[libclc] Move (add|sub)_sat to CLC; optimize

libclc/clc/include/clc/integer/clc_add_sat.h

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

libclc/clc/include/clc/integer/clc_sub_sat.h

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

libclc/clc/lib/clspv/SOURCES

@@ -1,3 +1,5 @@
+../generic/integer/clc_add_sat.cl
+../generic/integer/clc_sub_sat.cl
 ../generic/math/clc_ceil.cl
 ../generic/math/clc_copysign.cl
 ../generic/math/clc_fabs.cl

libclc/clc/lib/generic/SOURCES

@@ -4,6 +4,8 @@ common/clc_smoothstep.cl
 geometric/clc_dot.cl
 integer/clc_abs.cl
 integer/clc_abs_diff.cl
+integer/clc_add_sat.cl
+integer/clc_sub_sat.cl
 math/clc_ceil.cl
 math/clc_copysign.cl
 math/clc_fabs.cl

libclc/clc/lib/generic/integer/clc_add_sat.cl

@@ -0,0 +1,7 @@
+#include <clc/internal/clc.h>
+
+#define FUNCTION __clc_add_sat
+#define __CLC_FUNCTION(x) __builtin_elementwise_add_sat
+#define __CLC_BODY <clc/shared/binary_def.inc>
+
+#include <clc/integer/gentype.inc>

libclc/clc/lib/generic/integer/clc_sub_sat.cl

@@ -0,0 +1,7 @@
+#include <clc/internal/clc.h>
+
+#define FUNCTION __clc_sub_sat
+#define __CLC_FUNCTION(x) __builtin_elementwise_sub_sat
+#define __CLC_BODY <clc/shared/binary_def.inc>
+
+#include <clc/integer/gentype.inc>

libclc/clc/lib/spirv/SOURCES

@@ -2,6 +2,8 @@
 ../generic/common/clc_radians.cl
 ../generic/common/clc_smoothstep.cl
 ../generic/geometric/clc_dot.cl
+../generic/integer/clc_add_sat.cl
+../generic/integer/clc_sub_sat.cl
 ../generic/math/clc_ceil.cl
 ../generic/math/clc_copysign.cl
 ../generic/math/clc_fabs.cl

libclc/clc/lib/spirv64/SOURCES

@@ -2,6 +2,8 @@
 ../generic/common/clc_radians.cl
 ../generic/common/clc_smoothstep.cl
 ../generic/geometric/clc_dot.cl
+../generic/integer/clc_add_sat.cl
+../generic/integer/clc_sub_sat.cl
 ../generic/math/clc_ceil.cl
 ../generic/math/clc_copysign.cl
 ../generic/math/clc_fabs.cl

libclc/generic/lib/integer/add_sat.cl

@@ -1,73 +1,7 @@
 #include <clc/clc.h>
-#include <clc/clcmacro.h>
+#include <clc/integer/clc_add_sat.h>
 
-// From add_sat.ll
-_CLC_DECL char   __clc_add_sat_s8(char, char);
-_CLC_DECL uchar  __clc_add_sat_u8(uchar, uchar);
-_CLC_DECL short  __clc_add_sat_s16(short, short);
-_CLC_DECL ushort __clc_add_sat_u16(ushort, ushort);
-_CLC_DECL int    __clc_add_sat_s32(int, int);
-_CLC_DECL uint   __clc_add_sat_u32(uint, uint);
-_CLC_DECL long   __clc_add_sat_s64(long, long);
-_CLC_DECL ulong  __clc_add_sat_u64(ulong, ulong);
+#define FUNCTION add_sat
+#define __CLC_BODY <clc/shared/binary_def.inc>
 
