#endif
-
 #include_next <ctype.h>
 
-#if (defined(__NVPTX__) || defined(__AMDGPU__)) && defined(__GLIBC__)
-#pragma pop_macro("__USE_EXTERN_INLINES")
-#endif
-
-#if __has_include(<llvm-libc-decls/ctype.h>)
-
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
+#else
+#define __LIBC_ATTRS
 #endif
 
-// The GNU headers like to provide these as macros, we need to undefine them so
-// they do not conflict with the following definitions for the GPU.
-
-#pragma push_macro("isalnum")
-#pragma push_macro("isalpha")
-#pragma push_macro("isascii")
-#pragma push_macro("isblank")
-#pragma push_macro("iscntrl")
-#pragma push_macro("isdigit")
-#pragma push_macro("isgraph")
-#pragma push_macro("islower")
-#pragma push_macro("isprint")
-#pragma push_macro("ispunct")
-#pragma push_macro("isspace")
-#pragma push_macro("isupper")
-#pragma push_macro("isxdigit")
-#pragma push_macro("toascii")
-#pragma push_macro("tolower")
-#pragma push_macro("toupper")
-#pragma push_macro("isalnum_l")
-#pragma push_macro("isalpha_l")
-#pragma push_macro("isascii_l")
-#pragma push_macro("isblank_l")
-#pragma push_macro("iscntrl_l")
-#pragma push_macro("isdigit_l")
-#pragma push_macro("isgraph_l")
-#pragma push_macro("islower_l")
-#pragma push_macro("isprint_l")
-#pragma push_macro("ispunct_l")
-#pragma push_macro("isspace_l")
-#pragma push_macro("isupper_l")
-#pragma push_macro("isxdigit_l")
-
-#undef isalnum
-#undef isalpha
-#undef isascii
-#undef iscntrl
-#undef isdigit
-#undef islower
-#undef isgraph
-#undef isprint
-#undef ispunct
-#undef isspace
-#undef isupper
-#undef isblank
-#undef isxdigit
-#undef toascii
-#undef tolower
-#undef toupper
-#undef isalnum_l
-#undef isalpha_l
-#undef iscntrl_l
-#undef isdigit_l
-#undef islower_l
-#undef isgraph_l
-#undef isprint_l
-#undef ispunct_l
-#undef isspace_l
-#undef isupper_l
-#undef isblank_l
-#undef isxdigit_l
-
-#pragma omp begin declare target
-
-#include <llvm-libc-decls/ctype.h>
-
-#pragma omp end declare target
-
-// Restore the original macros when compiling on the host.
-#if !defined(__NVPTX__) && !defined(__AMDGPU__)
-#pragma pop_macro("isalnum")
-#pragma pop_macro("isalpha")
-#pragma pop_macro("isascii")
-#pragma pop_macro("isblank")
-#pragma pop_macro("iscntrl")
-#pragma pop_macro("isdigit")
-#pragma pop_macro("isgraph")
-#pragma pop_macro("islower")
-#pragma pop_macro("isprint")
-#pragma pop_macro("ispunct")
-#pragma pop_macro("isspace")
-#pragma pop_macro("isupper")
-#pragma pop_macro("isxdigit")
-#pragma pop_macro("toascii")
-#pragma pop_macro("tolower")
-#pragma pop_macro("toupper")
-#pragma pop_macro("isalnum_l")
-#pragma pop_macro("isalpha_l")
-#pragma pop_macro("isascii_l")
-#pragma pop_macro("isblank_l")
-#pragma pop_macro("iscntrl_l")
-#pragma pop_macro("isdigit_l")
-#pragma pop_macro("isgraph_l")
-#pragma pop_macro("islower_l")
-#pragma pop_macro("isprint_l")
-#pragma pop_macro("ispunct_l")
-#pragma pop_macro("isspace_l")
-#pragma pop_macro("isupper_l")
-#pragma pop_macro("isxdigit_l")
-#endif
+// TODO: Define these for CUDA / HIP.
 
 #undef __LIBC_ATTRS
 
-#endif
-
 #endif // __CLANG_LLVM_LIBC_WRAPPERS_CTYPE_H__

package/include/llvm_libc_wrappers/inttypes.h

@@ -19,13 +19,11 @@
 
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
+#else
+#define __LIBC_ATTRS
 #endif
 
-#pragma omp begin declare target
-
-#include <llvm-libc-decls/inttypes.h>
-
-#pragma omp end declare target
+// TODO: Define these for CUDA / HIP.
 
