diff --git a/Modules/blake2module.c b/Modules/blake2module.c
index 07aa89f573f05f..ae37e2d3383f9b 100644
--- a/Modules/blake2module.c
+++ b/Modules/blake2module.c
@@ -43,12 +43,11 @@
// SIMD256 can't be compiled on macOS ARM64, and performance of SIMD128 isn't
// great; but when compiling a universal2 binary, autoconf will set
-// HACL_CAN_COMPILE_SIMD128 and HACL_CAN_COMPILE_SIMD256 because they *can* be
-// compiled on x86_64. If we're on macOS ARM64, disable these preprocessor
-// symbols.
+// _Py_HACL_CAN_COMPILE_VEC{128,256} because they *can* be compiled on x86_64.
+// If we're on macOS ARM64, we however disable these preprocessor symbols.
#if defined(__APPLE__) && defined(__arm64__)
-# undef HACL_CAN_COMPILE_SIMD128
-# undef HACL_CAN_COMPILE_SIMD256
+# undef _Py_HACL_CAN_COMPILE_VEC128
+# undef _Py_HACL_CAN_COMPILE_VEC256
#endif
// ECX
@@ -114,31 +113,32 @@ void detect_cpu_features(cpu_flags *flags) {
}
}
-#ifdef HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
static inline bool has_simd128(cpu_flags *flags) {
- // For now this is Intel-only, could conceivably be #ifdef'd to something
+ // For now this is Intel-only, could conceivably be if'd to something
// else.
return flags->sse && flags->sse2 && flags->sse3 && flags->sse41 && flags->sse42 && flags->cmov;
}
#endif
-#ifdef HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
static inline bool has_simd256(cpu_flags *flags) {
return flags->avx && flags->avx2;
}
#endif
-// Small mismatch between the variable names Python defines as part of configure
-// at the ones HACL* expects to be set in order to enable those headers.
-#define HACL_CAN_COMPILE_VEC128 HACL_CAN_COMPILE_SIMD128
-#define HACL_CAN_COMPILE_VEC256 HACL_CAN_COMPILE_SIMD256
+// HACL* expects HACL_CAN_COMPILE_VEC* macros to be set in order to enable
+// the corresponding SIMD instructions so we need to "forward" the values
+// we just deduced above.
+#define HACL_CAN_COMPILE_VEC128 _Py_HACL_CAN_COMPILE_VEC128
+#define HACL_CAN_COMPILE_VEC256 _Py_HACL_CAN_COMPILE_VEC256
#include "_hacl/Hacl_Hash_Blake2b.h"
#include "_hacl/Hacl_Hash_Blake2s.h"
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
#include "_hacl/Hacl_Hash_Blake2b_Simd256.h"
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
#include "_hacl/Hacl_Hash_Blake2s_Simd128.h"
#endif
@@ -165,7 +165,7 @@ blake2_get_state(PyObject *module)
return (Blake2State *)state;
}
-#if defined(HACL_CAN_COMPILE_SIMD128) || defined(HACL_CAN_COMPILE_SIMD256)
+#if defined(_Py_HACL_CAN_COMPILE_VEC128) || defined(_Py_HACL_CAN_COMPILE_VEC256)
static inline Blake2State*
blake2_get_state_from_type(PyTypeObject *module)
{
@@ -329,18 +329,18 @@ static inline bool is_blake2s(blake2_impl impl) {
}
static inline blake2_impl type_to_impl(PyTypeObject *type) {
-#if defined(HACL_CAN_COMPILE_SIMD128) || defined(HACL_CAN_COMPILE_SIMD256)
+#if defined(_Py_HACL_CAN_COMPILE_VEC128) || defined(_Py_HACL_CAN_COMPILE_VEC256)
Blake2State* st = blake2_get_state_from_type(type);
#endif
if (!strcmp(type->tp_name, blake2b_type_spec.name)) {
-#ifdef HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
if (has_simd256(&st->flags))
return Blake2b_256;
else
#endif
return Blake2b;
} else if (!strcmp(type->tp_name, blake2s_type_spec.name)) {
-#ifdef HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
if (has_simd128(&st->flags))
return Blake2s_128;
else
@@ -356,10 +356,10 @@ typedef struct {
union {
Hacl_Hash_Blake2s_state_t *blake2s_state;
Hacl_Hash_Blake2b_state_t *blake2b_state;
-#ifdef HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
Hacl_Hash_Blake2s_Simd128_state_t *blake2s_128_state;
#endif
-#ifdef HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
Hacl_Hash_Blake2b_Simd256_state_t *blake2b_256_state;
#endif
};
@@ -425,14 +425,14 @@ static void
update(Blake2Object *self, uint8_t *buf, Py_ssize_t len)
{
switch (self->impl) {
- // These need to be ifdef'd out otherwise it's an unresolved symbol at
- // link-time.
