[WRAPPER] Add wrapper for libmvec.so.1

CMakeLists.txt

@@ -668,6 +668,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedlibice.c"
     "${BOX64_ROOT}/src/wrapped/wrappedliblapack.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibm.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibmvec.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibncurses.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibncurses6.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibncursesw.c"

src/library_list.h

@@ -406,6 +406,7 @@ GO("libandroid-shmem.so", androidshmem)
 
 GO("libc.so.6", libc)
 GO("libm.so.6", libm)
+GO("libmvec.so.1", libmvec)
 GO("libdl.so.2", libdl)
 GO("libpthread.so.0", libpthread)
 GO("libcups.so.2", libcups)

src/wrapped/generated/functions_list.txt

@@ -5992,6 +5992,10 @@ wrappedlibm:
   - __pow_finite
 - DFDD:
   - __powl_finite
+wrappedlibmvec:
+- vFv:
+  - _ZGVbN2v_cos
+  - _ZGVbN2v_sin
 wrappedlibncurses:
 - pFv:
   - initscr

src/wrapped/generated/wrappedlibmvecdefs.h

@@ -0,0 +1,8 @@
+/*******************************************************************
+ * File automatically generated by rebuild_wrappers.py (v2.5.0.24) *
+ *******************************************************************/
+#ifndef __wrappedlibmvecDEFS_H_
+#define __wrappedlibmvecDEFS_H_
+
+
+#endif // __wrappedlibmvecDEFS_H_

src/wrapped/generated/wrappedlibmvectypes.h

@@ -0,0 +1,20 @@
+/*******************************************************************
+ * File automatically generated by rebuild_wrappers.py (v2.5.0.24) *
+ *******************************************************************/
+#ifndef __wrappedlibmvecTYPES_H_
+#define __wrappedlibmvecTYPES_H_
+
+#ifndef LIBNAME
+#error You should only #include this file inside a wrapped*.c file
+#endif
+#ifndef ADDED_FUNCTIONS
+#define ADDED_FUNCTIONS() 
+#endif
+
+typedef void (*vFv_t)(void);
+
+#define SUPER() ADDED_FUNCTIONS() \
+	GO(_ZGVbN2v_cos, vFv_t) \
+	GO(_ZGVbN2v_sin, vFv_t)
+
+#endif // __wrappedlibmvecTYPES_H_

src/wrapped/generated/wrappedlibmvecundefs.h

@@ -0,0 +1,8 @@
+/*******************************************************************
+ * File automatically generated by rebuild_wrappers.py (v2.5.0.24) *
+ *******************************************************************/
+#ifndef __wrappedlibmvecUNDEFS_H_
+#define __wrappedlibmvecUNDEFS_H_
+
+
+#endif // __wrappedlibmvecUNDEFS_H_

