[libclc] [libclc] Move conversion builtins to the CLC library (PR #124727)
Fraser Cormack via cfe-commits
cfe-commits at lists.llvm.org
Tue Feb 11 03:20:49 PST 2025
https://github.com/frasercrmck updated https://github.com/llvm/llvm-project/pull/124727
>From c0252c8da6fb4ae68699a7e6fc38ed7c643338e8 Mon Sep 17 00:00:00 2001
From: Fraser Cormack <fraser at codeplay.com>
Date: Mon, 11 Nov 2024 15:46:56 +0000
Subject: [PATCH 1/2] [libclc] Move conversion builtins to the CLC library
This commit moves the implementations of conversion builtins to the CLC
library. It keeps the dichotomy of regular vs. clspv implementations of
the conversions. However, for the sake of a consistent interface all CLC
conversion routines are built, even the ones that clspv opts out of in
the user-facing OpenCL layer.
It simultaneously updates the python script to use f-strings for
formatting.
---
libclc/CMakeLists.txt | 31 +-
libclc/clc/include/clc/clc_convert.h | 98 +++++
libclc/clc/include/clc/float/definitions.h | 88 +++++
libclc/generic/include/clc/convert.h | 83 +++--
.../generic/include/clc/float/definitions.h | 88 -----
libclc/generic/lib/gen_convert.py | 337 ++++++++----------
6 files changed, 419 insertions(+), 306 deletions(-)
create mode 100644 libclc/clc/include/clc/clc_convert.h
create mode 100644 libclc/clc/include/clc/float/definitions.h
delete mode 100644 libclc/generic/include/clc/float/definitions.h
diff --git a/libclc/CMakeLists.txt b/libclc/CMakeLists.txt
index 2978fadc2c29fc3..c88ea9700d100b8 100644
--- a/libclc/CMakeLists.txt
+++ b/libclc/CMakeLists.txt
@@ -247,11 +247,27 @@ add_custom_target( "generate_convert.cl" DEPENDS convert.cl )
set_target_properties( "generate_convert.cl" PROPERTIES FOLDER "libclc/Sourcegenning" )
add_custom_command(
- OUTPUT clspv-convert.cl
- COMMAND ${Python3_EXECUTABLE} ${script_loc} --clspv > clspv-convert.cl
+ OUTPUT clc-convert.cl
+ COMMAND ${Python3_EXECUTABLE} ${script_loc} --clc > clc-convert.cl
DEPENDS ${script_loc} )
-add_custom_target( "clspv-generate_convert.cl" DEPENDS clspv-convert.cl )
-set_target_properties( "clspv-generate_convert.cl" PROPERTIES FOLDER "libclc/Sourcegenning" )
+add_custom_target( "clc-generate_convert.cl" DEPENDS clc-convert.cl )
+set_target_properties( "clc-generate_convert.cl" PROPERTIES FOLDER "libclc/Sourcegenning" )
+
+if ( clspv-- IN_LIST LIBCLC_TARGETS_TO_BUILD OR clspv64-- IN_LIST LIBCLC_TARGETS_TO_BUILD )
+ add_custom_command(
+ OUTPUT clspv-convert.cl
+ COMMAND ${Python3_EXECUTABLE} ${script_loc} --clspv > clspv-convert.cl
+ DEPENDS ${script_loc} )
+ add_custom_target( "clspv-generate_convert.cl" DEPENDS clspv-convert.cl )
+ set_target_properties( "clspv-generate_convert.cl" PROPERTIES FOLDER "libclc/Sourcegenning" )
+
+ add_custom_command(
+ OUTPUT clc-clspv-convert.cl
+ COMMAND ${Python3_EXECUTABLE} ${script_loc} --clc --clspv > clc-clspv-convert.cl
+ DEPENDS ${script_loc} )
+ add_custom_target( "clc-clspv-generate_convert.cl" DEPENDS clc-clspv-convert.cl )
+ set_target_properties( "clc-clspv-generate_convert.cl" PROPERTIES FOLDER "libclc/Sourcegenning" )
+endif()
enable_testing()
@@ -289,6 +305,12 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
set( clc_lib_files )
set( clc_dirs ${dirs} generic )
+ if( ARCH STREQUAL clspv OR ARCH STREQUAL clspv64 )
+ set( clc_gen_files clc-clspv-convert.cl )
+ else()
+ set( clc_gen_files clc-convert.cl )
+ endif()
+
libclc_configure_lib_source(
clc_lib_files
CLC_INTERNAL
@@ -372,6 +394,7 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
COMPILE_FLAGS ${clc_build_flags}
OPT_FLAGS ${opt_flags}
LIB_FILES ${clc_lib_files}
+ GEN_FILES ${clc_gen_files}
)
list( APPEND build_flags
diff --git a/libclc/clc/include/clc/clc_convert.h b/libclc/clc/include/clc/clc_convert.h
new file mode 100644
index 000000000000000..20bbd57540b3064
--- /dev/null
+++ b/libclc/clc/include/clc/clc_convert.h
@@ -0,0 +1,98 @@
+#ifndef __CLC_CLC_CONVERT_H__
+#define __CLC_CLC_CONVERT_H__
+
+#define _CLC_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
+ _CLC_OVERLOAD _CLC_DECL TO_TYPE __clc_convert_##TO_TYPE##SUFFIX(FROM_TYPE x);
+
+#define _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##2, TO_TYPE##2, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##3, TO_TYPE##3, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##4, TO_TYPE##4, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##8, TO_TYPE##8, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##16, TO_TYPE##16, SUFFIX)
+
