CUDA: Implement frexp, tgamma, ldexp, modf, remquo #836
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Closed
isuruf
changed the title
pjaaskel
requested changes
May 28, 2020
|
@pjaaskel, the formatter doesn't like the new lines I've added. diff --git a/lib/kernel/templates.h b/lib/kernel/templates.h
index d9fd018c..e0d1df9a 100644
--- a/lib/kernel/templates.h
+++ b/lib/kernel/templates.h
@@ -1878,21 +1878,22 @@
IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, VTYPE, __global)
#define DEFINE_BUILTIN_V_VPV(NAME) \
__IF_FP16 ( \
- half _CL_OVERLOADABLE _CL_READNONE NAME (half a, __private half *b) \
- { \
- /* use float builtin */ \
- __private float c; \
- __private float r = __builtin_##NAME##f (a, &c); \
- *b = c; \
- return r; \
- } \
- IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, half, __local) \
- IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, half, __global) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, half2, half, half, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, half3, half2, half, lo, s2) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, half4, half2, half2, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, half8, half4, half4, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, half16, half8, half8, lo, hi)) \
+ half _CL_OVERLOADABLE _CL_READNONE NAME (half a, __private half *b) { \
+ /* use float builtin */ \
+ __private float c; \
+ __private float r = __builtin_##NAME##f (a, &c); \
+ *b = c; \
+ return r; \
+ } IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, half, __local) \
+ IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, half, __global) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, half2, half, half, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, half3, half2, half, lo, s2) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, half4, half2, half2, lo, \
+ hi) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, half8, half4, half4, \
+ lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, half16, half8, \
+ half8, lo, hi)) \
float _CL_OVERLOADABLE _CL_READNONE NAME (float a, __private float *b) \
{ \
return __builtin_##NAME##f (a, b); \
@@ -1905,17 +1906,20 @@
IMPLEMENT_BUILTIN_V_VPV (NAME, float8, float4, float4, lo, hi) \
IMPLEMENT_BUILTIN_V_VPV (NAME, float16, float8, float8, lo, hi) \
__IF_FP64 ( \
- double _CL_OVERLOADABLE _CL_READNONE NAME (double a, __private double *b) \
- { \
- return __builtin_##NAME (a, b); \
- } \
- IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, double, __local) \
- IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, double, __global) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, double2, double, double, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, double3, double2, double, lo, s2) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, double4, double2, double2, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, double8, double4, double4, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPV (NAME, double16, double8, double8, lo, hi))
+ double _CL_OVERLOADABLE _CL_READNONE NAME (double a, \
+ __private double *b) { \
+ return __builtin_##NAME (a, b); \
+ } IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, double, __local) \
+ IMPLEMENT_BUILTIN_V_VPV_ADDRSPACE (NAME, double, __global) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, double2, double, double, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, double3, double2, double, \
+ lo, s2) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, double4, double2, \
+ double2, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPV (NAME, double8, double4, \
+ double4, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPV ( \
+ NAME, double16, double8, double8, lo, hi))
#define IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE(NAME, VTYPE, JTYPE, ADDRSPACE) \
VTYPE __attribute__ ((overloadable)) \
@@ -1941,21 +1945,23 @@
IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, VTYPE, JTYPE, __global)
#define DEFINE_BUILTIN_V_VVPJ(NAME) \
__IF_FP16 ( \
- half _CL_OVERLOADABLE _CL_READNONE NAME (half a, half b, int *c) \
- { \
- /* use float builtin */ \
- __private int d; \
- __private float r = __builtin_##NAME##f (a, b, &d); \
- *c = d; \
- return r; \
- } \
- IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, half, int, __local) \
- IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, half, int, __global) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, half2, int2, int, int, lo, hi) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, half3, int3, int2, int, lo, s2) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, half4, int4, int2, int2, lo, hi) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, half8, int8, int4, int4, lo, hi) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, half16, int16, int8, int8, lo, hi)) \
+ half _CL_OVERLOADABLE _CL_READNONE NAME (half a, half b, int *c) { \
+ /* use float builtin */ \
+ __private int d; \
+ __private float r = __builtin_##NAME##f (a, b, &d); \
+ *c = d; \
+ return r; \
+ } IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, half, int, __local) \
+ IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, half, int, __global) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, half2, int2, int, int, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, half3, int3, int2, int, lo, \
+ s2) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, half4, int4, int2, \
+ int2, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, half8, int8, int4, \