 #undef __LIBC_ATTRS
 

package/include/llvm_libc_wrappers/stdio.h

@@ -6,45 +6,19 @@
 //
 //===----------------------------------------------------------------------===//
 
+#ifndef __CLANG_LLVM_LIBC_WRAPPERS_STDIO_H__
+#define __CLANG_LLVM_LIBC_WRAPPERS_STDIO_H__
+
 #if !defined(_OPENMP) && !defined(__HIP__) && !defined(__CUDA__)
 #error "This file is for GPU offloading compilation only"
 #endif
 
 #include_next <stdio.h>
 
-// In some old versions of glibc, other standard headers sometimes define
-// special macros (e.g., __need_FILE) before including stdio.h to cause stdio.h
-// to produce special definitions.  Future includes of stdio.h when those
-// special macros are undefined are expected to produce the normal definitions
-// from stdio.h.
-//
-// We do not apply our include guard (__CLANG_LLVM_LIBC_WRAPPERS_STDIO_H__)
-// unconditionally to the above include_next.  Otherwise, after an occurrence of
-// the first glibc stdio.h use case described above, the include_next would be
-// skipped for remaining includes of stdio.h, leaving required symbols
-// undefined.
-//
-// We make the following assumptions to handle all use cases:
-//
-// 1. If the above include_next produces special glibc definitions, then (a) it
-//    does not produce the normal definitions that we must intercept below, (b)
-//    the current file was included from a glibc header that already defined
-//    __GLIBC__ (usually by including glibc's <features.h>), and (c) the above
-//    include_next does not define _STDIO_H.  In that case, we skip the rest of
-//    the current file and don't guard against future includes.
-// 2. If the above include_next produces the normal stdio.h definitions, then
-//    either (a) __GLIBC__ is not defined because C headers are from some other
-//    libc implementation or (b) the above include_next defines _STDIO_H to
-//    prevent the above include_next from having any effect in the future.
-#if !defined(__GLIBC__) || defined(_STDIO_H)
-
-#ifndef __CLANG_LLVM_LIBC_WRAPPERS_STDIO_H__
-#define __CLANG_LLVM_LIBC_WRAPPERS_STDIO_H__
-
-#if __has_include(<llvm-libc-decls/stdio.h>)
-
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
+#else
+#define __LIBC_ATTRS
 #endif
 
 // Some headers provide these as macros. Temporarily undefine them so they do
@@ -60,21 +34,19 @@
 
 #pragma omp begin declare target
 
-#include <llvm-libc-decls/stdio.h>
+__LIBC_ATTRS extern FILE *stderr;
+__LIBC_ATTRS extern FILE *stdin;
+__LIBC_ATTRS extern FILE *stdout;
 
 #pragma omp end declare target
 
-#undef __LIBC_ATTRS
-
 // Restore the original macros when compiling on the host.
 #if !defined(__NVPTX__) && !defined(__AMDGPU__)
-#pragma pop_macro("stdout")
 #pragma pop_macro("stderr")
 #pragma pop_macro("stdin")
+#pragma pop_macro("stdout")
 #endif
 
-#endif
+#undef __LIBC_ATTRS
 
 #endif // __CLANG_LLVM_LIBC_WRAPPERS_STDIO_H__
-
-#endif

package/include/llvm_libc_wrappers/stdlib.h

@@ -15,39 +15,18 @@
 
 #include_next <stdlib.h>
 
-#if __has_include(<llvm-libc-decls/stdlib.h>)
-
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
+#else
+#define __LIBC_ATTRS
 #endif
 
 #pragma omp begin declare target
 
-// The LLVM C library uses these named types so we forward declare them.
-typedef void (*__atexithandler_t)(void);
-typedef int (*__search_compare_t)(const void *, const void *);
-typedef int (*__qsortcompare_t)(const void *, const void *);
-typedef int (*__qsortrcompare_t)(const void *, const void *, void *);
-
-// Enforce ABI compatibility with the structs used by the LLVM C library.
-_Static_assert(__builtin_offsetof(div_t, quot) == 0, "ABI mismatch!");
-_Static_assert(__builtin_offsetof(ldiv_t, quot) == 0, "ABI mismatch!");
-_Static_assert(__builtin_offsetof(lldiv_t, quot) == 0, "ABI mismatch!");
-
-#if defined(__GLIBC__) && __cplusplus >= 201703L
-#define at_quick_exit atexit
-#endif
-
-#include <llvm-libc-decls/stdlib.h>
-
-#if defined(__GLIBC__) && __cplusplus >= 201703L
-#undef at_quick_exit
-#endif
+// TODO: Define these for CUDA / HIP.
 
