mirrors/llvm-project
0
0
Fork 0
mirror of https://github.com/llvm/llvm-project.git synced 2025-04-15 22:06:32 +00:00
Code Issues Projects Releases Wiki Activity

[libclc] Move hypot to CLC library; optimize (#129551)

Browse Source 
This was already nominally in the CLC library; this commit just formally
moves it over. It simultaneously optimizes it for vector types by
avoiding scalarization.
This commit is contained in:
Fraser Cormack 2025-03-04 14:16:16 +00:00 committed by GitHub
parent f838a5e96c
commit e5d5503e4e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
10 changed files with 161 additions and 118 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
libclc/clc/include/clc/math/clc_hypot.h Normal file
View File

@ -0,0 +1,12 @@
#ifndef __CLC_MATH_CLC_HYPOT_H__
#define __CLC_MATH_CLC_HYPOT_H__
#define __CLC_BODY <clc/shared/binary_decl.inc>
#define __CLC_FUNCTION __clc_hypot
#include <clc/math/gentype.inc>
#undef __CLC_BODY
#undef __CLC_FUNCTION
#endif // __CLC_MATH_CLC_HYPOT_H__

1
libclc/clc/lib/generic/SOURCES
View File

@ -23,6 +23,7 @@ math/clc_fabs.cl
math/clc_fma.cl
math/clc_floor.cl
math/clc_frexp.cl
math/clc_hypot.cl
math/clc_ldexp.cl
math/clc_log.cl
math/clc_log10.cl

37
libclc/clc/lib/generic/math/clc_hypot.cl Normal file
View File

@ -0,0 +1,37 @@
/*
* Copyright (c) 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include <clc/clc_convert.h>
#include <clc/clcmacro.h>
#include <clc/integer/clc_abs.h>
#include <clc/internal/clc.h>
#include <clc/math/clc_fma.h>
#include <clc/math/clc_mad.h>
#include <clc/math/clc_sqrt.h>
#include <clc/math/clc_subnormal_config.h>
#include <clc/math/math.h>
#include <clc/relational/clc_isnan.h>
#include <clc/shared/clc_clamp.h>
#define __CLC_BODY <clc_hypot.inc>
#include <clc/math/gentype.inc>
#undef __CLC_BODY

