AuroraMiddleware/gtk
1
0
Fork 1
mirror of https://gitlab.gnome.org/GNOME/gtk.git synced 2025-01-01 16:30:15 +00:00
Code Issues Projects Releases Wiki Activity

Merge branch 'f16c.msvc' into 'master'

Browse Source 
Fix F16C build/runtime detection on Visual Studio

See merge request GNOME/gtk!3429
This commit is contained in:
Matthias Clasen 2021-04-12 20:35:10 +00:00
commit 442f485591
2 changed files with 63 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

63
gsk/ngl/fp16.c
View File

@ -80,6 +80,12 @@ half_to_float4_c (const guint16 h[4],
#ifdef HAVE_F16C #ifdef HAVE_F16C
#if defined(_MSC_VER) && !defined(__clang__)
#define CAST_M128I_P(a) (__m128i const *) a
#else
#define CAST_M128I_P(a) (__m128i_u const *) a
#endif
static void static void
float_to_half4_f16c (const float f[4], float_to_half4_f16c (const float f[4],
guint16 h[4]) guint16 h[4])
@ -93,11 +99,61 @@ static void
half_to_float4_f16c (const guint16 h[4], half_to_float4_f16c (const guint16 h[4],
float f[4]) float f[4])
{ {
__m128i i = _mm_loadl_epi64 ((__m128i_u const *)h); __m128i i = _mm_loadl_epi64 (CAST_M128I_P (h));
__m128 s = _mm_cvtph_ps (i); __m128 s = _mm_cvtph_ps (i);
_mm_store_ps (f, s); _mm_store_ps (f, s);
} }
#undef CAST_M128I_P
#if defined(_MSC_VER) && !defined(__clang__)
/* based on info from https://walbourn.github.io/directxmath-f16c-and-fma/ */
static gboolean
have_f16c_msvc (void)
{
static gboolean result = FALSE;
static gsize inited = 0;
if (g_once_init_enter (&inited))
{
int cpuinfo[4] = { -1 };
__cpuid (cpuinfo, 0);
if (cpuinfo[0] > 0)
{
__cpuid (cpuinfo, 1);
if ((cpuinfo[2] & 0x8000000) != 0)
result = (cpuinfo[2] & 0x20000000) != 0;
}
g_once_init_leave (&inited, 1);
}
return result;
}
void
float_to_half4 (const float f[4], guint16 h[4])
{
if (have_f16c_msvc ())
float_to_half4_f16c (f, h);
else
float_to_half4_c (f, h);
}
void
half_to_float4 (const guint16 h[4], float f[4])
{
if (have_f16c_msvc ())
half_to_float4_f16c (h, f);
else
half_to_float4_c (h, f);
}
#else
void float_to_half4 (const float f[4], guint16 h[4]) __attribute__((ifunc ("resolve_float_to_half4"))); void float_to_half4 (const float f[4], guint16 h[4]) __attribute__((ifunc ("resolve_float_to_half4")));
void half_to_float4 (const guint16 h[4], float f[4]) __attribute__((ifunc ("resolve_half_to_float4"))); void half_to_float4 (const guint16 h[4], float f[4]) __attribute__((ifunc ("resolve_half_to_float4")));
@ -120,11 +176,12 @@ resolve_half_to_float4 (void)
else else
return half_to_float4_c; return half_to_float4_c;
} }
#endif
#else #else
#ifdef __APPLE__ #if defined(__APPLE__) || (defined(_MSC_VER) && !defined(__clang__))
// turns out aliases don't work on Darwin // turns out aliases don't work on Darwin nor Visual Studio
void void
float_to_half4 (const float f[4], float_to_half4 (const float f[4],

6
meson.build
View File

@ -692,10 +692,8 @@ if get_option('f16c').enabled()
# if !defined(__amd64__) && !defined(__x86_64__) # if !defined(__amd64__) && !defined(__x86_64__)
# error "F16C intrinsics are only available on x86_64" # error "F16C intrinsics are only available on x86_64"
# endif # endif
#elif defined (_MSC_VER) && !defined (_M_X64) && !defined (_M_AMD64)
# error "F16C intrinsics not supported on x86 MSVC builds"
#endif #endif
#if defined(__SSE__) || (_M_X64 > 0) #if defined(__SSE__) || defined(_MSC_VER)
# include <immintrin.h> # include <immintrin.h>
#else #else
# error "No F16C intrinsics available" # error "No F16C intrinsics available"
@ -707,8 +705,10 @@ int main () {
__m128i i = _mm_cvtps_ph (s, 0); __m128i i = _mm_cvtps_ph (s, 0);
_mm_storel_epi64 ((__m128i*)h, i); _mm_storel_epi64 ((__m128i*)h, i);
#if defined (__GNUC__) || defined (__clang__)
__builtin_cpu_init (); __builtin_cpu_init ();
__builtin_cpu_supports ("f16c"); __builtin_cpu_supports ("f16c");
#endif
return 0; return 0;
}''' }'''