-#ifdef HACL_CAN_COMPILE_SIMD256
+ // blake2b_256_state and blake2s_128_state must be if'd since
+ // otherwise this results in an unresolved symbol at link-time.
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256:
HACL_UPDATE(Hacl_Hash_Blake2b_Simd256_update,self->blake2b_256_state, buf, len);
return;
#endif
-#ifdef HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128:
HACL_UPDATE(Hacl_Hash_Blake2s_Simd128_update,self->blake2s_128_state, buf, len);
return;
@@ -468,12 +468,12 @@ py_blake2b_or_s_new(PyTypeObject *type, PyObject *data, int digest_size,
// Ensure that the states are NULL-initialized in case of an error.
// See: py_blake2_clear() for more details.
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256:
self->blake2b_256_state = NULL;
break;
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128:
self->blake2s_128_state = NULL;
break;
@@ -591,7 +591,7 @@ py_blake2b_or_s_new(PyTypeObject *type, PyObject *data, int digest_size,
};
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256: {
self->blake2b_256_state = Hacl_Hash_Blake2b_Simd256_malloc_with_params_and_key(¶ms, last_node, key->buf);
if (self->blake2b_256_state == NULL) {
@@ -601,7 +601,7 @@ py_blake2b_or_s_new(PyTypeObject *type, PyObject *data, int digest_size,
break;
}
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128: {
self->blake2s_128_state = Hacl_Hash_Blake2s_Simd128_malloc_with_params_and_key(¶ms, last_node, key->buf);
if (self->blake2s_128_state == NULL) {
@@ -733,7 +733,7 @@ blake2_blake2b_copy_locked(Blake2Object *self, Blake2Object *cpy)
{
assert(cpy != NULL);
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256: {
cpy->blake2b_256_state = Hacl_Hash_Blake2b_Simd256_copy(self->blake2b_256_state);
if (cpy->blake2b_256_state == NULL) {
@@ -742,7 +742,7 @@ blake2_blake2b_copy_locked(Blake2Object *self, Blake2Object *cpy)
break;
}
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128: {
cpy->blake2s_128_state = Hacl_Hash_Blake2s_Simd128_copy(self->blake2s_128_state);
if (cpy->blake2s_128_state == NULL) {
@@ -853,12 +853,12 @@ _blake2_blake2b_digest_impl(Blake2Object *self)
ENTER_HASHLIB(self);
uint8_t digest_length = 0;
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256:
digest_length = Hacl_Hash_Blake2b_Simd256_digest(self->blake2b_256_state, digest);
break;
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128:
digest_length = Hacl_Hash_Blake2s_Simd128_digest(self->blake2s_128_state, digest);
break;
@@ -891,12 +891,12 @@ _blake2_blake2b_hexdigest_impl(Blake2Object *self)
ENTER_HASHLIB(self);
uint8_t digest_length = 0;
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256:
digest_length = Hacl_Hash_Blake2b_Simd256_digest(self->blake2b_256_state, digest);
break;
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128:
digest_length = Hacl_Hash_Blake2s_Simd128_digest(self->blake2s_128_state, digest);
break;
@@ -947,11 +947,11 @@ py_blake2b_get_digest_size(PyObject *op, void *Py_UNUSED(closure))
{
Blake2Object *self = _Blake2Object_CAST(op);
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256:
return PyLong_FromLong(Hacl_Hash_Blake2b_Simd256_info(self->blake2b_256_state).digest_length);
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128:
return PyLong_FromLong(Hacl_Hash_Blake2s_Simd128_info(self->blake2s_128_state).digest_length);
#endif
@@ -982,7 +982,7 @@ py_blake2_clear(PyObject *op)
// it. If an error occurs in the constructor, we should only free
// states that were allocated (i.e. that are not NULL).