+#define _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, char, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, uchar, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, int, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, uint, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, short, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, ushort, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, long, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, ulong, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, float, SUFFIX)
+
+#if defined(cl_khr_fp64) && defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, double, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, half, SUFFIX)
+#elif defined(cl_khr_fp64)
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, double, SUFFIX)
+#elif defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, half, SUFFIX)
+#else
+#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX)
+#endif
+
+#define _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(char, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(uchar, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(int, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(uint, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(short, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(ushort, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(long, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(ulong, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(float, SUFFIX)
+
+#if defined(cl_khr_fp64) && defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
+ _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(double, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(half, SUFFIX)
+#elif defined(cl_khr_fp64)
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
+ _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(double, SUFFIX)
+#elif defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
+ _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(half, SUFFIX)
+#else
+#define _CLC_VECTOR_CONVERT_TO(SUFFIX) _CLC_VECTOR_CONVERT_TO1(SUFFIX)
+#endif
+
+#define _CLC_VECTOR_CONVERT_TO_SUFFIX(ROUND) \
+ _CLC_VECTOR_CONVERT_TO(_sat##ROUND) \
+ _CLC_VECTOR_CONVERT_TO(ROUND)
+
+_CLC_VECTOR_CONVERT_TO_SUFFIX(_rtn)
+_CLC_VECTOR_CONVERT_TO_SUFFIX(_rte)
+_CLC_VECTOR_CONVERT_TO_SUFFIX(_rtz)
+_CLC_VECTOR_CONVERT_TO_SUFFIX(_rtp)
+_CLC_VECTOR_CONVERT_TO_SUFFIX()
+
+#undef _CLC_VECTOR_CONVERT_TO_SUFFIX
+#undef _CLC_VECTOR_CONVERT_TO
+#undef _CLC_VECTOR_CONVERT_TO1
+#undef _CLC_VECTOR_CONVERT_FROM
+#undef _CLC_VECTOR_CONVERT_FROM1
+#undef _CLC_VECTOR_CONVERT_DECL
+#undef _CLC_CONVERT_DECL
+
+#endif // __CLC_CLC_CONVERT_H__
diff --git a/libclc/clc/include/clc/float/definitions.h b/libclc/clc/include/clc/float/definitions.h
new file mode 100644
index 000000000000000..6fea19a1cae98b7
--- /dev/null
+++ b/libclc/clc/include/clc/float/definitions.h
@@ -0,0 +1,88 @@
+#define MAXFLOAT 0x1.fffffep127f
+#define HUGE_VALF __builtin_huge_valf()
+#define INFINITY __builtin_inff()
+#define NAN __builtin_nanf("")
+
+#define FLT_DIG 6
+#define FLT_MANT_DIG 24
+#define FLT_MAX_10_EXP +38
+#define FLT_MAX_EXP +128
+#define FLT_MIN_10_EXP -37
+#define FLT_MIN_EXP -125
+#define FLT_RADIX 2
+#define FLT_MAX MAXFLOAT
+#define FLT_MIN 0x1.0p-126f
+#define FLT_EPSILON 0x1.0p-23f
+
+#define FP_ILOGB0 (-2147483647 - 1)
+#define FP_ILOGBNAN 2147483647
+
+#define M_E_F 0x1.5bf0a8p+1f
+#define M_LOG2E_F 0x1.715476p+0f
+#define M_LOG10E_F 0x1.bcb7b2p-2f
+#define M_LN2_F 0x1.62e430p-1f
+#define M_LN10_F 0x1.26bb1cp+1f
+#define M_PI_F 0x1.921fb6p+1f
+#define M_PI_2_F 0x1.921fb6p+0f
+#define M_PI_4_F 0x1.921fb6p-1f
+#define M_1_PI_F 0x1.45f306p-2f
+#define M_2_PI_F 0x1.45f306p-1f
+#define M_2_SQRTPI_F 0x1.20dd76p+0f
+#define M_SQRT2_F 0x1.6a09e6p+0f
+#define M_SQRT1_2_F 0x1.6a09e6p-1f
+
+#define M_LOG210_F 0x1.a934f0p+1f
+
+#ifdef cl_khr_fp64
+
+#define HUGE_VAL __builtin_huge_val()
+
+#define DBL_DIG 15
+#define DBL_MANT_DIG 53
+#define DBL_MAX_10_EXP +308
+#define DBL_MAX_EXP +1024
+#define DBL_MIN_10_EXP -307
+#define DBL_MIN_EXP -1021
+#define DBL_MAX 0x1.fffffffffffffp1023
+#define DBL_MIN 0x1.0p-1022
+#define DBL_EPSILON 0x1.0p-52
+
+#define M_E 0x1.5bf0a8b145769p+1
+#define M_LOG2E 0x1.71547652b82fep+0
+#define M_LOG10E 0x1.bcb7b1526e50ep-2
+#define M_LN2 0x1.62e42fefa39efp-1
+#define M_LN10 0x1.26bb1bbb55516p+1
+#define M_PI 0x1.921fb54442d18p+1
+#define M_PI_2 0x1.921fb54442d18p+0
+#define M_PI_4 0x1.921fb54442d18p-1
+#define M_1_PI 0x1.45f306dc9c883p-2
+#define M_2_PI 0x1.45f306dc9c883p-1
+#define M_2_SQRTPI 0x1.20dd750429b6dp+0
+#define M_SQRT2 0x1.6a09e667f3bcdp+0
+#define M_SQRT1_2 0x1.6a09e667f3bcdp-1
+
+#ifdef __CLC_INTERNAL
+#define M_LOG210 0x1.a934f0979a371p+1
+#endif
+
+#endif
+
+#ifdef cl_khr_fp16
+
+#if __OPENCL_VERSION__ >= 120
