[libclc] Enable clang fp reciprocal
in clc_native_divide/recip/rsqrt/tan (#149269)
The pragma adds `arcp` flag to `fdiv` instruction in these functions. The flag can provide better performance.
This commit is contained in:
parent
0b6df5485e
commit
cf36f49c04
@ -8,5 +8,6 @@
|
||||
|
||||
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_native_divide(__CLC_GENTYPE x,
|
||||
__CLC_GENTYPE y) {
|
||||
_Pragma("clang fp reciprocal(on)");
|
||||
return x / y;
|
||||
}
|
||||
|
@ -7,5 +7,6 @@
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_native_recip(__CLC_GENTYPE val) {
|
||||
_Pragma("clang fp reciprocal(on)");
|
||||
return 1.0f / val;
|
||||
}
|
||||
|
@ -7,5 +7,6 @@
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_native_rsqrt(__CLC_GENTYPE val) {
|
||||
_Pragma("clang fp reciprocal(on)");
|
||||
return 1.0f / __clc_native_sqrt(val);
|
||||
}
|
||||
|
@ -7,5 +7,6 @@
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_native_tan(__CLC_GENTYPE val) {
|
||||
_Pragma("clang fp reciprocal(on)");
|
||||
return __clc_native_sin(val) / __clc_native_cos(val);
|
||||
}
|
||||
|
Loading…
x
Reference in New Issue
Block a user