switch (self->impl) {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
case Blake2b_256:
if (self->blake2b_256_state != NULL) {
Hacl_Hash_Blake2b_Simd256_free(self->blake2b_256_state);
@@ -990,7 +990,7 @@ py_blake2_clear(PyObject *op)
}
break;
#endif
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
case Blake2s_128:
if (self->blake2s_128_state != NULL) {
Hacl_Hash_Blake2s_Simd128_free(self->blake2s_128_state);
diff --git a/Modules/hmacmodule.c b/Modules/hmacmodule.c
index 059e15ba0d35fd..8cd470f4f80b3a 100644
--- a/Modules/hmacmodule.c
+++ b/Modules/hmacmodule.c
@@ -31,14 +31,15 @@
#endif
#if defined(__APPLE__) && defined(__arm64__)
-# undef HACL_CAN_COMPILE_SIMD128
-# undef HACL_CAN_COMPILE_SIMD256
+# undef _Py_HACL_CAN_COMPILE_VEC128
+# undef _Py_HACL_CAN_COMPILE_VEC256
#endif
-// Small mismatch between the variable names Python defines as part of configure
-// at the ones HACL* expects to be set in order to enable those headers.
-#define HACL_CAN_COMPILE_VEC128 HACL_CAN_COMPILE_SIMD128
-#define HACL_CAN_COMPILE_VEC256 HACL_CAN_COMPILE_SIMD256
+// HACL* expects HACL_CAN_COMPILE_VEC* macros to be set in order to enable
+// the corresponding SIMD instructions so we need to "forward" the values
+// we just deduced above.
+#define HACL_CAN_COMPILE_VEC128 _Py_HACL_CAN_COMPILE_VEC128
+#define HACL_CAN_COMPILE_VEC256 _Py_HACL_CAN_COMPILE_VEC256
#include "_hacl/Hacl_HMAC.h"
#include "_hacl/Hacl_Streaming_HMAC.h" // Hacl_Agile_Hash_* identifiers
@@ -464,7 +465,7 @@ narrow_hmac_hash_kind(hmacmodule_state *state, HMAC_Hash_Kind kind)
{
switch (kind) {
case Py_hmac_kind_hmac_blake2s_32: {
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
if (state->can_run_simd128) {
return Py_hmac_kind_hmac_vectorized_blake2s_32;
}
@@ -472,7 +473,7 @@ narrow_hmac_hash_kind(hmacmodule_state *state, HMAC_Hash_Kind kind)
return kind;
}
case Py_hmac_kind_hmac_blake2b_32: {
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
if (state->can_run_simd256) {
return Py_hmac_kind_hmac_vectorized_blake2b_32;
}
@@ -1761,7 +1762,7 @@ hmacmodule_init_cpu_features(hmacmodule_state *state)
#undef ECX_SSE3
#undef EBX_AVX2
-#if HACL_CAN_COMPILE_SIMD128
+#if _Py_HACL_CAN_COMPILE_VEC128
// TODO(picnixz): use py_cpuid_features (gh-125022) to improve detection
state->can_run_simd128 = sse && sse2 && sse3 && sse41 && sse42 && cmov;
#else
@@ -1771,7 +1772,7 @@ hmacmodule_init_cpu_features(hmacmodule_state *state)
state->can_run_simd128 = false;
#endif
-#if HACL_CAN_COMPILE_SIMD256
+#if _Py_HACL_CAN_COMPILE_VEC256
// TODO(picnixz): use py_cpuid_features (gh-125022) to improve detection
state->can_run_simd256 = state->can_run_simd128 && avx && avx2;
#else
diff --git a/PCbuild/pythoncore.vcxproj b/PCbuild/pythoncore.vcxproj
index 32a8f2dbad3d5e..b911c9385634d7 100644
--- a/PCbuild/pythoncore.vcxproj
+++ b/PCbuild/pythoncore.vcxproj
@@ -419,8 +419,12 @@
- HACL_CAN_COMPILE_SIMD128;%(PreprocessorDefinitions)