+
+#define HALF_DIG 3
+#define HALF_MANT_DIG 11
+#define HALF_MAX_10_EXP +4
+#define HALF_MAX_EXP +16
+#define HALF_MIN_10_EXP -4
+#define HALF_MIN_EXP -13
+
+#define HALF_RADIX 2
+#define HALF_MAX 0x1.ffcp15h
+#define HALF_MIN 0x1.0p-14h
+#define HALF_EPSILON 0x1.0p-10h
+
+#endif
+
+#endif
diff --git a/libclc/generic/include/clc/convert.h b/libclc/generic/include/clc/convert.h
index 8219df47ad2c68c..4e96c874b5cd854 100644
--- a/libclc/generic/include/clc/convert.h
+++ b/libclc/generic/include/clc/convert.h
@@ -1,74 +1,81 @@
-#define _CLC_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
+#define _CLC_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
_CLC_OVERLOAD _CLC_DECL TO_TYPE convert_##TO_TYPE##SUFFIX(FROM_TYPE x);
-#define _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
- _CLC_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
- _CLC_CONVERT_DECL(FROM_TYPE##2, TO_TYPE##2, SUFFIX) \
- _CLC_CONVERT_DECL(FROM_TYPE##3, TO_TYPE##3, SUFFIX) \
- _CLC_CONVERT_DECL(FROM_TYPE##4, TO_TYPE##4, SUFFIX) \
- _CLC_CONVERT_DECL(FROM_TYPE##8, TO_TYPE##8, SUFFIX) \
+#define _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE, TO_TYPE, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##2, TO_TYPE##2, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##3, TO_TYPE##3, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##4, TO_TYPE##4, SUFFIX) \
+ _CLC_CONVERT_DECL(FROM_TYPE##8, TO_TYPE##8, SUFFIX) \
_CLC_CONVERT_DECL(FROM_TYPE##16, TO_TYPE##16, SUFFIX)
-#define _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, char, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, uchar, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, int, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, uint, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, short, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, ushort, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, long, SUFFIX) \
- _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, ulong, SUFFIX) \
+#define _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, char, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, uchar, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, int, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, uint, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, short, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, ushort, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, long, SUFFIX) \
+ _CLC_VECTOR_CONVERT_DECL(FROM_TYPE, ulong, SUFFIX) \
_CLC_VECTOR_CONVERT_DECL(FROM_TYPE, float, SUFFIX)
#if defined(cl_khr_fp64) && defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
_CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
_CLC_VECTOR_CONVERT_DECL(FROM_TYPE, double, SUFFIX) \
_CLC_VECTOR_CONVERT_DECL(FROM_TYPE, half, SUFFIX)
#elif defined(cl_khr_fp64)
-#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
_CLC_VECTOR_CONVERT_DECL(FROM_TYPE, double, SUFFIX)
#elif defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
_CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX) \
_CLC_VECTOR_CONVERT_DECL(FROM_TYPE, half, SUFFIX)
#else
-#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
+#define _CLC_VECTOR_CONVERT_FROM(FROM_TYPE, SUFFIX) \
_CLC_VECTOR_CONVERT_FROM1(FROM_TYPE, SUFFIX)
#endif
-#define _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(char, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(uchar, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(int, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(uint, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(short, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(ushort, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(long, SUFFIX) \
- _CLC_VECTOR_CONVERT_FROM(ulong, SUFFIX) \
+#define _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(char, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(uchar, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(int, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(uint, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(short, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(ushort, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(long, SUFFIX) \
+ _CLC_VECTOR_CONVERT_FROM(ulong, SUFFIX) \
_CLC_VECTOR_CONVERT_FROM(float, SUFFIX)
#if defined(cl_khr_fp64) && defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
_CLC_VECTOR_CONVERT_TO1(SUFFIX) \
_CLC_VECTOR_CONVERT_FROM(double, SUFFIX) \
_CLC_VECTOR_CONVERT_FROM(half, SUFFIX)
#elif defined(cl_khr_fp64)
-#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
- _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
+ _CLC_VECTOR_CONVERT_TO1(SUFFIX) \