src/wrapped/wrappedlibmvec.c

@@ -0,0 +1,86 @@
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <dlfcn.h>
+#include <math.h>
+
+#include "wrappedlibs.h"
+
+#include "wrapper.h"
+#include "bridge.h"
+#include "librarian/library_private.h"
+#include "x64emu.h"
+#include "debug.h"
+#include "emu/x64emu_private.h"
+
+const char* libmvecName = "libmvec.so.1";
+#define LIBNAME libmvec
+
+// GCC vector type: 2 packed doubles in 128 bits.
+// Matches x86 SSE2 __m128d and aarch64 NEON float64x2_t layout.
+typedef double v2df __attribute__((vector_size(16)));
+typedef v2df (*v2df_func_t)(v2df);
+
+static v2df_func_t native_v2d_sin = NULL;
+static v2df_func_t native_v2d_cos = NULL;
+
+// Native libmvec symbol names per architecture.
+// aarch64 NEON: _ZGVnN2v_* (n=NEON, N=no mask, 2=two doubles, v=vector arg)
+// x86_64 SSE2:  _ZGVbN2v_* (b=SSE,  N=no mask, 2=two doubles, v=vector arg)
+#ifdef ARM64
+#define NATIVE_SIN_NAME "_ZGVnN2v_sin"
+#define NATIVE_COS_NAME "_ZGVnN2v_cos"
+#else
+#define NATIVE_SIN_NAME "_ZGVbN2v_sin"
+#define NATIVE_COS_NAME "_ZGVbN2v_cos"
+#endif
+
+// _ZGVbN2v_sin: SSE2 vectorized sin(__m128d) -> __m128d
+// Input: xmm0 = {double, double}  (2 packed doubles)
+// Output: xmm0 = {sin(d0), sin(d1)}
+EXPORT void my__ZGVbN2v_sin(x64emu_t* emu)
+{
+    if (native_v2d_sin) {
+        v2df input, result;
+        memcpy(&input, &emu->xmm[0], sizeof(v2df));
+        result = native_v2d_sin(input);
+        memcpy(&emu->xmm[0], &result, sizeof(v2df));
+    } else {
+        emu->xmm[0].d[0] = sin(emu->xmm[0].d[0]);
+        emu->xmm[0].d[1] = sin(emu->xmm[0].d[1]);
+    }
+}
+
+// _ZGVbN2v_cos: SSE2 vectorized cos(__m128d) -> __m128d
+// Input: xmm0 = {double, double}  (2 packed doubles)
+// Output: xmm0 = {cos(d0), cos(d1)}
+EXPORT void my__ZGVbN2v_cos(x64emu_t* emu)
+{
+    if (native_v2d_cos) {
+        v2df input, result;
+        memcpy(&input, &emu->xmm[0], sizeof(v2df));
+        result = native_v2d_cos(input);
+        memcpy(&emu->xmm[0], &result, sizeof(v2df));
+    } else {
+        emu->xmm[0].d[0] = cos(emu->xmm[0].d[0]);
+        emu->xmm[0].d[1] = cos(emu->xmm[0].d[1]);
+    }
+}
+
+// Try to load the native libmvec.so.1 and resolve vectorized sin/cos.
+// On aarch64 and x86_64: native libmvec exists in glibc, use it.
+// On ppc64le and others: dlopen fails, fall back to scalar sin/cos.
+#define PRE_INIT                                                          \
+    if (1) {                                                              \
+        void* native = dlopen("libmvec.so.1", RTLD_LAZY);                 \
+        if (native) {                                                     \
+            lib->w.lib = native;                                          \
+            native_v2d_sin = (v2df_func_t)dlsym(native, NATIVE_SIN_NAME); \
+            native_v2d_cos = (v2df_func_t)dlsym(native, NATIVE_COS_NAME); \
+        } else {                                                          \
+            lib->w.lib = dlopen(NULL, RTLD_LAZY | RTLD_GLOBAL);           \
+        }                                                                 \
+    } else
+
+#include "wrappedlib_init.h"

