Skip to content

Commit f651866

Browse files
vtjnashvchuravy
vtjnash
authored and
vchuravy
committed
Prefix Float16 intrinsics
1 parent db570df commit f651866

File tree

4 files changed

+52
-38
lines changed

4 files changed

+52
-38
lines changed

‎src/APInt-C.cpp‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -316,7 +316,7 @@ void LLVMByteSwap(unsigned numbits, integerPart *pa, integerPart *pr) {
316316
void LLVMFPtoInt(unsigned numbits, void *pa, unsigned onumbits, integerPart *pr, bool isSigned, bool *isExact) {
317317
double Val;
318318
if (numbits == 16)
319-
Val = __gnu_h2f_ieee(*(uint16_t*)pa);
319+
Val = julia__gnu_h2f_ieee(*(uint16_t*)pa);
320320
else if (numbits == 32)
321321
Val = *(float*)pa;
322322
else if (numbits == 64)
@@ -391,7 +391,7 @@ void LLVMSItoFP(unsigned numbits, integerPart *pa, unsigned onumbits, integerPar
391391
val = a.roundToDouble(true);
392392
}
393393
if (onumbits == 16)
394-
*(uint16_t*)pr = __gnu_f2h_ieee(val);
394+
*(uint16_t*)pr = julia__gnu_f2h_ieee(val);
395395
else if (onumbits == 32)
396396
*(float*)pr = val;
397397
else if (onumbits == 64)
@@ -408,7 +408,7 @@ void LLVMUItoFP(unsigned numbits, integerPart *pa, unsigned onumbits, integerPar
408408
val = a.roundToDouble(false);
409409
}
410410
if (onumbits == 16)
411-
*(uint16_t*)pr = __gnu_f2h_ieee(val);
411+
*(uint16_t*)pr = julia__gnu_f2h_ieee(val);
412412
else if (onumbits == 32)
413413
*(float*)pr = val;
414414
else if (onumbits == 64)

‎src/julia.expmap‎

Lines changed: 0 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -38,12 +38,6 @@
3838
environ;
3939
__progname;
4040

41-
/* compiler run-time intrinsics */
42-
__gnu_h2f_ieee;
43-
__extendhfsf2;
44-
__gnu_f2h_ieee;
45-
__truncdfhf2;
46-
4741
local:
4842
*;
4943
};

‎src/julia_internal.h‎

Lines changed: 12 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1543,8 +1543,18 @@ jl_sym_t *_jl_symbol(const char *str, size_t len) JL_NOTSAFEPOINT;
15431543
#define JL_GC_ASSERT_LIVE(x) (void)(x)
15441544
#endif
15451545

1546-
float __gnu_h2f_ieee(uint16_t param) JL_NOTSAFEPOINT;
1547-
uint16_t __gnu_f2h_ieee(float param) JL_NOTSAFEPOINT;
1546+
JL_DLLEXPORT float julia__gnu_h2f_ieee(uint16_t param) JL_NOTSAFEPOINT;
1547+
JL_DLLEXPORT uint16_t julia__gnu_f2h_ieee(float param) JL_NOTSAFEPOINT;
1548+
JL_DLLEXPORT uint16_t julia__truncdfhf2(double param) JL_NOTSAFEPOINT;
1549+
//JL_DLLEXPORT double julia__extendhfdf2(uint16_t n) JL_NOTSAFEPOINT;
1550+
//JL_DLLEXPORT int32_t julia__fixhfsi(uint16_t n) JL_NOTSAFEPOINT;
1551+
//JL_DLLEXPORT int64_t julia__fixhfdi(uint16_t n) JL_NOTSAFEPOINT;
1552+
//JL_DLLEXPORT uint32_t julia__fixunshfsi(uint16_t n) JL_NOTSAFEPOINT;
1553+
//JL_DLLEXPORT uint64_t julia__fixunshfdi(uint16_t n) JL_NOTSAFEPOINT;
1554+
//JL_DLLEXPORT uint16_t julia__floatsihf(int32_t n) JL_NOTSAFEPOINT;
1555+
//JL_DLLEXPORT uint16_t julia__floatdihf(int64_t n) JL_NOTSAFEPOINT;
1556+
//JL_DLLEXPORT uint16_t julia__floatunsihf(uint32_t n) JL_NOTSAFEPOINT;
1557+
//JL_DLLEXPORT uint16_t julia__floatundihf(uint64_t n) JL_NOTSAFEPOINT;
15481558