108
libclc/clc/lib/generic/math/clc_hypot.inc Normal file
View File

@ -0,0 +1,108 @@
/*
* Copyright (c) 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
// Returns sqrt(x*x + y*y) with no overflow or underflow unless the result
// warrants it
#if __CLC_FPSIZE == 32
_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_hypot(__CLC_GENTYPE x,
__CLC_GENTYPE y) {
__CLC_UINTN ux = __CLC_AS_UINTN(x);
__CLC_UINTN aux = ux & EXSIGNBIT_SP32;
__CLC_UINTN uy = __CLC_AS_UINTN(y);
__CLC_UINTN auy = uy & EXSIGNBIT_SP32;
__CLC_INTN c = aux > auy;
ux = c ? aux : auy;
uy = c ? auy : aux;
__CLC_INTN xexp = __clc_clamp(
__CLC_AS_INTN(ux >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32, -126, 126);
__CLC_GENTYPE fx_exp =
__CLC_AS_GENTYPE((xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
__CLC_GENTYPE fi_exp =
__CLC_AS_GENTYPE((-xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
__CLC_GENTYPE fx = __CLC_AS_GENTYPE(ux) * fi_exp;
__CLC_GENTYPE fy = __CLC_AS_GENTYPE(uy) * fi_exp;
__CLC_GENTYPE retval = __clc_sqrt(__clc_mad(fx, fx, fy * fy)) * fx_exp;
retval = (ux > PINFBITPATT_SP32 || uy == 0) ? __CLC_AS_GENTYPE(ux) : retval;
retval = (ux == PINFBITPATT_SP32 || uy == PINFBITPATT_SP32)
? __CLC_AS_GENTYPE((__CLC_UINTN)PINFBITPATT_SP32)
: retval;
return retval;
}
#elif __CLC_FPSIZE == 64
_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_hypot(__CLC_GENTYPE x,
__CLC_GENTYPE y) {
__CLC_ULONGN ux = __CLC_AS_ULONGN(x) & ~SIGNBIT_DP64;
__CLC_INTN xexp = __CLC_CONVERT_INTN(ux >> EXPSHIFTBITS_DP64);
x = __CLC_AS_GENTYPE(ux);
__CLC_ULONGN uy = __CLC_AS_ULONGN(y) & ~SIGNBIT_DP64;
__CLC_INTN yexp = __CLC_CONVERT_INTN(uy >> EXPSHIFTBITS_DP64);
y = __CLC_AS_GENTYPE(uy);
__CLC_LONGN c = __CLC_CONVERT_LONGN(xexp > EXPBIAS_DP64 + 500 ||
yexp > EXPBIAS_DP64 + 500);
__CLC_GENTYPE preadjust = c ? 0x1.0p-600 : 1.0;
__CLC_GENTYPE postadjust = c ? 0x1.0p+600 : 1.0;
c = __CLC_CONVERT_LONGN(xexp < EXPBIAS_DP64 - 500 ||
yexp < EXPBIAS_DP64 - 500);
preadjust = c ? 0x1.0p+600 : preadjust;
postadjust = c ? 0x1.0p-600 : postadjust;
__CLC_GENTYPE ax = x * preadjust;
__CLC_GENTYPE ay = y * preadjust;
// The post adjust may overflow, but this can't be avoided in any case
__CLC_GENTYPE r = __clc_sqrt(__clc_fma(ax, ax, ay * ay)) * postadjust;
// If the difference in exponents between x and y is large
__CLC_GENTYPE s = x + y;
c = __CLC_CONVERT_LONGN(__clc_abs(xexp - yexp) > MANTLENGTH_DP64 + 1);
r = c ? s : r;
// Check for NaN
c = __clc_isnan(x) || __clc_isnan(y);
r = c ? __CLC_AS_GENTYPE((__CLC_ULONGN)QNANBITPATT_DP64) : r;
// If either is Inf, we must return Inf
c = x == __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64) ||
y == __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64);
r = c ? __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64) : r;
return r;
}
#elif __CLC_FPSIZE == 16
_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_hypot(__CLC_GENTYPE x,
__CLC_GENTYPE y) {
return __CLC_CONVERT_GENTYPE(
__clc_hypot(__CLC_CONVERT_FLOATN(x), __CLC_CONVERT_FLOATN(y)));
}
#endif

1
libclc/clspv/lib/SOURCES
View File

@ -18,7 +18,6 @@ subnormal_config.cl
../../generic/lib/math/cbrt.cl
../../generic/lib/math/clc_exp10.cl
../../generic/lib/math/clc_fmod.cl
../../generic/lib/math/clc_hypot.cl
../../generic/lib/math/clc_pow.cl
../../generic/lib/math/clc_pown.cl
../../generic/lib/math/clc_powr.cl

5
libclc/generic/include/math/clc_hypot.h
View File

@ -1,5 +0,0 @@
#define __CLC_FUNCTION __clc_hypot
#define __CLC_BODY <clc/math/binary_decl_tt.inc>
#include <clc/math/gentype.inc>
#undef __CLC_BODY
#undef __CLC_FUNCTION

1
libclc/generic/lib/SOURCES
View File

@ -127,7 +127,6 @@ math/half_rsqrt.cl
math/half_sin.cl
math/half_sqrt.cl
math/half_tan.cl
math/clc_hypot.cl
math/hypot.cl
math/ilogb.cl
math/ldexp.cl