src/wrapped/wrappedlibmvec_private.h

@@ -0,0 +1,235 @@
+#if !(defined(GO) && defined(GOM) && defined(GO2) && defined(DATA))
+#error meh!
+#endif
+
+// SSE2 2x double (b = SSE, N2 = 2 elements, v = vector)
+//GO(_ZGVbN2v_acos, vFv)
+//GO(_ZGVbN2v_acosh, vFv)
+//GO(_ZGVbN2v_asin, vFv)
+//GO(_ZGVbN2v_asinh, vFv)
+//GO(_ZGVbN2v_atan, vFv)
+//GO(_ZGVbN2v_atanh, vFv)
+//GO(_ZGVbN2v_cbrt, vFv)
+GOM(_ZGVbN2v_cos, vFEv)
+//GO(_ZGVbN2v_cosh, vFv)
+//GO(_ZGVbN2v_erf, vFv)
+//GO(_ZGVbN2v_erfc, vFv)
+//GO(_ZGVbN2v_exp, vFv)
+//GO(_ZGVbN2v_exp10, vFv)
+//GO(_ZGVbN2v_exp2, vFv)
+//GO(_ZGVbN2v_expm1, vFv)
+//GO(_ZGVbN2v_log, vFv)
+//GO(_ZGVbN2v_log10, vFv)
+//GO(_ZGVbN2v_log1p, vFv)
+//GO(_ZGVbN2v_log2, vFv)
+GOM(_ZGVbN2v_sin, vFEv)
+//GO(_ZGVbN2v_sinh, vFv)
+//GO(_ZGVbN2v_tan, vFv)
+//GO(_ZGVbN2v_tanh, vFv)
+//GO(_ZGVbN2vv_atan2, vFv)
+//GO(_ZGVbN2vv_hypot, vFv)
+//GO(_ZGVbN2vv_pow, vFv)
+//GO(_ZGVbN2vvv_sincos, vFv)
+
+// SSE2 4x float (b = SSE, N4 = 4 elements, v = vector)
+//GO(_ZGVbN4v_acosf, vFv)
+//GO(_ZGVbN4v_acoshf, vFv)
+//GO(_ZGVbN4v_asinf, vFv)
+//GO(_ZGVbN4v_asinhf, vFv)
+//GO(_ZGVbN4v_atanf, vFv)
+//GO(_ZGVbN4v_atanhf, vFv)
+//GO(_ZGVbN4v_cbrtf, vFv)
+//GO(_ZGVbN4v_cosf, vFv)
+//GO(_ZGVbN4v_coshf, vFv)
+//GO(_ZGVbN4v_erfcf, vFv)
+//GO(_ZGVbN4v_erff, vFv)
+//GO(_ZGVbN4v_exp10f, vFv)
+//GO(_ZGVbN4v_exp2f, vFv)
+//GO(_ZGVbN4v_expf, vFv)
+//GO(_ZGVbN4v_expm1f, vFv)
+//GO(_ZGVbN4v_log10f, vFv)
+//GO(_ZGVbN4v_log1pf, vFv)
+//GO(_ZGVbN4v_log2f, vFv)
+//GO(_ZGVbN4v_logf, vFv)
+//GO(_ZGVbN4v_sinf, vFv)
+//GO(_ZGVbN4v_sinhf, vFv)
+//GO(_ZGVbN4v_tanf, vFv)
+//GO(_ZGVbN4v_tanhf, vFv)
+//GO(_ZGVbN4vv_atan2f, vFv)
+//GO(_ZGVbN4vv_hypotf, vFv)
+//GO(_ZGVbN4vv_powf, vFv)
+//GO(_ZGVbN4vvv_sincosf, vFv)
+
+// AVX 4x double (c = AVX, N4 = 4 elements, v = vector)
+//GO(_ZGVcN4v_acos, vFv)
+//GO(_ZGVcN4v_acosh, vFv)
+//GO(_ZGVcN4v_asin, vFv)
+//GO(_ZGVcN4v_asinh, vFv)
+//GO(_ZGVcN4v_atan, vFv)
+//GO(_ZGVcN4v_atanh, vFv)
+//GO(_ZGVcN4v_cbrt, vFv)
+//GO(_ZGVcN4v_cos, vFv)
+//GO(_ZGVcN4v_cosh, vFv)
+//GO(_ZGVcN4v_erf, vFv)
+//GO(_ZGVcN4v_erfc, vFv)
+//GO(_ZGVcN4v_exp, vFv)
+//GO(_ZGVcN4v_exp10, vFv)
+//GO(_ZGVcN4v_exp2, vFv)
+//GO(_ZGVcN4v_expm1, vFv)
