Skip to content

[libclc] Optimize generic CLC fmin/fmax #128506

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jul 29, 2025
Merged

[libclc] Optimize generic CLC fmin/fmax #128506

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
414c6cf
[libclc] Optimize generic CLC fmin/fmax
frasercrmck Feb 24, 2025
001f427
spirv fmin/fmax
frasercrmck Jul 28, 2025
8cd4a8e
amdgcn/r600 simplify
frasercrmck Jul 28, 2025
ce099c8
use builtins
frasercrmck Jul 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions libclc/clc/lib/amdgcn/SOURCES
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,3 @@
math/clc_fmax.cl
math/clc_fmin.cl
math/clc_ldexp_override.cl
workitem/clc_get_global_offset.cl
workitem/clc_get_global_size.cl
Expand Down
47 changes: 2 additions & 45 deletions libclc/clc/lib/generic/math/clc_fmax.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,53 +6,10 @@
//
//===----------------------------------------------------------------------===//

#include <clc/clcmacro.h>
#include <clc/internal/clc.h>
#include <clc/relational/clc_isnan.h>

#define __FLOAT_ONLY
#define __CLC_MIN_VECSIZE 1
#define FUNCTION __clc_fmax
#define __IMPL_FUNCTION __builtin_fmaxf
#define __CLC_BODY <clc/shared/binary_def_scalarize.inc>
#include <clc/math/gentype.inc>
#undef __CLC_MIN_VECSIZE
#undef FUNCTION
#undef __IMPL_FUNCTION

#ifdef cl_khr_fp64

#pragma OPENCL EXTENSION cl_khr_fp64 : enable

#define __DOUBLE_ONLY
#define __CLC_MIN_VECSIZE 1
#define FUNCTION __clc_fmax
#define __IMPL_FUNCTION __builtin_fmax
#define __CLC_BODY <clc/shared/binary_def_scalarize.inc>
#include <clc/math/gentype.inc>
#undef __CLC_MIN_VECSIZE
#undef FUNCTION
#undef __IMPL_FUNCTION
#define __IMPL_FUNCTION(x) __builtin_elementwise_maximumnum
#define __CLC_BODY <clc/shared/binary_def.inc>

#endif

#ifdef cl_khr_fp16

#pragma OPENCL EXTENSION cl_khr_fp16 : enable

_CLC_DEF _CLC_OVERLOAD half __clc_fmax(half x, half y) {
if (__clc_isnan(x))
return y;
if (__clc_isnan(y))
return x;
return (x < y) ? y : x;
}

#define __HALF_ONLY
#define __CLC_SUPPORTED_VECSIZE_OR_1 2
#define FUNCTION __clc_fmax
#define __CLC_BODY <clc/shared/binary_def_scalarize.inc>
#include <clc/math/gentype.inc>
#undef FUNCTION

#endif
46 changes: 2 additions & 44 deletions libclc/clc/lib/generic/math/clc_fmin.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,52 +6,10 @@
//
//===----------------------------------------------------------------------===//

#include <clc/clcmacro.h>
#include <clc/internal/clc.h>
#include <clc/relational/clc_isnan.h>

#define __FLOAT_ONLY
#define __CLC_MIN_VECSIZE 1
#define FUNCTION __clc_fmin
#define __IMPL_FUNCTION __builtin_fminf
#define __CLC_BODY <clc/shared/binary_def_scalarize.inc>
#include <clc/math/gentype.inc>
#undef __CLC_MIN_VECSIZE
#undef FUNCTION
#undef __IMPL_FUNCTION

#ifdef cl_khr_fp64

#pragma OPENCL EXTENSION cl_khr_fp64 : enable

#define __DOUBLE_ONLY
#define __CLC_MIN_VECSIZE 1
#define FUNCTION __clc_fmin
#define __IMPL_FUNCTION __builtin_fmin
#define __CLC_BODY <clc/shared/binary_def_scalarize.inc>
#include <clc/math/gentype.inc>
#undef __CLC_MIN_VECSIZE
#undef FUNCTION
#undef __IMPL_FUNCTION
#define __IMPL_FUNCTION(x) __builtin_elementwise_minimumnum
#define __CLC_BODY <clc/shared/binary_def.inc>