_CLC_VECTOR_CONVERT_FROM(double, SUFFIX)
#elif defined(cl_khr_fp16)
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
_CLC_VECTOR_CONVERT_TO1(SUFFIX) \
_CLC_VECTOR_CONVERT_FROM(half, SUFFIX)
#else
-#define _CLC_VECTOR_CONVERT_TO(SUFFIX) \
- _CLC_VECTOR_CONVERT_TO1(SUFFIX)
+#define _CLC_VECTOR_CONVERT_TO(SUFFIX) _CLC_VECTOR_CONVERT_TO1(SUFFIX)
#endif
-#define _CLC_VECTOR_CONVERT_TO_SUFFIX(ROUND) \
- _CLC_VECTOR_CONVERT_TO(_sat##ROUND) \
+#define _CLC_VECTOR_CONVERT_TO_SUFFIX(ROUND) \
+ _CLC_VECTOR_CONVERT_TO(_sat##ROUND) \
_CLC_VECTOR_CONVERT_TO(ROUND)
_CLC_VECTOR_CONVERT_TO_SUFFIX(_rtn)
@@ -76,3 +83,11 @@ _CLC_VECTOR_CONVERT_TO_SUFFIX(_rte)
_CLC_VECTOR_CONVERT_TO_SUFFIX(_rtz)
_CLC_VECTOR_CONVERT_TO_SUFFIX(_rtp)
_CLC_VECTOR_CONVERT_TO_SUFFIX()
+
+#undef _CLC_VECTOR_CONVERT_TO_SUFFIX
+#undef _CLC_VECTOR_CONVERT_TO
+#undef _CLC_VECTOR_CONVERT_TO1
+#undef _CLC_VECTOR_CONVERT_FROM
+#undef _CLC_VECTOR_CONVERT_FROM1
+#undef _CLC_VECTOR_CONVERT_DECL
+#undef _CLC_CONVERT_DECL
diff --git a/libclc/generic/include/clc/float/definitions.h b/libclc/generic/include/clc/float/definitions.h
deleted file mode 100644
index be3d0130f3e614c..000000000000000
--- a/libclc/generic/include/clc/float/definitions.h
+++ /dev/null
@@ -1,88 +0,0 @@
-#define MAXFLOAT 0x1.fffffep127f
-#define HUGE_VALF __builtin_huge_valf()
-#define INFINITY __builtin_inff()
-#define NAN __builtin_nanf("")
-
-#define FLT_DIG 6
-#define FLT_MANT_DIG 24
-#define FLT_MAX_10_EXP +38
-#define FLT_MAX_EXP +128
-#define FLT_MIN_10_EXP -37
-#define FLT_MIN_EXP -125
-#define FLT_RADIX 2
-#define FLT_MAX MAXFLOAT
-#define FLT_MIN 0x1.0p-126f
-#define FLT_EPSILON 0x1.0p-23f
-
-#define FP_ILOGB0 (-2147483647 - 1)
-#define FP_ILOGBNAN 2147483647
-
-#define M_E_F 0x1.5bf0a8p+1f
-#define M_LOG2E_F 0x1.715476p+0f
-#define M_LOG10E_F 0x1.bcb7b2p-2f
-#define M_LN2_F 0x1.62e430p-1f
-#define M_LN10_F 0x1.26bb1cp+1f
-#define M_PI_F 0x1.921fb6p+1f
-#define M_PI_2_F 0x1.921fb6p+0f
-#define M_PI_4_F 0x1.921fb6p-1f
-#define M_1_PI_F 0x1.45f306p-2f
-#define M_2_PI_F 0x1.45f306p-1f
-#define M_2_SQRTPI_F 0x1.20dd76p+0f
-#define M_SQRT2_F 0x1.6a09e6p+0f
-#define M_SQRT1_2_F 0x1.6a09e6p-1f
-
-#define M_LOG210_F 0x1.a934f0p+1f
-
-#ifdef cl_khr_fp64
-
-#define HUGE_VAL __builtin_huge_val()
-
-#define DBL_DIG 15
-#define DBL_MANT_DIG 53
-#define DBL_MAX_10_EXP +308
-#define DBL_MAX_EXP +1024
-#define DBL_MIN_10_EXP -307
-#define DBL_MIN_EXP -1021
-#define DBL_MAX 0x1.fffffffffffffp1023
-#define DBL_MIN 0x1.0p-1022
-#define DBL_EPSILON 0x1.0p-52
-
-#define M_E 0x1.5bf0a8b145769p+1
-#define M_LOG2E 0x1.71547652b82fep+0
-#define M_LOG10E 0x1.bcb7b1526e50ep-2
-#define M_LN2 0x1.62e42fefa39efp-1
-#define M_LN10 0x1.26bb1bbb55516p+1
-#define M_PI 0x1.921fb54442d18p+1
-#define M_PI_2 0x1.921fb54442d18p+0
-#define M_PI_4 0x1.921fb54442d18p-1
-#define M_1_PI 0x1.45f306dc9c883p-2
-#define M_2_PI 0x1.45f306dc9c883p-1
-#define M_2_SQRTPI 0x1.20dd750429b6dp+0
-#define M_SQRT2 0x1.6a09e667f3bcdp+0
-#define M_SQRT1_2 0x1.6a09e667f3bcdp-1
-
-#ifdef __CLC_INTERNAL
-#define M_LOG210 0x1.a934f0979a371p+1
-#endif
-
-#endif
-
-#ifdef cl_khr_fp16
-
-#if __OPENCL_VERSION__ >= 120
-
-#define HALF_DIG 3
-#define HALF_MANT_DIG 11
-#define HALF_MAX_10_EXP +4
-#define HALF_MAX_EXP +16
-#define HALF_MIN_10_EXP -4
-#define HALF_MIN_EXP -13
-
-#define HALF_RADIX 2
-#define HALF_MAX 0x1.ffcp15h
-#define HALF_MIN 0x1.0p-14h
-#define HALF_EPSILON 0x1.0p-10h
-
-#endif
-
-#endif
diff --git a/libclc/generic/lib/gen_convert.py b/libclc/generic/lib/gen_convert.py
index d2f69e602a29dac..d90678f83ea48ca 100644
--- a/libclc/generic/lib/gen_convert.py
+++ b/libclc/generic/lib/gen_convert.py
@@ -30,11 +30,15 @@
import argparse
parser = argparse.ArgumentParser()
+parser.add_argument(
+ "--clc", action="store_true", help="Generate clc internal conversions"
+)
parser.add_argument(
"--clspv", action="store_true", help="Generate the clspv variant of the code"
)
args = parser.parse_args()
+clc = args.clc
clspv = args.clspv
types = [
@@ -158,8 +162,30 @@ def conditional_guard(src, dst):
return False
+nl = "\n"
+includes = []
+if not clc:
+ includes = ["<clc/clc.h>"]
+else:
+ includes = sorted([
+ "<clc/internal/clc.h>",
+ "<clc/integer/definitions.h>",
+ "<clc/float/definitions.h>",
+ "<clc/integer/clc_abs.h>",
+ "<clc/common/clc_sign.h>",
+ "<clc/shared/clc_clamp.h>",
+ "<clc/shared/clc_min.h>",
+ "<clc/shared/clc_max.h>",
+ "<clc/math/clc_fabs.h>",
+ "<clc/math/clc_rint.h>",
+ "<clc/math/clc_ceil.h>",
+ "<clc/math/clc_floor.h>",
+ "<clc/math/clc_nextafter.h>",
+ "<clc/relational/clc_select.h>",
+ ])
+
print(
- """/* !!!! AUTOGENERATED FILE generated by convert_type.py !!!!!