+//GO(_ZGVcN4v_log, vFv)
+//GO(_ZGVcN4v_log10, vFv)
+//GO(_ZGVcN4v_log1p, vFv)
+//GO(_ZGVcN4v_log2, vFv)
+//GO(_ZGVcN4v_sin, vFv)
+//GO(_ZGVcN4v_sinh, vFv)
+//GO(_ZGVcN4v_tan, vFv)
+//GO(_ZGVcN4v_tanh, vFv)
+//GO(_ZGVcN4vv_atan2, vFv)
+//GO(_ZGVcN4vv_hypot, vFv)
+//GO(_ZGVcN4vv_pow, vFv)
+//GO(_ZGVcN4vvv_sincos, vFv)
+
+// AVX 8x float (c = AVX, N8 = 8 elements, v = vector)
+//GO(_ZGVcN8v_acosf, vFv)
+//GO(_ZGVcN8v_acoshf, vFv)
+//GO(_ZGVcN8v_asinf, vFv)
+//GO(_ZGVcN8v_asinhf, vFv)
+//GO(_ZGVcN8v_atanf, vFv)
+//GO(_ZGVcN8v_atanhf, vFv)
+//GO(_ZGVcN8v_cbrtf, vFv)
+//GO(_ZGVcN8v_cosf, vFv)
+//GO(_ZGVcN8v_coshf, vFv)
+//GO(_ZGVcN8v_erfcf, vFv)
+//GO(_ZGVcN8v_erff, vFv)
+//GO(_ZGVcN8v_exp10f, vFv)
+//GO(_ZGVcN8v_exp2f, vFv)
+//GO(_ZGVcN8v_expf, vFv)
+//GO(_ZGVcN8v_expm1f, vFv)
+//GO(_ZGVcN8v_log10f, vFv)
+//GO(_ZGVcN8v_log1pf, vFv)
+//GO(_ZGVcN8v_log2f, vFv)
+//GO(_ZGVcN8v_logf, vFv)
+//GO(_ZGVcN8v_sinf, vFv)
+//GO(_ZGVcN8v_sinhf, vFv)
+//GO(_ZGVcN8v_tanf, vFv)
+//GO(_ZGVcN8v_tanhf, vFv)
+//GO(_ZGVcN8vv_atan2f, vFv)
+//GO(_ZGVcN8vv_hypotf, vFv)
+//GO(_ZGVcN8vv_powf, vFv)
+//GO(_ZGVcN8vvv_sincosf, vFv)
+
+// AVX2 4x double (d = AVX2, N4 = 4 elements, v = vector)
+//GO(_ZGVdN4v_acos, vFv)
+//GO(_ZGVdN4v_acosh, vFv)
+//GO(_ZGVdN4v_asin, vFv)
+//GO(_ZGVdN4v_asinh, vFv)
+//GO(_ZGVdN4v_atan, vFv)
+//GO(_ZGVdN4v_atanh, vFv)
+//GO(_ZGVdN4v_cbrt, vFv)
+//GO(_ZGVdN4v_cos, vFv)
+//GO(_ZGVdN4v_cosh, vFv)
+//GO(_ZGVdN4v_erf, vFv)
+//GO(_ZGVdN4v_erfc, vFv)
+//GO(_ZGVdN4v_exp, vFv)
+//GO(_ZGVdN4v_exp10, vFv)
+//GO(_ZGVdN4v_exp2, vFv)
+//GO(_ZGVdN4v_expm1, vFv)
+//GO(_ZGVdN4v_log, vFv)
+//GO(_ZGVdN4v_log10, vFv)
+//GO(_ZGVdN4v_log1p, vFv)
+//GO(_ZGVdN4v_log2, vFv)
+//GO(_ZGVdN4v_sin, vFv)
+//GO(_ZGVdN4v_sinh, vFv)
+//GO(_ZGVdN4v_tan, vFv)
+//GO(_ZGVdN4v_tanh, vFv)
+//GO(_ZGVdN4vv_atan2, vFv)
+//GO(_ZGVdN4vv_hypot, vFv)
+//GO(_ZGVdN4vv_pow, vFv)
+//GO(_ZGVdN4vvv_sincos, vFv)
+
+// AVX2 8x float (d = AVX2, N8 = 8 elements, v = vector)
+//GO(_ZGVdN8v_acosf, vFv)
+//GO(_ZGVdN8v_acoshf, vFv)
+//GO(_ZGVdN8v_asinf, vFv)
+//GO(_ZGVdN8v_asinhf, vFv)
+//GO(_ZGVdN8v_atanf, vFv)
+//GO(_ZGVdN8v_atanhf, vFv)
+//GO(_ZGVdN8v_cbrtf, vFv)