#endif

#ifdef cl_khr_fp16

#pragma OPENCL EXTENSION cl_khr_fp16 : enable

_CLC_DEF _CLC_OVERLOAD half __clc_fmin(half x, half y) {
if (__clc_isnan(x))
return y;
if (__clc_isnan(y))
return x;
return (y < x) ? y : x;
}

#define __HALF_ONLY
#define __CLC_SUPPORTED_VECSIZE_OR_1 2
#define FUNCTION __clc_fmin
#define __CLC_BODY <clc/shared/binary_def_scalarize.inc>
#include <clc/math/gentype.inc>

#endif
2 changes: 0 additions & 2 deletions libclc/clc/lib/r600/SOURCES
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,2 @@
math/clc_fmax.cl
math/clc_fmin.cl
math/clc_native_rsqrt.cl
math/clc_rsqrt_override.cl
41 changes: 0 additions & 41 deletions libclc/clc/lib/r600/math/clc_fmax.cl
View file
Open in desktop

This file was deleted.

42 changes: 0 additions & 42 deletions libclc/clc/lib/r600/math/clc_fmin.cl
View file
Open in desktop

This file was deleted.

2 changes: 2 additions & 0 deletions libclc/clc/lib/spirv/SOURCES
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1,3 @@
math/clc_fmax.cl
math/clc_fmin.cl
math/clc_runtime_has_hw_fma32.cl
21 changes: 2 additions & 19 deletions libclc/clc/lib/amdgcn/math/clc_fmax.cl → libclc/clc/lib/spirv/math/clc_fmax.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,40 +8,23 @@

#include <clc/clcmacro.h>
#include <clc/internal/clc.h>
#include <clc/relational/clc_isnan.h>

_CLC_DEF _CLC_OVERLOAD float __clc_fmax(float x, float y) {
// fcanonicalize removes sNaNs and flushes denormals if not enabled. Otherwise
// fmax instruction flushes the values for comparison, but outputs original
// denormal
x = __builtin_canonicalizef(x);
y = __builtin_canonicalizef(y);
return __builtin_fmaxf(x, y);
}

#ifdef cl_khr_fp64

#pragma OPENCL EXTENSION cl_khr_fp64 : enable

_CLC_DEF _CLC_OVERLOAD double __clc_fmax(double x, double y) {
x = __builtin_canonicalize(x);
y = __builtin_canonicalize(y);
return __builtin_fmax(x, y);
}

#endif
#ifdef cl_khr_fp16

#ifdef cl_khr_fp16
#pragma OPENCL EXTENSION cl_khr_fp16 : enable

_CLC_DEF _CLC_OVERLOAD half __clc_fmax(half x, half y) {
if (__clc_isnan(x))
return y;
if (__clc_isnan(y))
return x;
return (y < x) ? x : y;
return __builtin_fmaxf16(x, y);
}

#endif

#define FUNCTION __clc_fmax
Expand Down
20 changes: 1 addition & 19 deletions libclc/clc/lib/amdgcn/math/clc_fmin.cl → libclc/clc/lib/spirv/math/clc_fmin.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,41 +8,23 @@

#include <clc/clcmacro.h>
#include <clc/internal/clc.h>
#include <clc/relational/clc_isnan.h>

_CLC_DEF _CLC_OVERLOAD float __clc_fmin(float x, float y) {
// fcanonicalize removes sNaNs and flushes denormals if not enabled. Otherwise
// fmin instruction flushes the values for comparison, but outputs original
// denormal
x = __builtin_canonicalizef(x);
y = __builtin_canonicalizef(y);
return __builtin_fminf(x, y);
}

#ifdef cl_khr_fp64

#pragma OPENCL EXTENSION cl_khr_fp64 : enable

_CLC_DEF _CLC_OVERLOAD double __clc_fmin(double x, double y) {
x = __builtin_canonicalize(x);
y = __builtin_canonicalize(y);
return __builtin_fmin(x, y);
}

#endif

#ifdef cl_khr_fp16

#pragma OPENCL EXTENSION cl_khr_fp16 : enable

_CLC_DEF _CLC_OVERLOAD half __clc_fmin(half x, half y) {
if (__clc_isnan(x))
return y;
if (__clc_isnan(y))
return x;
return (y < x) ? y : x;
return __builtin_fminf16(x, y);
}

#endif

#define FUNCTION __clc_fmin
Expand Down
Loading