 #pragma omp end declare target
 
 #undef __LIBC_ATTRS
 
-#endif
-
 #endif // __CLANG_LLVM_LIBC_WRAPPERS_STDLIB_H__

package/include/llvm_libc_wrappers/string.h

@@ -15,82 +15,14 @@
 
 #include_next <string.h>
 
-#if __has_include(<llvm-libc-decls/string.h>)
-
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
-#endif
-
-#pragma omp begin declare target
-
-// The GNU headers provide C++ standard compliant headers when in C++ mode and
-// the LLVM libc does not. We need to manually provide the definitions using the
-// same prototypes.
-#if defined(__cplusplus) && defined(__GLIBC__) &&                              \
-    defined(__CORRECT_ISO_CPP_STRING_H_PROTO)
-
-#ifndef __LIBC_ATTRS
-#define __LIBC_ATTRS
-#endif
-
-extern "C" {
-void *memccpy(void *__restrict, const void *__restrict, int,
-              size_t) __LIBC_ATTRS;
-int memcmp(const void *, const void *, size_t) __LIBC_ATTRS;
-void *memcpy(void *__restrict, const void *__restrict, size_t) __LIBC_ATTRS;
-void *memmem(const void *, size_t, const void *, size_t) __LIBC_ATTRS;
-void *memmove(void *, const void *, size_t) __LIBC_ATTRS;
-void *mempcpy(void *__restrict, const void *__restrict, size_t) __LIBC_ATTRS;
-void *memset(void *, int, size_t) __LIBC_ATTRS;
-char *stpcpy(char *__restrict, const char *__restrict) __LIBC_ATTRS;
-char *stpncpy(char *__restrict, const char *__restrict, size_t) __LIBC_ATTRS;
-char *strcat(char *__restrict, const char *__restrict) __LIBC_ATTRS;
-int strcmp(const char *, const char *) __LIBC_ATTRS;
-int strcoll(const char *, const char *) __LIBC_ATTRS;
-char *strcpy(char *__restrict, const char *__restrict) __LIBC_ATTRS;
-size_t strcspn(const char *, const char *) __LIBC_ATTRS;
-char *strdup(const char *) __LIBC_ATTRS;
-size_t strlen(const char *) __LIBC_ATTRS;
-char *strncat(char *__restrict, const char *__restrict, size_t) __LIBC_ATTRS;
-int strncmp(const char *, const char *, size_t) __LIBC_ATTRS;
-char *strncpy(char *__restrict, const char *__restrict, size_t) __LIBC_ATTRS;
-char *strndup(const char *, size_t) __LIBC_ATTRS;
-size_t strnlen(const char *, size_t) __LIBC_ATTRS;
-size_t strspn(const char *, const char *) __LIBC_ATTRS;
-char *strtok(char *__restrict, const char *__restrict) __LIBC_ATTRS;
-char *strtok_r(char *__restrict, const char *__restrict,
-               char **__restrict) __LIBC_ATTRS;
-size_t strxfrm(char *__restrict, const char *__restrict, size_t) __LIBC_ATTRS;
-}
-
-extern "C++" {
-char *strstr(char *, const char *) noexcept __LIBC_ATTRS;
-const char *strstr(const char *, const char *) noexcept __LIBC_ATTRS;
-char *strpbrk(char *, const char *) noexcept __LIBC_ATTRS;
-const char *strpbrk(const char *, const char *) noexcept __LIBC_ATTRS;
-char *strrchr(char *, int) noexcept __LIBC_ATTRS;
-const char *strrchr(const char *, int) noexcept __LIBC_ATTRS;
-char *strchr(char *, int) noexcept __LIBC_ATTRS;
-const char *strchr(const char *, int) noexcept __LIBC_ATTRS;
-char *strchrnul(char *, int) noexcept __LIBC_ATTRS;
-const char *strchrnul(const char *, int) noexcept __LIBC_ATTRS;
-char *strcasestr(char *, const char *) noexcept __LIBC_ATTRS;
-const char *strcasestr(const char *, const char *) noexcept __LIBC_ATTRS;
-void *memrchr(void *__s, int __c, size_t __n) noexcept __LIBC_ATTRS;
-const void *memrchr(const void *__s, int __c, size_t __n) noexcept __LIBC_ATTRS;
-void *memchr(void *__s, int __c, size_t __n) noexcept __LIBC_ATTRS;
-const void *memchr(const void *__s, int __c, size_t __n) noexcept __LIBC_ATTRS;
-}
-
 #else
-#include <llvm-libc-decls/string.h>
-
+#define __LIBC_ATTRS
 #endif
 
-#pragma omp end declare target
+// TODO: Define these for CUDA / HIP.
 
 #undef __LIBC_ATTRS
 
-#endif
-
 #endif // __CLANG_LLVM_LIBC_WRAPPERS_STRING_H__

package/include/llvm_libc_wrappers/time.h

@@ -15,20 +15,14 @@
 
 #include_next <time.h>
 
-#if __has_include(<llvm-libc-decls/time.h>)
-
 #if defined(__HIP__) || defined(__CUDA__)
 #define __LIBC_ATTRS __attribute__((device))
+#else
+#define __LIBC_ATTRS
 #endif
 
-#pragma omp begin declare target
-
-_Static_assert(sizeof(clock_t) == sizeof(long), "ABI mismatch!");
-
-#include <llvm-libc-decls/time.h>
+// TODO: Define these for CUDA / HIP.
 
-#pragma omp end declare target
-
-#endif
+#undef __LIBC_ATTRS
 
 #endif // __CLANG_LLVM_LIBC_WRAPPERS_TIME_H__