-_CLC_OVERLOAD _CLC_DEF char add_sat(char x, char y) {
-  short r = x + y;
-  return convert_char_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF uchar add_sat(uchar x, uchar y) {
-  ushort r = x + y;
-  return convert_uchar_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF short add_sat(short x, short y) {
-  int r = x + y;
-  return convert_short_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF ushort add_sat(ushort x, ushort y) {
-  uint r = x + y;
-  return convert_ushort_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF int add_sat(int x, int y) {
-  int r;
-  if (__builtin_sadd_overflow(x, y, &r))
-    // The oveflow can only occur if both are pos or both are neg,
-    // thus we only need to check one operand
-    return x > 0 ? INT_MAX : INT_MIN;
-  return r;
-}
-
-_CLC_OVERLOAD _CLC_DEF uint add_sat(uint x, uint y) {
-  uint r;
-  if (__builtin_uadd_overflow(x, y, &r))
-	return UINT_MAX;
-  return r;
-}
-
-_CLC_OVERLOAD _CLC_DEF long add_sat(long x, long y) {
-  long r;
-  if (__builtin_saddl_overflow(x, y, &r))
-    // The oveflow can only occur if both are pos or both are neg,
-    // thus we only need to check one operand
-    return x > 0 ? LONG_MAX : LONG_MIN;
-  return r;
-}
-
-_CLC_OVERLOAD _CLC_DEF ulong add_sat(ulong x, ulong y) {
-  ulong r;
-  if (__builtin_uaddl_overflow(x, y, &r))
-	return ULONG_MAX;
-  return r;
-}
-
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, char, add_sat, char, char)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, uchar, add_sat, uchar, uchar)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, short, add_sat, short, short)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, ushort, add_sat, ushort, ushort)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, int, add_sat, int, int)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, uint, add_sat, uint, uint)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, long, add_sat, long, long)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, ulong, add_sat, ulong, ulong)
+#include <clc/integer/gentype.inc>

libclc/generic/lib/integer/sub_sat.cl

@@ -1,61 +1,7 @@
 #include <clc/clc.h>
-#include <clc/clcmacro.h>
+#include <clc/integer/clc_sub_sat.h>
 
-_CLC_OVERLOAD _CLC_DEF char sub_sat(char x, char y) {
-  short r = x - y;
-  return convert_char_sat(r);
-}
+#define FUNCTION sub_sat
+#define __CLC_BODY <clc/shared/binary_def.inc>
 
-_CLC_OVERLOAD _CLC_DEF uchar sub_sat(uchar x, uchar y) {
-  short r = x - y;
-  return convert_uchar_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF short sub_sat(short x, short y) {
-  int r = x - y;
-  return convert_short_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF ushort sub_sat(ushort x, ushort y) {
-  int r = x - y;
-  return convert_ushort_sat(r);
-}
-
-_CLC_OVERLOAD _CLC_DEF int sub_sat(int x, int y) {
-  int r;
-  if (__builtin_ssub_overflow(x, y, &r))
-    // The oveflow can only occur in the direction of the first operand
-    return x > 0 ? INT_MAX : INT_MIN;
-  return r;
-}
-
-_CLC_OVERLOAD _CLC_DEF uint sub_sat(uint x, uint y) {
-  uint r;
-  if (__builtin_usub_overflow(x, y, &r))
-	return 0;
-  return r;
-}
-
-_CLC_OVERLOAD _CLC_DEF long sub_sat(long x, long y) {
-  long r;
-  if (__builtin_ssubl_overflow(x, y, &r))
-    // The oveflow can only occur in the direction of the first operand
-    return x > 0 ? LONG_MAX : LONG_MIN;
-  return r;
-}
-
-_CLC_OVERLOAD _CLC_DEF ulong sub_sat(ulong x, ulong y) {
-  ulong r;
-  if (__builtin_usubl_overflow(x, y, &r))
-	return 0;
-  return r;
-}
-
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, char, sub_sat, char, char)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, uchar, sub_sat, uchar, uchar)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, short, sub_sat, short, short)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, ushort, sub_sat, ushort, ushort)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, int, sub_sat, int, int)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, uint, sub_sat, uint, uint)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, long, sub_sat, long, long)
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, ulong, sub_sat, ulong, ulong)
+#include <clc/integer/gentype.inc>

libclc/generic/lib/math/clc_ldexp.cl

@@ -22,6 +22,7 @@
 
 #include <clc/clc.h>
 #include <clc/clcmacro.h>
+#include <clc/integer/clc_add_sat.h>
 #include <clc/math/clc_subnormal_config.h>
 #include <clc/math/math.h>
 #include <clc/relational/clc_isinf.h>
@@ -37,7 +38,7 @@ _CLC_DEF _CLC_OVERLOAD float __clc_ldexp(float x, int n) {
     int e = (i >> 23) & 0xff;
     int m = i & 0x007fffff;
     int s = i & 0x80000000;
-    int v = add_sat(e, n);
+    int v = __clc_add_sat(e, n);
     v = __clc_clamp(v, 0, 0xff);
     int mr = e == 0 | v == 0 | v == 0xff ? 0 : m;
     int c = e == 0xff;