+//GO(_ZGVdN8v_cosf, vFv)
+//GO(_ZGVdN8v_coshf, vFv)
+//GO(_ZGVdN8v_erfcf, vFv)
+//GO(_ZGVdN8v_erff, vFv)
+//GO(_ZGVdN8v_exp10f, vFv)
+//GO(_ZGVdN8v_exp2f, vFv)
+//GO(_ZGVdN8v_expf, vFv)
+//GO(_ZGVdN8v_expm1f, vFv)
+//GO(_ZGVdN8v_log10f, vFv)
+//GO(_ZGVdN8v_log1pf, vFv)
+//GO(_ZGVdN8v_log2f, vFv)
+//GO(_ZGVdN8v_logf, vFv)
+//GO(_ZGVdN8v_sinf, vFv)
+//GO(_ZGVdN8v_sinhf, vFv)
+//GO(_ZGVdN8v_tanf, vFv)
+//GO(_ZGVdN8v_tanhf, vFv)
+//GO(_ZGVdN8vv_atan2f, vFv)
+//GO(_ZGVdN8vv_hypotf, vFv)
+//GO(_ZGVdN8vv_powf, vFv)
+//GO(_ZGVdN8vvv_sincosf, vFv)
+
+// AVX-512 8x double (e = AVX-512, N8 = 8 elements, v = vector)
+//GO(_ZGVeN8v_acos, vFv)
+//GO(_ZGVeN8v_acosh, vFv)
+//GO(_ZGVeN8v_asin, vFv)
+//GO(_ZGVeN8v_asinh, vFv)
+//GO(_ZGVeN8v_atan, vFv)
+//GO(_ZGVeN8v_atanh, vFv)
+//GO(_ZGVeN8v_cbrt, vFv)
+//GO(_ZGVeN8v_cos, vFv)
+//GO(_ZGVeN8v_cosh, vFv)
+//GO(_ZGVeN8v_erf, vFv)
+//GO(_ZGVeN8v_erfc, vFv)
+//GO(_ZGVeN8v_exp, vFv)
+//GO(_ZGVeN8v_exp10, vFv)
+//GO(_ZGVeN8v_exp2, vFv)
+//GO(_ZGVeN8v_expm1, vFv)
+//GO(_ZGVeN8v_log, vFv)
+//GO(_ZGVeN8v_log10, vFv)
+//GO(_ZGVeN8v_log1p, vFv)
+//GO(_ZGVeN8v_log2, vFv)
+//GO(_ZGVeN8v_sin, vFv)
+//GO(_ZGVeN8v_sinh, vFv)
+//GO(_ZGVeN8v_tan, vFv)
+//GO(_ZGVeN8v_tanh, vFv)
+//GO(_ZGVeN8vv_atan2, vFv)
+//GO(_ZGVeN8vv_hypot, vFv)
+//GO(_ZGVeN8vv_pow, vFv)
+//GO(_ZGVeN8vvv_sincos, vFv)
+
+// AVX-512 16x float (e = AVX-512, N16 = 16 elements, v = vector)
+//GO(_ZGVeN16v_acosf, vFv)
+//GO(_ZGVeN16v_acoshf, vFv)
+//GO(_ZGVeN16v_asinf, vFv)
+//GO(_ZGVeN16v_asinhf, vFv)
+//GO(_ZGVeN16v_atanf, vFv)
+//GO(_ZGVeN16v_atanhf, vFv)
+//GO(_ZGVeN16v_cbrtf, vFv)
+//GO(_ZGVeN16v_cosf, vFv)
+//GO(_ZGVeN16v_coshf, vFv)
+//GO(_ZGVeN16v_erfcf, vFv)
+//GO(_ZGVeN16v_erff, vFv)
+//GO(_ZGVeN16v_exp10f, vFv)
+//GO(_ZGVeN16v_exp2f, vFv)
+//GO(_ZGVeN16v_expf, vFv)
+//GO(_ZGVeN16v_expm1f, vFv)
+//GO(_ZGVeN16v_log10f, vFv)
+//GO(_ZGVeN16v_log1pf, vFv)
+//GO(_ZGVeN16v_log2f, vFv)
+//GO(_ZGVeN16v_logf, vFv)
+//GO(_ZGVeN16v_sinf, vFv)
+//GO(_ZGVeN16v_sinhf, vFv)
+//GO(_ZGVeN16v_tanf, vFv)
+//GO(_ZGVeN16v_tanhf, vFv)
+//GO(_ZGVeN16vv_atan2f, vFv)
+//GO(_ZGVeN16vv_hypotf, vFv)
+//GO(_ZGVeN16vv_powf, vFv)
+//GO(_ZGVeN16vvv_sincosf, vFv)