+ f"""/* !!!! AUTOGENERATED FILE generated by convert_type.py !!!!!
DON'T CHANGE THIS FILE. MAKE YOUR CHANGES TO convert_type.py AND RUN:
$ ./generate-conversion-type-cl.sh
@@ -188,7 +214,8 @@ def conditional_guard(src, dst):
THE SOFTWARE.
*/
-#include <clc/clc.h>
+{nl.join(['#include ' + f for f in includes])}
+#include <clc/clc_convert.h>
#ifdef cl_khr_fp16
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
@@ -210,6 +237,7 @@ def conditional_guard(src, dst):
"""
)
+
#
# Default Conversions
#
@@ -236,6 +264,13 @@ def conditional_guard(src, dst):
# even for integer-to-integer conversions. When such a conversion
# is used, the rounding mode is ignored.
#
+def print_passthru_conversion(src_ty, dst_ty, fn_name):
+ print(
+ f"""_CLC_DEF _CLC_OVERLOAD {dst_ty} {fn_name}({src_ty} x) {{
+ return __clc_{fn_name}(x);
+}}
+"""
+ )
def generate_default_conversion(src, dst, mode):
@@ -243,26 +278,35 @@ def generate_default_conversion(src, dst, mode):
for size in vector_sizes:
if not size:
- print(
- f"""_CLC_DEF _CLC_OVERLOAD {dst} convert_{dst}{mode}({src} x) {{
+ if clc:
+ print(
+ f"""_CLC_DEF _CLC_OVERLOAD {dst} __clc_convert_{dst}{mode}({src} x) {{
return ({dst})x;
}}
"""
- )
+ )
+ else:
+ print_passthru_conversion(src, dst, f"convert_{dst}{mode}")
else:
- print(
- f"""_CLC_DEF _CLC_OVERLOAD {dst}{size} convert_{dst}{size}{mode}({src}{size} x) {{
+ if clc:
+ print(
+ f"""_CLC_DEF _CLC_OVERLOAD {dst}{size} __clc_convert_{dst}{size}{mode}({src}{size} x) {{
return __builtin_convertvector(x, {dst}{size});
}}
"""
- )
+ )
+ else:
+ print_passthru_conversion(
+ f"{src}{size}", f"{dst}{size}", f"convert_{dst}{size}{mode}"
+ )
if close_conditional:
print("#endif")
-# Do not generate default conversion for clspv as they are handled natively
-if not clspv:
+# Do not generate user-facing default conversions for clspv as they are handled
+# natively
+if clc or not clspv:
for src in types:
for dst in types:
generate_default_conversion(src, dst, "")
@@ -270,15 +314,16 @@ def generate_default_conversion(src, dst, mode):
for src in int_types:
for dst in int_types:
for mode in rounding_modes:
- # Do not generate "_rte" conversion for clspv as they are handled
- # natively
- if clspv and mode == "_rte":
+ # Do not generate user-facing "_rte" conversions for clspv as they
+ # are handled natively
+ if clspv and not clc and mode == "_rte":
continue
generate_default_conversion(src, dst, mode)
#
# Saturated Conversions To Integers
-#
+
+
# These functions are dependent on the unsaturated conversion functions
# generated above, and use clamp, max, min, and select to eliminate
# branching and vectorize the conversions.
@@ -286,35 +331,37 @@ def generate_default_conversion(src, dst, mode):
# Again, as above, we allow all rounding modes for integer-to-integer
# conversions with saturation.
#
-
-
def generate_saturated_conversion(src, dst, size):
# Header
close_conditional = conditional_guard(src, dst)
- print(
- """_CLC_DEF _CLC_OVERLOAD
-{DST}{N} convert_{DST}{N}_sat({SRC}{N} x)
-{{""".format(
- DST=dst, SRC=src, N=size
- )
- )
- # FIXME: This is a work around for lack of select function with
- # signed third argument when the first two arguments are unsigned types.
- # We cast to the signed type for sign-extension, then do a bitcast to
- # the unsigned type.
+ dstn = f"{dst}{size}"
+ srcn = f"{src}{size}"
+
+ if not clc:
+ print_passthru_conversion(f"{srcn}", f"{dstn}", f"convert_{dstn}_sat")
+ if close_conditional:
+ print("#endif")
+ return
+
+ print(f"_CLC_DEF _CLC_OVERLOAD {dstn} __clc_convert_{dstn}_sat({srcn} x) {{")
+
+ # FIXME: This is a work around for lack of select function with signed
+ # third argument when the first two arguments are unsigned types. We cast
+ # to the signed type for sign-extension, then do a bitcast to the unsigned
+ # type.