- HACL_CAN_COMPILE_SIMD256;%(PreprocessorDefinitions)
+
+ _Py_HACL_CAN_COMPILE_VEC128;%(PreprocessorDefinitions)
+
+
+ _Py_HACL_CAN_COMPILE_VEC256;%(PreprocessorDefinitions)
+
diff --git a/configure b/configure
index 47b6c0ebc88a01..cc2636348ad4cc 100755
--- a/configure
+++ b/configure
@@ -32625,7 +32625,7 @@ then :
LIBHACL_SIMD128_FLAGS="-msse -msse2 -msse3 -msse4.1 -msse4.2"
-printf "%s\n" "#define HACL_CAN_COMPILE_SIMD128 1" >>confdefs.h
+printf "%s\n" "#define _Py_HACL_CAN_COMPILE_VEC128 1" >>confdefs.h
# macOS universal2 builds *support* the -msse etc flags because they're
@@ -32701,7 +32701,7 @@ then :
LIBHACL_SIMD256_FLAGS="-mavx2"
-printf "%s\n" "#define HACL_CAN_COMPILE_SIMD256 1" >>confdefs.h
+printf "%s\n" "#define _Py_HACL_CAN_COMPILE_VEC256 1" >>confdefs.h
# macOS universal2 builds *support* the -mavx2 compiler flag because it's
diff --git a/configure.ac b/configure.ac
index a4a747180bfdd1..912f77b9ae2cc5 100644
--- a/configure.ac
+++ b/configure.ac
@@ -8017,7 +8017,8 @@ then
AX_CHECK_COMPILE_FLAG([-msse -msse2 -msse3 -msse4.1 -msse4.2],[
[LIBHACL_SIMD128_FLAGS="-msse -msse2 -msse3 -msse4.1 -msse4.2"]
- AC_DEFINE([HACL_CAN_COMPILE_SIMD128], [1], [HACL* library can compile SIMD128 implementations])
+ AC_DEFINE([_Py_HACL_CAN_COMPILE_VEC128], [1], [
+ HACL* library can compile SIMD128 implementations])
# macOS universal2 builds *support* the -msse etc flags because they're
# available on x86_64. However, performance of the HACL SIMD128 implementation
@@ -8048,7 +8049,8 @@ if test "$ac_sys_system" != "Linux-android" -a "$ac_sys_system" != "WASI" || \
then
AX_CHECK_COMPILE_FLAG([-mavx2],[
[LIBHACL_SIMD256_FLAGS="-mavx2"]
- AC_DEFINE([HACL_CAN_COMPILE_SIMD256], [1], [HACL* library can compile SIMD256 implementations])
+ AC_DEFINE([_Py_HACL_CAN_COMPILE_VEC256], [1], [
+ HACL* library can compile SIMD256 implementations])
# macOS universal2 builds *support* the -mavx2 compiler flag because it's
# available on x86_64; but the HACL SIMD256 build then fails because the
diff --git a/pyconfig.h.in b/pyconfig.h.in
index d4f1da7fb10776..d7c496fccc682c 100644
--- a/pyconfig.h.in
+++ b/pyconfig.h.in
@@ -50,12 +50,6 @@
/* Define if getpgrp() must be called as getpgrp(0). */
#undef GETPGRP_HAVE_ARG
-/* HACL* library can compile SIMD128 implementations */
-#undef HACL_CAN_COMPILE_SIMD128
-
-/* HACL* library can compile SIMD256 implementations */
-#undef HACL_CAN_COMPILE_SIMD256
-
/* Define if you have the 'accept' function. */
#undef HAVE_ACCEPT
@@ -2026,6 +2020,12 @@
/* Defined if _Complex C type can be used with libffi. */
#undef _Py_FFI_SUPPORT_C_COMPLEX
+/* HACL* library can compile SIMD128 implementations */
+#undef _Py_HACL_CAN_COMPILE_VEC128
+
+/* HACL* library can compile SIMD256 implementations */
+#undef _Py_HACL_CAN_COMPILE_VEC256
+
/* Define to force use of thread-safe errno, h_errno, and other functions */
#undef _REENTRANT