According to opencl spec, The mandated minimum single precision floating-point capability given by CL_DEVICE_SINGLE_FP_CONFIG is CL_FP_ROUND_TO_ZERO or CL_FP_ROUND_TO_NEAREST. We set the single float mode to IEEE 754 and set the rounding mode to RTN.
Signed-off-by: Zhigang Gong <[email protected]> --- src/cl_gt_device.h | 2 ++ src/intel/intel_gpgpu.c | 2 ++ 2 files changed, 4 insertions(+) diff --git a/src/cl_gt_device.h b/src/cl_gt_device.h index f58e1fd..7f67697 100644 --- a/src/cl_gt_device.h +++ b/src/cl_gt_device.h @@ -62,6 +62,8 @@ .execution_capabilities = CL_EXEC_KERNEL, .queue_properties = CL_QUEUE_PROFILING_ENABLE, .platform = NULL, /* == intel_platform (set when requested) */ +/* IEEE 754, XXX does IVB support CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT? */ +.single_fp_config = CL_FP_DENORM | CL_FP_INF_NAN | CL_FP_ROUND_TO_NEAREST , /* IEEE 754. */ #define DECL_INFO_STRING(FIELD, STRING) \ .FIELD = STRING, \ diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c index 1301b66..073e255 100644 --- a/src/intel/intel_gpgpu.c +++ b/src/intel/intel_gpgpu.c @@ -608,6 +608,8 @@ intel_gpgpu_build_idrt(intel_gpgpu_t *gpgpu, cl_gpgpu_kernel *kernel) ker_bo = (drm_intel_bo *) kernel->bo; desc->desc0.kernel_start_pointer = ker_bo->offset >> 6; /* reloc */ desc->desc1.single_program_flow = 1; + desc->desc1.floating_point_mode = 0; /* use IEEE-754 rule */ + desc->desc5.rounding_mode = 0; /* round to nearest even */ desc->desc2.sampler_state_pointer = gpgpu->sampler_state_b.bo->offset >> 5; desc->desc3.binding_table_entry_count = 0; /* no prefetch */ desc->desc3.binding_table_pointer = 0; -- 1.7.9.5 _______________________________________________ Beignet mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/beignet
