[Beignet] [PATCH 1/2] [opencl-1.2] Add the support for 1D image from buffer.
junyan.he at inbox.com
junyan.he at inbox.com
Thu May 15 23:52:46 PDT 2014
From: Junyan He <junyan.he at linux.intel.com>
Signed-off-by: Junyan He <junyan.he at linux.intel.com>
---
backend/src/ocl_stdlib.tmpl.h | 1 +
src/cl_device_id.c | 1 +
src/cl_device_id.h | 1 +
src/cl_gt_device.h | 1 +
src/cl_mem.c | 148 +++++++++++++++++++++++++++++++++++++++++-
src/cl_mem.h | 1 +
src/intel/intel_gpgpu.c | 3 +-
7 files changed, 154 insertions(+), 2 deletions(-)
diff --git a/backend/src/ocl_stdlib.tmpl.h b/backend/src/ocl_stdlib.tmpl.h
index 8ab8b31..83a5b5b 100755
--- a/backend/src/ocl_stdlib.tmpl.h
+++ b/backend/src/ocl_stdlib.tmpl.h
@@ -92,6 +92,7 @@ typedef __texture struct _image3d_t* __image3d_t;
typedef const ushort __sampler_t;
typedef size_t __event_t;
#define image1d_t __image1d_t
+#define image1d_buffer_t __image1d_t
#define image2d_t __image2d_t
#define image3d_t __image3d_t
#define sampler_t __sampler_t
diff --git a/src/cl_device_id.c b/src/cl_device_id.c
index 2b443c6..f60ad51 100644
--- a/src/cl_device_id.c
+++ b/src/cl_device_id.c
@@ -327,6 +327,7 @@ cl_get_device_info(cl_device_id device,
DECL_FIELD(GLOBAL_MEM_CACHE_SIZE, global_mem_cache_size)
DECL_FIELD(GLOBAL_MEM_SIZE, global_mem_size)
DECL_FIELD(MAX_CONSTANT_BUFFER_SIZE, max_constant_buffer_size)
+ DECL_FIELD(IMAGE_MAX_BUFFER_SIZE, image_mem_size)
DECL_FIELD(MAX_CONSTANT_ARGS, max_constant_args)
DECL_FIELD(LOCAL_MEM_TYPE, local_mem_type)
DECL_FIELD(LOCAL_MEM_SIZE, local_mem_size)
diff --git a/src/cl_device_id.h b/src/cl_device_id.h
index 5f7c9fe..2bbe98e 100644
--- a/src/cl_device_id.h
+++ b/src/cl_device_id.h
@@ -55,6 +55,7 @@ struct _cl_device_id {
size_t image3d_max_width;
size_t image3d_max_height;
size_t image3d_max_depth;
+ cl_ulong image_mem_size;
cl_uint max_samplers;
size_t max_parameter_size;
cl_uint mem_base_addr_align;
diff --git a/src/cl_gt_device.h b/src/cl_gt_device.h
index 110988a..327e2a9 100644
--- a/src/cl_gt_device.h
+++ b/src/cl_gt_device.h
@@ -46,6 +46,7 @@
.image3d_max_width = 8192,
.image3d_max_height = 8192,
.image3d_max_depth = 2048,
+.image_mem_size = 8192,
.max_samplers = 16,
.mem_base_addr_align = sizeof(cl_long) * 16 * 8,
.min_data_type_align_size = sizeof(cl_long) * 16,
diff --git a/src/cl_mem.c b/src/cl_mem.c
index 0250f0a..08e4f66 100644
--- a/src/cl_mem.c
+++ b/src/cl_mem.c
@@ -151,10 +151,17 @@ cl_get_image_info(cl_mem mem,
FIELD_SIZE(IMAGE_WIDTH, size_t);
FIELD_SIZE(IMAGE_HEIGHT, size_t);
FIELD_SIZE(IMAGE_DEPTH, size_t);
+ FIELD_SIZE(IMAGE_BUFFER, cl_mem);
default:
return CL_INVALID_VALUE;
}
+ /* Do some further check. */
+ if (param_name == CL_IMAGE_BUFFER &&
+ image->image_type != CL_MEM_OBJECT_IMAGE1D_BUFFER) {
+ return CL_INVALID_VALUE;
+ }
+
switch(param_name)
{
case CL_IMAGE_FORMAT:
@@ -178,6 +185,9 @@ cl_get_image_info(cl_mem mem,
case CL_IMAGE_DEPTH:
*(size_t *)param_value = image->depth;
