[Beignet] [PATCH V2 1/7] Runtime: disable some unecessary image formats.
He Junyan
junyan.he at inbox.com
Tue Sep 17 22:32:42 PDT 2013
As you said, compiler_copy_image1 will fail because of the GPU
limitation, we may overcome it by some hack of software in the future.
and these patch set is LGTM,
On Fri, 2013-09-13 at 14:01 +0800, Zhigang Gong wrote:
> Per OpenCL, the minimum list of supported format is as below:
> CL_RGBA:
> CL_UNORM_INT8
> CL_UNORM_INT16
> CL_SIGNED_INT8
> CL_SIGNED_INT16
> CL_SIGNED_INT32
> CL_UNSIGNED_INT8
> CL_UNSIGNED_INT16
> CL_UNSIGNED_INT32
> CL_HALF_FLOAT
> CL_FLOAT
>
> CL_BGRA:
> CL_UNORM_INT8
>
> Let's only support this type and CL_R currently.
>
> Also removed an unnecessary assertion. And fix the CL_Rx's type size.
>
> Signed-off-by: Zhigang Gong <zhigang.gong at linux.intel.com>
> ---
> src/cl_image.c | 20 +++++++++++++-------
> 1 file changed, 13 insertions(+), 7 deletions(-)
>
> diff --git a/src/cl_image.c b/src/cl_image.c
> index 1fe8d08..f89bcae 100644
> --- a/src/cl_image.c
> +++ b/src/cl_image.c
> @@ -61,11 +61,11 @@ cl_image_byte_per_pixel(const cl_image_format *fmt, uint32_t *bpp)
> };
>
> switch (order) {
> + case CL_Rx: break;
> case CL_R: break;
> case CL_A: break;
> case CL_RA: *bpp *= 2; break;
> case CL_RG: *bpp *= 2; break;
> - case CL_Rx: *bpp *= 2; break;
> case CL_INTENSITY:
> case CL_LUMINANCE:
> if (type != CL_UNORM_INT8 && type != CL_UNORM_INT16 &&
> @@ -101,6 +101,8 @@ cl_image_get_intel_format(const cl_image_format *fmt)
> const uint32_t order = fmt->image_channel_order;
> switch (order) {
> case CL_R:
> +#if 0
> + case CL_Rx:
> case CL_A:
> case CL_INTENSITY:
> case CL_LUMINANCE:
> @@ -109,12 +111,16 @@ cl_image_get_intel_format(const cl_image_format *fmt)
> && type != CL_SNORM_INT8 && type != CL_SNORM_INT16
> && type != CL_HALF_FLOAT && type != CL_FLOAT))
> return INTEL_UNSUPPORTED_FORMAT;
> +#endif
> +
> +/* XXX it seems we have some acuracy compatible issue with snomr_int8/16,
> + * have to disable those formats currently. */
>
> switch (type) {
> case CL_HALF_FLOAT: return I965_SURFACEFORMAT_R16_FLOAT;
> case CL_FLOAT: return I965_SURFACEFORMAT_R32_FLOAT;
> - case CL_SNORM_INT16: return I965_SURFACEFORMAT_R16_SNORM;
> - case CL_SNORM_INT8: return I965_SURFACEFORMAT_R8_SNORM;
> +// case CL_SNORM_INT16: return I965_SURFACEFORMAT_R16_SNORM;
> +// case CL_SNORM_INT8: return I965_SURFACEFORMAT_R8_SNORM;
> case CL_UNORM_INT8: return I965_SURFACEFORMAT_R8_UNORM;
> case CL_UNORM_INT16: return I965_SURFACEFORMAT_R16_UNORM;
> case CL_SIGNED_INT8: return I965_SURFACEFORMAT_R8_SINT;
> @@ -125,9 +131,9 @@ cl_image_get_intel_format(const cl_image_format *fmt)
> case CL_UNSIGNED_INT32: return I965_SURFACEFORMAT_R32_UINT;
> default: return INTEL_UNSUPPORTED_FORMAT;
> };
> +#if 0
> case CL_RG:
> case CL_RA:
> - case CL_Rx:
> switch (type) {
> case CL_HALF_FLOAT: return I965_SURFACEFORMAT_R16G16_FLOAT;
> case CL_FLOAT: return I965_SURFACEFORMAT_R32G32_FLOAT;
> @@ -151,12 +157,13 @@ cl_image_get_intel_format(const cl_image_format *fmt)
> case CL_UNORM_SHORT_555:
> default: return INTEL_UNSUPPORTED_FORMAT;
> };
> +#endif
> case CL_RGBA:
> switch (type) {
> case CL_HALF_FLOAT: return I965_SURFACEFORMAT_R16G16B16A16_FLOAT;
> case CL_FLOAT: return I965_SURFACEFORMAT_R32G32B32A32_FLOAT;
> - case CL_SNORM_INT16: return I965_SURFACEFORMAT_R16G16B16A16_SNORM;
> - case CL_SNORM_INT8: return I965_SURFACEFORMAT_R8G8B8A8_SNORM;
> +// case CL_SNORM_INT16: return I965_SURFACEFORMAT_R16G16B16A16_SNORM;
> +// case CL_SNORM_INT8: return I965_SURFACEFORMAT_R8G8B8A8_SNORM;
> case CL_UNORM_INT8: return I965_SURFACEFORMAT_R8G8B8A8_UNORM;
> case CL_UNORM_INT16: return I965_SURFACEFORMAT_R16G16B16A16_UNORM;
> case CL_SIGNED_INT8: return I965_SURFACEFORMAT_R8G8B8A8_SINT;
> @@ -201,7 +208,6 @@ cl_image_get_supported_fmt(cl_context ctx,
> cl_uint *num_image_formats)
> {
> size_t i, j, n = 0;
> - assert(image_formats);
> for (i = 0; i < cl_image_order_n; ++i)
> for (j = 0; j < cl_image_type_n; ++j) {
> const cl_image_format fmt = {
More information about the Beignet
mailing list