if dst in unsigned_types:
- bool_prefix = "as_{DST}{N}(convert_{BOOL}{N}".format(
- DST=dst, BOOL=bool_type[dst], N=size
- )
+ bool_prefix = f"__clc_as_{dstn}(__clc_convert_{bool_type[dst]}{size}"
bool_suffix = ")"
else:
- bool_prefix = "convert_{BOOL}{N}".format(BOOL=bool_type[dst], N=size)
+ bool_prefix = f"__clc_convert_{bool_type[dst]}{size}"
bool_suffix = ""
+ dst_max = limit_max[dst]
+ dst_min = limit_min[dst]
+
# Body
if src == dst:
-
# Conversion between same types
print(" return x;")
@@ -323,69 +370,40 @@ def generate_saturated_conversion(src, dst, size):
if clspv:
# Conversion from float to int
print(
- """ {DST}{N} y = convert_{DST}{N}(x);
- y = select(y, ({DST}{N}){DST_MIN}, {BP}(x <= ({SRC}{N}){DST_MIN}){BS});
- y = select(y, ({DST}{N}){DST_MAX}, {BP}(x >= ({SRC}{N}){DST_MAX}){BS});
- return y;""".format(
- SRC=src,
- DST=dst,
- N=size,
- DST_MIN=limit_min[dst],
- DST_MAX=limit_max[dst],
- BP=bool_prefix,
- BS=bool_suffix,
- )
+ f""" {dstn} y = __clc_convert_{dstn}(x);
+ y = __clc_select(y, ({dstn}){dst_min}, {bool_prefix}(x <= ({srcn}){dst_min}){bool_suffix});
+ y = __clc_select(y, ({dstn}){dst_max}, {bool_prefix}(x >= ({srcn}){dst_max}){bool_suffix});
+ return y;"""
)
else:
# Conversion from float to int
print(
- """ {DST}{N} y = convert_{DST}{N}(x);
- y = select(y, ({DST}{N}){DST_MIN}, {BP}(x < ({SRC}{N}){DST_MIN}){BS});
- y = select(y, ({DST}{N}){DST_MAX}, {BP}(x > ({SRC}{N}){DST_MAX}){BS});
- return y;""".format(
- SRC=src,
- DST=dst,
- N=size,
- DST_MIN=limit_min[dst],
- DST_MAX=limit_max[dst],
- BP=bool_prefix,
- BS=bool_suffix,
- )
+ f""" {dstn} y = __clc_convert_{dstn}(x);
+ y = __clc_select(y, ({dstn}){dst_min}, {bool_prefix}(x < ({srcn}){dst_min}){bool_suffix});
+ y = __clc_select(y, ({dstn}){dst_max}, {bool_prefix}(x > ({srcn}){dst_max}){bool_suffix});
+ return y;"""
)
-
else:
# Integer to integer convesion with sizeof(src) == sizeof(dst)
if sizeof_type[src] == sizeof_type[dst]:
if src in unsigned_types:
- print(
- " x = min(x, ({SRC}){DST_MAX});".format(
- SRC=src, DST_MAX=limit_max[dst]
- )
- )
+ print(f" x = __clc_min(x, ({src}){dst_max});")
else:
- print(" x = max(x, ({SRC})0);".format(SRC=src))
+ print(f" x = __clc_max(x, ({src})0);")
# Integer to integer conversion where sizeof(src) > sizeof(dst)
elif sizeof_type[src] > sizeof_type[dst]:
if src in unsigned_types:
- print(
- " x = min(x, ({SRC}){DST_MAX});".format(
- SRC=src, DST_MAX=limit_max[dst]
- )
- )
+ print(f" x = __clc_min(x, ({src}){dst_max});")
else:
- print(
- " x = clamp(x, ({SRC}){DST_MIN}, ({SRC}){DST_MAX});".format(
- SRC=src, DST_MIN=limit_min[dst], DST_MAX=limit_max[dst]
- )
- )
+ print(f" x = __clc_clamp(x, ({src}){dst_min}, ({src}){dst_max});")
# Integer to integer conversion where sizeof(src) < sizeof(dst)
elif src not in unsigned_types and dst in unsigned_types:
- print(" x = max(x, ({SRC})0);".format(SRC=src))
+ print(f" x = __clc_max(x, ({src})0);")
- print(" return convert_{DST}{N}(x);".format(DST=dst, N=size))
+ print(f" return __clc_convert_{dstn}(x);")
# Footer
print("}")
@@ -403,17 +421,19 @@ def generate_saturated_conversion_with_rounding(src, dst, size, mode):
# Header
close_conditional = conditional_guard(src, dst)
- # Body
- print(
- """_CLC_DEF _CLC_OVERLOAD
-{DST}{N} convert_{DST}{N}_sat{M}({SRC}{N} x)
-{{
- return convert_{DST}{N}_sat(x);
+ dstn = f"{dst}{size}"
+ srcn = f"{src}{size}"
+
+ if not clc:
+ print_passthru_conversion(f"{srcn}", f"{dstn}", f"convert_{dstn}_sat{mode}")
+ else:
+ # Body
+ print(
+ f"""_CLC_DEF _CLC_OVERLOAD {dstn} __clc_convert_{dstn}_sat{mode}({srcn} x) {{
+ return __clc_convert_{dstn}_sat(x);
}}
-""".format(
- DST=dst, SRC=src, N=size, M=mode
+"""
)
- )
# Footer
if close_conditional:
@@ -426,6 +446,7 @@ def generate_saturated_conversion_with_rounding(src, dst, size, mode):
for mode in rounding_modes:
generate_saturated_conversion_with_rounding(src, dst, size, mode)
+
#
# Conversions To/From Floating-Point With Rounding
#
@@ -439,134 +460,90 @@ def generate_saturated_conversion_with_rounding(src, dst, size, mode):