break;
+ case CL_IMAGE_BUFFER:
+ *(cl_mem *)param_value = image->buffer_1d;
+ break;
}
return CL_SUCCESS;
@@ -641,6 +651,131 @@ error:
goto exit;
}
+static cl_mem
+_cl_mem_new_image_from_buffer(cl_context ctx,
+ cl_mem_flags flags,
+ const cl_image_format* image_format,
+ const cl_image_desc *image_desc,
+ cl_int *errcode_ret)
+{
+ cl_mem image = NULL;
+ cl_mem buffer = image_desc->buffer;
+ cl_int err = CL_SUCCESS;
+ *errcode_ret = err;
+ cl_ulong max_size;
+ cl_mem_flags merged_flags;
+ uint32_t bpp;
+ uint32_t intel_fmt = INTEL_UNSUPPORTED_FORMAT;
+ size_t offset = 0;
+
+ /* Get the size of each pixel */
+ if (UNLIKELY((err = cl_image_byte_per_pixel(image_format, &bpp)) != CL_SUCCESS))
+ goto error;
+
+ /* Only a sub-set of the formats are supported */
+ intel_fmt = cl_image_get_intel_format(image_format);
+ if (UNLIKELY(intel_fmt == INTEL_UNSUPPORTED_FORMAT)) {
+ err = CL_INVALID_IMAGE_FORMAT_DESCRIPTOR;
+ goto error;
+ }
+
+ if (!buffer) {
+ err = CL_INVALID_IMAGE_DESCRIPTOR;
+ goto error;
+ }
+
+ if (flags & (CL_MEM_USE_HOST_PTR|CL_MEM_ALLOC_HOST_PTR|CL_MEM_COPY_HOST_PTR)) {
+ err = CL_INVALID_IMAGE_DESCRIPTOR;
+ goto error;
+ }
+
+ /* access check. */
+ if ((buffer->flags & CL_MEM_WRITE_ONLY) &&
+ (flags & (CL_MEM_READ_WRITE|CL_MEM_READ_ONLY))) {
+ err = CL_INVALID_VALUE;
+ goto error;
+ }
+ if ((buffer->flags & CL_MEM_READ_ONLY) &&
+ (flags & (CL_MEM_READ_WRITE|CL_MEM_WRITE_ONLY))) {
+ err = CL_INVALID_VALUE;
+ goto error;
+ }
+ if ((buffer->flags & CL_MEM_HOST_WRITE_ONLY) &&
+ (flags & CL_MEM_HOST_READ_ONLY)) {
+ err = CL_INVALID_VALUE;
+ goto error;
+ }
+ if ((buffer->flags & CL_MEM_HOST_READ_ONLY) &&
+ (flags & CL_MEM_HOST_WRITE_ONLY)) {
+ err = CL_INVALID_VALUE;
+ goto error;
+ }
+ if ((buffer->flags & CL_MEM_HOST_NO_ACCESS) &&
+ (flags & (CL_MEM_HOST_READ_ONLY | CL_MEM_HOST_WRITE_ONLY))) {
+ err = CL_INVALID_VALUE;
+ goto error;
+ }
+
+ if ((err = cl_get_device_info(ctx->device,
+ CL_DEVICE_IMAGE_MAX_BUFFER_SIZE,
+ sizeof(max_size),
+ &max_size,
+ NULL)) != CL_SUCCESS) {
+ goto error;
+ }
+
+ if (image_desc->image_width > max_size) {
+ err = CL_INVALID_IMAGE_DESCRIPTOR;
+ goto error;
+ }
+
+ if (image_desc->image_width*bpp > buffer->size) {
+ err = CL_INVALID_IMAGE_DESCRIPTOR;
+ goto error;
+ }
+
+ merged_flags = buffer->flags;
+ if (flags & (CL_MEM_READ_WRITE|CL_MEM_READ_WRITE|CL_MEM_WRITE_ONLY)) {
+ merged_flags &= ~(CL_MEM_READ_WRITE|CL_MEM_READ_WRITE|CL_MEM_WRITE_ONLY);
+ merged_flags |= flags & (CL_MEM_READ_WRITE|CL_MEM_READ_WRITE|CL_MEM_WRITE_ONLY);
+ }
+ if (flags & (CL_MEM_HOST_WRITE_ONLY|CL_MEM_HOST_READ_ONLY|CL_MEM_HOST_NO_ACCESS)) {
+ merged_flags &= ~(CL_MEM_HOST_WRITE_ONLY|CL_MEM_HOST_READ_ONLY|CL_MEM_HOST_NO_ACCESS);
+ merged_flags |= flags & (CL_MEM_HOST_WRITE_ONLY|CL_MEM_HOST_READ_ONLY|CL_MEM_HOST_NO_ACCESS);
+ }
+
+ /* Because the buffer is NO_TILING, the image should be no tiling. */