106
libclc/generic/lib/math/clc_hypot.cl
View File

@ -1,106 +0,0 @@
/*
* Copyright (c) 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include <clc/clc.h>
#include <clc/clcmacro.h>
#include <clc/integer/clc_abs.h>
#include <clc/math/clc_fma.h>
#include <clc/math/clc_mad.h>
#include <clc/math/clc_subnormal_config.h>
#include <clc/math/math.h>
#include <clc/math/clc_sqrt.h>
#include <clc/relational/clc_isnan.h>
#include <clc/shared/clc_clamp.h>
#include <math/clc_hypot.h>
// Returns sqrt(x*x + y*y) with no overflow or underflow unless the result
// warrants it
_CLC_DEF _CLC_OVERLOAD float __clc_hypot(float x, float y) {
uint ux = as_uint(x);
uint aux = ux & EXSIGNBIT_SP32;
uint uy = as_uint(y);
uint auy = uy & EXSIGNBIT_SP32;
float retval;
int c = aux > auy;
ux = c ? aux : auy;
uy = c ? auy : aux;
int xexp =
__clc_clamp((int)(ux >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32, -126, 126);
float fx_exp = as_float((xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
float fi_exp = as_float((-xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
float fx = as_float(ux) * fi_exp;
float fy = as_float(uy) * fi_exp;
retval = __clc_sqrt(__clc_mad(fx, fx, fy * fy)) * fx_exp;
retval = ux > PINFBITPATT_SP32 | uy == 0 ? as_float(ux) : retval;
retval = ux == PINFBITPATT_SP32 | uy == PINFBITPATT_SP32
? as_float(PINFBITPATT_SP32)
: retval;
return retval;
}
_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, float, __clc_hypot, float, float)
#ifdef cl_khr_fp64
_CLC_DEF _CLC_OVERLOAD double __clc_hypot(double x, double y) {
ulong ux = as_ulong(x) & ~SIGNBIT_DP64;
int xexp = ux >> EXPSHIFTBITS_DP64;
x = as_double(ux);
ulong uy = as_ulong(y) & ~SIGNBIT_DP64;
int yexp = uy >> EXPSHIFTBITS_DP64;
y = as_double(uy);
int c = xexp > EXPBIAS_DP64 + 500 | yexp > EXPBIAS_DP64 + 500;
double preadjust = c ? 0x1.0p-600 : 1.0;
double postadjust = c ? 0x1.0p+600 : 1.0;
c = xexp < EXPBIAS_DP64 - 500 | yexp < EXPBIAS_DP64 - 500;
preadjust = c ? 0x1.0p+600 : preadjust;
postadjust = c ? 0x1.0p-600 : postadjust;
double ax = x * preadjust;
double ay = y * preadjust;
// The post adjust may overflow, but this can't be avoided in any case
double r = __clc_sqrt(__clc_fma(ax, ax, ay * ay)) * postadjust;
// If the difference in exponents between x and y is large
double s = x + y;
c = __clc_abs(xexp - yexp) > MANTLENGTH_DP64 + 1;
r = c ? s : r;
// Check for NaN
// c = x != x | y != y;
c = __clc_isnan(x) | __clc_isnan(y);
r = c ? as_double(QNANBITPATT_DP64) : r;
// If either is Inf, we must return Inf
c = x == as_double(PINFBITPATT_DP64) | y == as_double(PINFBITPATT_DP64);
r = c ? as_double(PINFBITPATT_DP64) : r;
return r;
}
_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, double, __clc_hypot, double,
double)
#endif

7
libclc/generic/lib/math/hypot.cl
View File

@ -1,7 +1,6 @@
#include <clc/clc.h>
#include <clc/math/clc_hypot.h>
#include <math/clc_hypot.h>
#define __CLC_FUNC hypot
#define __CLC_BODY <clc_sw_binary.inc>
#define FUNCTION hypot
#define __CLC_BODY <clc/shared/binary_def.inc>
#include <clc/math/gentype.inc>

1
libclc/spirv/lib/SOURCES
View File

@ -48,7 +48,6 @@ math/fma.cl
../../generic/lib/math/frexp.cl
../../generic/lib/math/half_rsqrt.cl
../../generic/lib/math/half_sqrt.cl
../../generic/lib/math/clc_hypot.cl
../../generic/lib/math/hypot.cl
../../generic/lib/math/ilogb.cl
../../generic/lib/math/ldexp.cl