#
# Only conversions to integers can have saturation.
#
-
-
def generate_float_conversion(src, dst, size, mode, sat):
# Header
close_conditional = conditional_guard(src, dst)
- print(
- """_CLC_DEF _CLC_OVERLOAD
-{DST}{N} convert_{DST}{N}{S}{M}({SRC}{N} x)
-{{""".format(
- SRC=src, DST=dst, N=size, M=mode, S=sat
- )
- )
+
+ dstn = f"{dst}{size}"
+ srcn = f"{src}{size}"
+ booln = f"{bool_type[dst]}{size}"
+ src_max = limit_max[src] if src in limit_max else ""
+ dst_min = limit_min[dst] if dst in limit_min else ""
+
+ if not clc:
+ print_passthru_conversion(f"{srcn}", f"{dstn}", f"convert_{dstn}{sat}{mode}")
+ # Footer
+ if close_conditional:
+ print("#endif")
+ return
+
+ print(f"_CLC_DEF _CLC_OVERLOAD {dstn} __clc_convert_{dstn}{sat}{mode}({srcn} x) {{")
# Perform conversion
if dst in int_types:
if mode == "_rte":
- print(" x = rint(x);")
+ print(" x = __clc_rint(x);")
elif mode == "_rtp":
- print(" x = ceil(x);")
+ print(" x = __clc_ceil(x);")
elif mode == "_rtn":
- print(" x = floor(x);")
- print(" return convert_{DST}{N}{S}(x);".format(DST=dst, N=size, S=sat))
+ print(" x = __clc_floor(x);")
+ print(f" return __clc_convert_{dstn}{sat}(x);")
elif mode == "_rte":
- print(" return convert_{DST}{N}(x);".format(DST=dst, N=size))
+ print(f" return __clc_convert_{dstn}(x);")
else:
- print(" {DST}{N} r = convert_{DST}{N}(x);".format(DST=dst, N=size))
+ print(f" {dstn} r = __clc_convert_{dstn}(x);")
if clspv:
- print(" {SRC}{N} y = convert_{SRC}{N}_sat(r);".format(SRC=src, N=size))
+ print(f" {srcn} y = __clc_convert_{srcn}_sat(r);")
else:
- print(" {SRC}{N} y = convert_{SRC}{N}(r);".format(SRC=src, N=size))
+ print(f" {srcn} y = __clc_convert_{srcn}(r);")
if mode == "_rtz":
if src in int_types:
- print(
- " {USRC}{N} abs_x = abs(x);".format(
- USRC=unsigned_type[src], N=size
- )
- )
- print(
- " {USRC}{N} abs_y = abs(y);".format(
- USRC=unsigned_type[src], N=size
- )
- )
- else:
- print(" {SRC}{N} abs_x = fabs(x);".format(SRC=src, N=size))
- print(" {SRC}{N} abs_y = fabs(y);".format(SRC=src, N=size))
- if clspv:
- print(
- " {BOOL}{N} c = convert_{BOOL}{N}(abs_y > abs_x);".format(
- BOOL=bool_type[dst], N=size
- )
- )
- if sizeof_type[src] >= 4 and src in int_types:
- print(
- " c = c || convert_{BOOL}{N}(({SRC}{N}){SRC_MAX} == x);".format(
- BOOL=bool_type[dst], N=size, SRC=src, SRC_MAX=limit_max[src]
- )
- )
- print(
- " {DST}{N} sel = select(r, nextafter(r, sign(r) * ({DST}{N})-INFINITY), c);".format(
- DST=dst, N=size, BOOL=bool_type[dst], SRC=src
- )
- )
+ usrcn = f"{unsigned_type[src]}{size}"
+ print(f" {usrcn} abs_x = __clc_abs(x);")
+ print(f" {usrcn} abs_y = __clc_abs(y);")
else:
- print(
- " {DST}{N} sel = select(r, nextafter(r, sign(r) * ({DST}{N})-INFINITY), convert_{BOOL}{N}(abs_y > abs_x));".format(
- DST=dst, N=size, BOOL=bool_type[dst]
- )
- )
+ print(f" {srcn} abs_x = __clc_fabs(x);")
+ print(f" {srcn} abs_y = __clc_fabs(y);")
+ print(f" {booln} c = __clc_convert_{booln}(abs_y > abs_x);")
+ if clspv and sizeof_type[src] >= 4 and src in int_types:
+ print(f" c = c || __clc_convert_{booln}(({srcn}){src_max} == x);")
+ print(
+ f" {dstn} sel = __clc_select(r, __clc_nextafter(r, __clc_sign(r) * ({dstn})-INFINITY), c);"
+ )
if dst == "half" and src in int_types and sizeof_type[src] >= 2:
dst_max = limit_max[dst]
- # short is 16 bits signed, so the maximum value rounded to zero is 0x1.ffcp+14 (0x1p+15 == 32768 > 0x7fff == 32767)
+ # short is 16 bits signed, so the maximum value rounded to zero
+ # is 0x1.ffcp+14 (0x1p+15 == 32768 > 0x7fff == 32767)
if src == "short":
dst_max = "0x1.ffcp+14"
print(
- " return clamp(sel, ({DST}{N}){DST_MIN}, ({DST}{N}){DST_MAX});".format(
- DST=dst, N=size, DST_MIN=limit_min[dst], DST_MAX=dst_max
- )
+ f" return __clc_clamp(sel, ({dstn}){dst_min}, ({dstn}){dst_max});"
)
else:
print(" return sel;")
if mode == "_rtp":
print(
- " {DST}{N} sel = select(r, nextafter(r, ({DST}{N})INFINITY), convert_{BOOL}{N}(y < x));".format(
- DST=dst, N=size, BOOL=bool_type[dst]