+ image = cl_mem_allocate(CL_MEM_IMAGE_TYPE, ctx, flags, merged_flags, CL_FALSE, &err);
+ if (image == NULL || err != CL_SUCCESS)
+ goto error;
+
+ cl_buffer_reference(buffer->bo);
+ image->bo = buffer->bo;
+ image->size = buffer->size;
+ /* If it is a sub buffer, we need to start from the sub offset. */
+ if (buffer->type == CL_MEM_SUBBUFFER_TYPE) {
+ offset = ((struct _cl_mem_buffer *)buffer)->sub_offset;
+ }
+ if (image->flags & CL_MEM_USE_HOST_PTR) {
+ /* Now point to the right offset if buffer is a SUB_BUFFER. */
+ image->host_ptr = buffer->host_ptr + offset;
+ }
+
+ cl_mem_image_init(cl_mem_image(image), image_desc->image_width, 1, image_desc->image_type,
+ 1, *image_format, intel_fmt, bpp, image_desc->image_width*bpp, 0, CL_NO_TILE,
+ 0, 0, offset);
+ cl_mem_add_ref(buffer);
+ cl_mem_image(image)->buffer_1d = buffer;
+ return image;
+
+error:
+ if (image)
+ cl_mem_delete(image);
+ image = NULL;
+ *errcode_ret = err;
+ return image;
+}
+
LOCAL cl_mem
cl_mem_new_image(cl_context context,
cl_mem_flags flags,
@@ -659,8 +794,10 @@ cl_mem_new_image(cl_context context,
host_ptr, errcode_ret);
case CL_MEM_OBJECT_IMAGE2D_ARRAY:
case CL_MEM_OBJECT_IMAGE1D_ARRAY:
- case CL_MEM_OBJECT_IMAGE1D_BUFFER:
NOT_IMPLEMENTED;
+ case CL_MEM_OBJECT_IMAGE1D_BUFFER:
+ return _cl_mem_new_image_from_buffer(context, flags, image_format,
+ image_desc, errcode_ret);
break;
case CL_MEM_OBJECT_BUFFER:
default:
@@ -683,6 +820,15 @@ cl_mem_delete(cl_mem mem)
}
#endif
+ /* iff we are a image, delete the 1d buffer if has. */
+ if (IS_IMAGE(mem)) {
+ if (cl_mem_image(mem)->buffer_1d) {
+ assert(cl_mem_image(mem)->image_type == CL_MEM_OBJECT_IMAGE1D_BUFFER);
+ cl_mem_delete(cl_mem_image(mem)->buffer_1d);
+ cl_mem_image(mem)->buffer_1d = NULL;
+ }
+ }
+
/* Remove it from the list */
assert(mem->ctx);
pthread_mutex_lock(&mem->ctx->buffer_lock);
diff --git a/src/cl_mem.h b/src/cl_mem.h
index afa48fe..a3dcdc6 100644
--- a/src/cl_mem.h
+++ b/src/cl_mem.h
@@ -83,6 +83,7 @@ struct _cl_mem_image {
cl_image_tiling_t tiling; /* only IVB+ supports TILE_[X,Y] (image only) */
size_t tile_x, tile_y; /* tile offset, used for mipmap images. */
size_t offset; /* offset for dri_bo, used when it's reloc. */
+ cl_mem buffer_1d; /* if the image is created from buffer, it point to the buffer.*/
};
struct _cl_mem_gl_image {
diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c
index 014b95f..7a24854 100644
--- a/src/intel/intel_gpgpu.c
+++ b/src/intel/intel_gpgpu.c
@@ -570,10 +570,11 @@ static int
intel_get_surface_type(cl_mem_object_type type)
{
switch (type) {
+ case CL_MEM_OBJECT_IMAGE1D_BUFFER:
case CL_MEM_OBJECT_IMAGE1D: return I965_SURFACE_1D;
+
case CL_MEM_OBJECT_IMAGE2D: return I965_SURFACE_2D;
case CL_MEM_OBJECT_IMAGE3D: return I965_SURFACE_3D;
- case CL_MEM_OBJECT_IMAGE1D_BUFFER:
case CL_MEM_OBJECT_IMAGE2D_ARRAY:
case CL_MEM_OBJECT_IMAGE1D_ARRAY:
NOT_IMPLEMENTED;
--
1.8.3.2
More information about the Beignet
mailing list