[oneMKL samples][computed tomography] Revisions, refactoring and addition of a functional verification

Libraries/oneMKL/computed_tomography/GNUmakefile

@@ -5,7 +5,7 @@ default: run
 all: run
 
 run: computed_tomography
-	./computed_tomography 400 400 input.bmp radon.bmp restored.bmp
+	./computed_tomography
 
 MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl=dft
 
@@ -15,6 +15,6 @@ computed_tomography: computed_tomography.cpp
 	icpx $< -fsycl -o $@ $(DPCPP_OPTS)
 
 clean:
-	-rm -f computed_tomography radon.bmp restored.bmp
+	-rm -f computed_tomography radon.bmp restored.bmp errors.bmp
 
 .PHONY: clean run all

Libraries/oneMKL/computed_tomography/README.md

@@ -16,15 +16,13 @@ For more information on oneMKL and complete documentation of all oneMKL routines
 
 Computed Tomography uses oneMKL discrete Fourier transform (DFT) routines to transform simulated raw CT data (as collected by a CT scanner) into a reconstructed image of the scanned object.
 
-In computed tomography, the raw imaging data is a set of line integrals over the actual object, also known as its _Radon transform_. From this data, the original image must be recovered by approximately inverting the Radon transform. This sample uses the filtered back projection method for inverting the Radon transform, which involves a 1D DFT, followed by filtering, then an inverse 2D DFT to perform the final reconstruction.
+In computed tomography, the raw imaging data is a set of line integrals over the actual object, also known as its _Radon transform_. From this data, the original image must be recovered by approximately inverting the Radon transform. This sample uses Fourier reconstruction for inverting the Radon transform of a user-provided input image: using batched 1D real DFT of Radon transform data points, samples of the input image's Fourier spectrum may be estimated on a polar grid; after interpolating the latter onto a Cartesian grid, an inverse 2D real DFT produces a fair reproduction of the original image.
 
-This sample performs its computations on the default SYCL* device. You can set the `SYCL_DEVICE_TYPE` environment variable to `cpu` or `gpu` to select the device to use.
-
-This article explains in detail how oneMKL fast Fourier transform (FFT) functions can be used to reconstruct the original image from the Computer Tomography (CT) data: https://www.intel.com/content/www/us/en/docs/onemkl/cookbook/current/ffts-for-computer-tomography-image-reconstruction.html.
+This sample performs its computations on the default SYCL device. You can set the `ONEAPI_DEVICE_SELECTOR` environment variable to `*:cpu` or `*:gpu` to select the device to use.
 
 ## Key Implementation Details
 
-To use oneMKL DFT routines, the sample creates a descriptor object for the given precision and domain (real-to-complex or complex-to-complex), calls the `commit` method, and provides a `sycl::queue` object to define the device and context. The `compute_*` routines are then called to perform the actual computation with the appropriate descriptor object and input/output buffers.
+To use oneMKL DFT routines, the sample creates double-precision real DFT descriptor objects, calls the `commit` member function with a `sycl::queue` object to define the device and context. The `compute_*` routines are then called to perform the actual computation with the appropriate descriptor object and input data.
 
 ## Using Visual Studio Code* (Optional)
 You can use Visual Studio Code (VS Code) extensions to set your environment, create launch configurations,
@@ -70,18 +68,18 @@ Run `nmake` to build and run the sample. `nmake clean` removes temporary files.
 ## Running the Computed Tomography Reconstruction Sample
 
 ### Example of Output
-If everything is working correctly, the example program will start with the 400x400 example image `input.bmp` then create simulated CT data from it (stored as `restored.bmp`). It will then reconstruct the original image in grayscale and store it as `restored.bmp`.
+If everything is working correctly, the example program will start with the 400x400 example image `input.bmp` then create simulated CT data from it (stored as `radon.bmp`). It will then reconstruct the original image in grayscale and store it as `restored.bmp`.
 
 ```
-./computed_tomography 400 400 input.bmp radon.bmp restored.bmp
+./computed_tomography
 Reading original image from input.bmp
-Allocating radonImage for backprojection
-Performing backprojection
-Restoring original: step1 - fft_1d in-place
-Allocating array for radial->cartesian interpolation
-Restoring original: step2 - interpolation
-Restoring original: step3 - ifft_2d in-place
-Saving restored image to restored.bmp
+Generating Radon transform data from input.bmp
+Saving Radon transform data in radon.bmp
+Reconstructing image from the Radon projection data
+        Step 1 - Batch of 400 real 1D in-place forward DFTs of length 400
+        Step 2 - Interpolating spectrum from polar to cartesian grid
+        Step 3 - In-place backward real 2D DFT of size 400x400
+Saving restored image in restored.bmp
 ```
 
 ### Troubleshooting