15491559
#ifdef __cplusplus
15501560
}

‎src/runtime_intrinsics.c‎

Lines changed: 37 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -15,9 +15,6 @@
1515
const unsigned int host_char_bit = 8;
1616

1717
// float16 intrinsics
18-
// TODO: use LLVM's compiler-rt on all platforms (Xcode already links compiler-rt)
19-
20-
#if !defined(_OS_DARWIN_)
2118

2219
static inline float half_to_float(uint16_t ival) JL_NOTSAFEPOINT
2320
{
@@ -188,22 +185,17 @@ static inline uint16_t float_to_half(float param) JL_NOTSAFEPOINT
188185
return h;
189186
}
190187

191-
JL_DLLEXPORT float __gnu_h2f_ieee(uint16_t param)
188+
JL_DLLEXPORT float julia__gnu_h2f_ieee(uint16_t param)
192189
{
193190
return half_to_float(param);
194191
}
195192

196-
JL_DLLEXPORT float __extendhfsf2(uint16_t param)
197-
{
198-
return half_to_float(param);
199-
}
200-
201-
JL_DLLEXPORT uint16_t __gnu_f2h_ieee(float param)
193+
JL_DLLEXPORT uint16_t julia__gnu_f2h_ieee(float param)
202194
{
203195
return float_to_half(param);
204196
}
205197

206-
JL_DLLEXPORT uint16_t __truncdfhf2(double param)
198+
JL_DLLEXPORT uint16_t julia__truncdfhf2(double param)
207199
{
208200
float res = (float)param;
209201
uint32_t resi;
@@ -225,7 +217,25 @@ JL_DLLEXPORT uint16_t __truncdfhf2(double param)
225217
return float_to_half(res);
226218
}
227219

228-
#endif
220+
//JL_DLLEXPORT double julia__extendhfdf2(uint16_t n) { return (double)julia__gnu_h2f_ieee(n); }
221+
//JL_DLLEXPORT int32_t julia__fixhfsi(uint16_t n) { return (int32_t)julia__gnu_h2f_ieee(n); }
222+
//JL_DLLEXPORT int64_t julia__fixhfdi(uint16_t n) { return (int64_t)julia__gnu_h2f_ieee(n); }
223+
//JL_DLLEXPORT uint32_t julia__fixunshfsi(uint16_t n) { return (uint32_t)julia__gnu_h2f_ieee(n); }
224+
//JL_DLLEXPORT uint64_t julia__fixunshfdi(uint16_t n) { return (uint64_t)julia__gnu_h2f_ieee(n); }
225+
//JL_DLLEXPORT uint16_t julia__floatsihf(int32_t n) { return julia__gnu_f2h_ieee((float)n); }
226+
//JL_DLLEXPORT uint16_t julia__floatdihf(int64_t n) { return julia__gnu_f2h_ieee((float)n); }
227+
//JL_DLLEXPORT uint16_t julia__floatunsihf(uint32_t n) { return julia__gnu_f2h_ieee((float)n); }
228+
//JL_DLLEXPORT uint16_t julia__floatundihf(uint64_t n) { return julia__gnu_f2h_ieee((float)n); }
229+
//HANDLE_LIBCALL(F16, F128, __extendhftf2)
230+
//HANDLE_LIBCALL(F16, F80, __extendhfxf2)
231+
//HANDLE_LIBCALL(F80, F16, __truncxfhf2)
232+
//HANDLE_LIBCALL(F128, F16, __trunctfhf2)
233+
//HANDLE_LIBCALL(PPCF128, F16, __trunctfhf2)
234+
//HANDLE_LIBCALL(F16, I128, __fixhfti)
235+
//HANDLE_LIBCALL(F16, I128, __fixunshfti)
236+
//HANDLE_LIBCALL(I128, F16, __floattihf)
237+
//HANDLE_LIBCALL(I128, F16, __floatuntihf)
238+
229239

230240
// run time version of bitcast intrinsic
231241
JL_DLLEXPORT jl_value_t *jl_bitcast(jl_value_t *ty, jl_value_t *v)
@@ -551,9 +561,9 @@ static inline unsigned select_by_size(unsigned sz) JL_NOTSAFEPOINT
551561
}
552562

553563
#define fp_select(a, func) \
554-
sizeof(a) == sizeof(float) ? func##f((float)a) : func(a)
564+
sizeof(a) <= sizeof(float) ? func##f((float)a) : func(a)
555565
#define fp_select2(a, b, func) \
556-
sizeof(a) == sizeof(float) ? func##f(a, b) : func(a, b)
566+
sizeof(a) <= sizeof(float) ? func##f(a, b) : func(a, b)
557567