- )
+ f" {dstn} sel = __clc_select(r, __clc_nextafter(r, ({dstn})INFINITY), __clc_convert_{booln}(y < x));"
)
if dst == "half" and src in int_types and sizeof_type[src] >= 2:
- print(
- " return max(sel, ({DST}{N}){DST_MIN});".format(
- DST=dst, N=size, DST_MIN=limit_min[dst]
- )
- )
+ print(f" return __clc_max(sel, ({dstn}){dst_min});")
else:
print(" return sel;")
if mode == "_rtn":
- if clspv:
- print(
- " {BOOL}{N} c = convert_{BOOL}{N}(y > x);".format(
- BOOL=bool_type[dst], N=size
- )
- )
- if sizeof_type[src] >= 4 and src in int_types:
- print(
- " c = c || convert_{BOOL}{N}(({SRC}{N}){SRC_MAX} == x);".format(
- BOOL=bool_type[dst], N=size, SRC=src, SRC_MAX=limit_max[src]
- )
- )
- print(
- " {DST}{N} sel = select(r, nextafter(r, ({DST}{N})-INFINITY), c);".format(
- DST=dst, N=size, BOOL=bool_type[dst], SRC=src
- )
- )
- else:
- print(
- " {DST}{N} sel = select(r, nextafter(r, ({DST}{N})-INFINITY), convert_{BOOL}{N}(y > x));".format(
- DST=dst, N=size, BOOL=bool_type[dst]
- )
- )
+ print(f" {booln} c = __clc_convert_{booln}(y > x);")
+ if clspv and sizeof_type[src] >= 4 and src in int_types:
+ print(f" c = c || __clc_convert_{booln}(({srcn}){src_max} == x);")
+ print(
+ f" {dstn} sel = __clc_select(r, __clc_nextafter(r, ({dstn})-INFINITY), c);"
+ )
if dst == "half" and src in int_types and sizeof_type[src] >= 2:
dst_max = limit_max[dst]
- # short is 16 bits signed, so the maximum value rounded to negative infinity is 0x1.ffcp+14 (0x1p+15 == 32768 > 0x7fff == 32767)
+ # short is 16 bits signed, so the maximum value rounded to
+ # negative infinity is 0x1.ffcp+14 (0x1p+15 == 32768 > 0x7fff
+ # == 32767)
if src == "short":
dst_max = "0x1.ffcp+14"
- print(
- " return min(sel, ({DST}{N}){DST_MAX});".format(
- DST=dst, N=size, DST_MAX=dst_max
- )
- )
+ print(f" return __clc_min(sel, ({dstn}){dst_max});")
else:
print(" return sel;")
@@ -588,8 +565,8 @@ def generate_float_conversion(src, dst, size, mode, sat):
for dst in float_types:
for size in vector_sizes:
for mode in rounding_modes:
- # Do not generate "_rte" conversion for clspv as they are
- # handled natively
- if clspv and mode == "_rte":
+ # Do not generate user-facing "_rte" conversions for clspv as
+ # they are handled natively
+ if clspv and not clc and mode == "_rte":
continue
generate_float_conversion(src, dst, size, mode, "")
>From 585b1f7dac004e132e4becdf0c57bf4f73dbbc9c Mon Sep 17 00:00:00 2001
From: Fraser Cormack <fraser at codeplay.com>
Date: Tue, 28 Jan 2025 10:44:09 +0000
Subject: [PATCH 2/2] fix formatting
---
libclc/generic/lib/gen_convert.py | 34 ++++++++++++++++---------------
1 file changed, 18 insertions(+), 16 deletions(-)
diff --git a/libclc/generic/lib/gen_convert.py b/libclc/generic/lib/gen_convert.py
index d90678f83ea48ca..92f32694a52afd1 100644
--- a/libclc/generic/lib/gen_convert.py
+++ b/libclc/generic/lib/gen_convert.py
@@ -167,22 +167,24 @@ def conditional_guard(src, dst):
if not clc:
includes = ["<clc/clc.h>"]
else:
- includes = sorted([
- "<clc/internal/clc.h>",
- "<clc/integer/definitions.h>",
- "<clc/float/definitions.h>",
- "<clc/integer/clc_abs.h>",
- "<clc/common/clc_sign.h>",
- "<clc/shared/clc_clamp.h>",
- "<clc/shared/clc_min.h>",
- "<clc/shared/clc_max.h>",
- "<clc/math/clc_fabs.h>",
- "<clc/math/clc_rint.h>",
- "<clc/math/clc_ceil.h>",
- "<clc/math/clc_floor.h>",
- "<clc/math/clc_nextafter.h>",
- "<clc/relational/clc_select.h>",
- ])
+ includes = sorted(
+ [
+ "<clc/internal/clc.h>",
+ "<clc/integer/definitions.h>",
+ "<clc/float/definitions.h>",
+ "<clc/integer/clc_abs.h>",
+ "<clc/common/clc_sign.h>",
+ "<clc/shared/clc_clamp.h>",
+ "<clc/shared/clc_min.h>",
+ "<clc/shared/clc_max.h>",
+ "<clc/math/clc_fabs.h>",
+ "<clc/math/clc_rint.h>",
+ "<clc/math/clc_ceil.h>",
+ "<clc/math/clc_floor.h>",
+ "<clc/math/clc_nextafter.h>",
+ "<clc/relational/clc_select.h>",
+ ]
+ )
print(
f"""/* !!!! AUTOGENERATED FILE generated by convert_type.py !!!!!
More information about the cfe-commits
mailing list