diff options
author | Jan Vesely <jan.vesely@rutgers.edu> | 2018-06-07 20:27:58 +0000 |
---|---|---|
committer | Jan Vesely <jan.vesely@rutgers.edu> | 2018-06-07 20:27:58 +0000 |
commit | e0edcaa4a95bf9979ee31de635824c7125799a86 (patch) | |
tree | ce5639b7fcf1d31c326140ff9dfa94c729e5436f /libclc/r600 | |
parent | e23c0ec0863870bc4823c731091f478c4bc781cf (diff) | |
download | llvm-e0edcaa4a95bf9979ee31de635824c7125799a86.zip llvm-e0edcaa4a95bf9979ee31de635824c7125799a86.tar.gz llvm-e0edcaa4a95bf9979ee31de635824c7125799a86.tar.bz2 |
r600/fmin: Flush denormals before calling builtin.
Same reason as amdgcn.
Fixes fmin, minmag CTS on turks.
Reviewer: Tom Stellard <tstellar@redhat.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 334228
Diffstat (limited to 'libclc/r600')
-rw-r--r-- | libclc/r600/lib/SOURCES | 1 | ||||
-rw-r--r-- | libclc/r600/lib/math/fmin.cl | 30 |
2 files changed, 31 insertions, 0 deletions
diff --git a/libclc/r600/lib/SOURCES b/libclc/r600/lib/SOURCES index 132b151..e69be4a 100644 --- a/libclc/r600/lib/SOURCES +++ b/libclc/r600/lib/SOURCES @@ -1,4 +1,5 @@ math/fmax.cl +math/fmin.cl synchronization/barrier_impl.ll workitem/get_global_offset.cl workitem/get_group_id.cl diff --git a/libclc/r600/lib/math/fmin.cl b/libclc/r600/lib/math/fmin.cl new file mode 100644 index 0000000..09f1e4c --- /dev/null +++ b/libclc/r600/lib/math/fmin.cl @@ -0,0 +1,30 @@ +#include <clc/clc.h> + +#include "../../../generic/lib/clcmacro.h" +#include "../../../generic/lib/math/math.h" + +_CLC_DEF _CLC_OVERLOAD float 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 = __clc_flush_denormal_if_not_supported(x); + y = __clc_flush_denormal_if_not_supported(y); + return __builtin_fminf(x, y); +} +_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, fmin, float, float) + +#ifdef cl_khr_fp64 + +#pragma OPENCL EXTENSION cl_khr_fp64 : enable + +_CLC_DEF _CLC_OVERLOAD double fmin(double x, double y) +{ + return __builtin_fmin(x, y); +} +_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, fmin, double, double) + +#endif + +#define __CLC_BODY <../../../generic/lib/math/fmin.inc> +#include <clc/math/gentype.inc> |