558568
// fast-function generators //
559569

@@ -597,11 +607,11 @@ static inline void name(unsigned osize, void *pa, void *pr) JL_NOTSAFEPOINT \
597607
static inline void name(unsigned osize, void *pa, void *pr) JL_NOTSAFEPOINT \
598608
{ \
599609
uint16_t a = *(uint16_t*)pa; \
600-
float A = __gnu_h2f_ieee(a); \
610+
float A = julia__gnu_h2f_ieee(a); \
601611
if (osize == 16) { \
602612
float R; \
603613
OP(&R, A); \
604-
*(uint16_t*)pr = __gnu_f2h_ieee(R); \
614+
*(uint16_t*)pr = julia__gnu_f2h_ieee(R); \
605615
} else { \
606616
OP((uint16_t*)pr, A); \
607617
} \
@@ -625,11 +635,11 @@ static void jl_##name##16(unsigned runtime_nbits, void *pa, void *pb, void *pr)
625635
{ \
626636
uint16_t a = *(uint16_t*)pa; \
627637
uint16_t b = *(uint16_t*)pb; \
628-
float A = __gnu_h2f_ieee(a); \
629-
float B = __gnu_h2f_ieee(b); \
638+
float A = julia__gnu_h2f_ieee(a); \
639+
float B = julia__gnu_h2f_ieee(b); \
630640
runtime_nbits = 16; \
631641
float R = OP(A, B); \
632-
*(uint16_t*)pr = __gnu_f2h_ieee(R); \
642+
*(uint16_t*)pr = julia__gnu_f2h_ieee(R); \
633643
}
634644

635645
// float or integer inputs, bool output
@@ -650,8 +660,8 @@ static int jl_##name##16(unsigned runtime_nbits, void *pa, void *pb) JL_NOTSAFEP
650660
{ \
651661
uint16_t a = *(uint16_t*)pa; \
652662
uint16_t b = *(uint16_t*)pb; \
653-
float A = __gnu_h2f_ieee(a); \
654-
float B = __gnu_h2f_ieee(b); \
663+
float A = julia__gnu_h2f_ieee(a); \
664+
float B = julia__gnu_h2f_ieee(b); \
655665
runtime_nbits = 16; \
656666
return OP(A, B); \
657667
}
@@ -691,12 +701,12 @@ static void jl_##name##16(unsigned runtime_nbits, void *pa, void *pb, void *pc,
691701
uint16_t a = *(uint16_t*)pa; \
692702
uint16_t b = *(uint16_t*)pb; \
693703
uint16_t c = *(uint16_t*)pc; \
694-
float A = __gnu_h2f_ieee(a); \
695-
float B = __gnu_h2f_ieee(b); \
696-
float C = __gnu_h2f_ieee(c); \
704+
float A = julia__gnu_h2f_ieee(a); \
705+
float B = julia__gnu_h2f_ieee(b); \
706+
float C = julia__gnu_h2f_ieee(c); \
697707
runtime_nbits = 16; \
698708
float R = OP(A, B, C); \
699-
*(uint16_t*)pr = __gnu_f2h_ieee(R); \
709+
*(uint16_t*)pr = julia__gnu_f2h_ieee(R); \
700710
}
701711

702712

@@ -1318,7 +1328,7 @@ static inline int fpiseq##nbits(c_type a, c_type b) JL_NOTSAFEPOINT { \
13181328
fpiseq_n(float, 32)
13191329
fpiseq_n(double, 64)
13201330
#define fpiseq(a,b) \
1321-
sizeof(a) == sizeof(float) ? fpiseq32(a, b) : fpiseq64(a, b)
1331+
sizeof(a) <= sizeof(float) ? fpiseq32(a, b) : fpiseq64(a, b)
13221332

13231333
bool_fintrinsic(eq,eq_float)
13241334
bool_fintrinsic(ne,ne_float)
@@ -1367,7 +1377,7 @@ cvt_iintrinsic(LLVMFPtoUI, fptoui)
13671377
if (!(osize < 8 * sizeof(a))) \
13681378
jl_error("fptrunc: output bitsize must be < input bitsize"); \
13691379
else if (osize == 16) \
1370-
*(uint16_t*)pr = __gnu_f2h_ieee(a); \
1380+
*(uint16_t*)pr = julia__gnu_f2h_ieee(a); \
13711381
else if (osize == 32) \
13721382
*(float*)pr = a; \
13731383
else if (osize == 64) \

0 commit comments

Comments
 (0)