+ int4, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, half16, int16, \
+ int8, int8, lo, hi)) \
float _CL_OVERLOADABLE _CL_READNONE NAME (float a, float b, int *c) \
{ \
return __builtin_##NAME##f (a, b, c); \
@@ -1968,17 +1974,21 @@
IMPLEMENT_BUILTIN_V_VVPJ (NAME, float8, int8, int4, int4, lo, hi) \
IMPLEMENT_BUILTIN_V_VVPJ (NAME, float16, int16, int8, int8, lo, hi) \
__IF_FP64 ( \
- double _CL_OVERLOADABLE _CL_READNONE NAME (double a, double b, int *c) \
- { \
- return __builtin_##NAME (a, b, c); \
- } \
- IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, double, int, __local) \
- IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, double, int, __global) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, double2, int2, int, int, lo, hi) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, double3, int3, int2, int, lo, s2) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, double4, int4, int2, int2, lo, hi) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, double8, int8, int4, int4, lo, hi) \
- IMPLEMENT_BUILTIN_V_VVPJ (NAME, double16, int16, int8, int8, lo, hi))
+ double _CL_OVERLOADABLE _CL_READNONE NAME (double a, double b, \
+ int *c) { \
+ return __builtin_##NAME (a, b, c); \
+ } IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, double, int, __local) \
+ IMPLEMENT_BUILTIN_V_VVPJ_ADDRSPACE (NAME, double, int, __global) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, double2, int2, int, int, lo, \
+ hi) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, double3, int3, int2, int, \
+ lo, s2) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, double4, int4, int2, \
+ int2, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VVPJ (NAME, double8, int8, \
+ int4, int4, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VVPJ ( \
+ NAME, double16, int16, int8, int8, lo, hi))
#define IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE(NAME, VTYPE, JTYPE, ADDRSPACE) \
VTYPE __attribute__ ((overloadable)) NAME (VTYPE a, ADDRSPACE JTYPE *c) \
@@ -2003,21 +2013,23 @@
IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, VTYPE, JTYPE, __global)
#define DEFINE_BUILTIN_V_VPJ(NAME) \
__IF_FP16 ( \
- half _CL_OVERLOADABLE _CL_READNONE NAME (half a, int *c) \
- { \
- /* use float builtin */ \
- __private int d; \
- __private float r = __builtin_##NAME##f (a, &d); \
- *c = d; \
- return r; \
- } \
- IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, half, int, __local) \
- IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, half, int, __global) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, half2, int2, int, int, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, half3, int3, int2, int, lo, s2) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, half4, int4, int2, int2, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, half8, int8, int4, int4, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, half16, int16, int8, int8, lo, hi)) \
+ half _CL_OVERLOADABLE _CL_READNONE NAME (half a, int *c) { \
+ /* use float builtin */ \
+ __private int d; \
+ __private float r = __builtin_##NAME##f (a, &d); \
+ *c = d; \
+ return r; \
+ } IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, half, int, __local) \
+ IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, half, int, __global) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, half2, int2, int, int, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, half3, int3, int2, int, lo, \
+ s2) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, half4, int4, int2, int2, \
+ lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, half8, int8, int4, \
+ int4, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, half16, int16, \
+ int8, int8, lo, hi)) \
float _CL_OVERLOADABLE _CL_READNONE NAME (float a, int *c) \
{ \
return __builtin_##NAME##f (a, c); \
@@ -2030,17 +2042,19 @@
IMPLEMENT_BUILTIN_V_VPJ (NAME, float8, int8, int4, int4, lo, hi) \
IMPLEMENT_BUILTIN_V_VPJ (NAME, float16, int16, int8, int8, lo, hi) \
__IF_FP64 ( \
- double _CL_OVERLOADABLE _CL_READNONE NAME (double a, int *c) \
- { \
- return __builtin_##NAME (a, c); \
- } \
- IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, double, int, __local) \
- IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, double, int, __global) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, double2, int2, int, int, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, double3, int3, int2, int, lo, s2) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, double4, int4, int2, int2, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, double8, int8, int4, int4, lo, hi) \
- IMPLEMENT_BUILTIN_V_VPJ (NAME, double16, int16, int8, int8, lo, hi))
+ double _CL_OVERLOADABLE _CL_READNONE NAME (double a, int *c) { \
+ return __builtin_##NAME (a, c); \
+ } IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, double, int, __local) \
+ IMPLEMENT_BUILTIN_V_VPJ_ADDRSPACE (NAME, double, int, __global) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, double2, int2, int, int, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, double3, int3, int2, int, \
+ lo, s2) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, double4, int4, int2, \
+ int2, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, double8, int8, int4, \
+ int4, lo, hi) \
+ IMPLEMENT_BUILTIN_V_VPJ (NAME, double16, int16, \
+ int8, int8, lo, hi))
#define __SINGLE_WI \
if (get_local_id(0) == 0 && \ |
Yeah, that's what I guessed: the macro magic is too complex for it. So, pls just revert it manually for now. |
pjaaskel
approved these changes
May 29, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #835