[SYCL] Use full compilation for the multi-device AOT test (#603)

* [SYCL] Use full compilation for the multi-device AOT test

Since the introduction of the original test, many changes have been
made:
- the tests' runtime and the compilation time have been reasonably
  improved;
- the compiler pipeline has been complicated by a bunch of DPC++
  tools;
- default compiler behavior for many features includes dependency on
  device libraries, which makes the test increasingly hard to
  maintain.

Therefore, it makes sense to refrain from running the compiler steps
manually from now on, relying on E2E compilation as our users
would. We will produce a single fat object for all supported device
targets + host, and then rely on the driver/tools to unbundle the
object & compile only for the needed targets.

Drop requirements for standalone tools

Signed-off-by: Artem Gindinson <[email protected]>

Author: Artem Gindinson

SYCL/AOT/multiple-devices.cpp

@@ -6,74 +6,54 @@
 //
 //===----------------------------------------------------------------------===//
 
-// REQUIRES: opencl-aot, ocloc, aoc, cpu, gpu, accelerator, llvm-link, llvm-spirv
+// REQUIRES: opencl-aot, ocloc, aoc, cpu, gpu, accelerator
 // UNSUPPORTED: cuda
 // CUDA is not compatible with SPIR.
 
-// 1-command compilation case
-// Targeting CPU, GPU, FPGA
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen,spir64_fpga -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %S/Inputs/aot.cpp -o %t_all.out
-// RUN: %HOST_RUN_PLACEHOLDER %t_all.out
-// RUN: %CPU_RUN_PLACEHOLDER %t_all.out
-// RUN: %GPU_RUN_PLACEHOLDER %t_all.out
-// RUN: %ACC_RUN_PLACEHOLDER %t_all.out
+// Produce a fat object for all targets (generic SPIR-V, CPU, GPU, FPGA)
+// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_x86_64,spir64_gen,spir64_fpga %S/Inputs/aot.cpp -c -o %t.o
 
-// FIXME: Change the behavior when proper automation for assert support is
-// introduced. For the time being, AOT flow can't detect if specific extension
-// is available for this or that device. The automation to be introduced is to
-// query native binary generator on specific features.
-//
-// Produce object file, spirv, device images to combine these differently
-// at link-time, thus testing various AOT-compiled images configurations
-// Disable DAE when performing separate compilation here, as we are not
-// using sycl-post-link to fully enable.
-// RUN: %clangxx -fsycl -fno-sycl-dead-args-optimization %S/Inputs/aot.cpp -c -o %t.o
-// RUN: %clangxx -fsycl -fsycl-link-targets=spir64 %t.o -o %t.spv
-// AOT-compile device binary images
-// Neither of AOT tools can compile several files, hence, here is this
-// workaround
-// RUN: %llvm_spirv -r %sycl_libs_dir/libsycl-fallback-cassert.spv -o=%T/fallback-cassert.bc
-// RUN: %llvm_spirv -r %t.spv -o=%t.bc
-// RUN: %llvm_link %t.bc %T/fallback-cassert.bc -o=%t2.bc
-// RUN: %llvm_spirv %t2.bc -o=%t.spv
-// RUN: opencl-aot %t.spv -o=%t_cpu.ir --device=cpu
-// RUN: ocloc -file %t.spv -spirv_input -output %t_gen.out -output_no_suffix -device cfl
-// RUN: aoc %t.spv -o %t_fpga.aocx -sycl -dep-files=%t.d
+// CPU, GPU, FPGA
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen,spir64_fpga -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %t.o -o %t_all_aot.out
+// RUN: %HOST_RUN_PLACEHOLDER %t_all_aot.out
+// RUN: %CPU_RUN_PLACEHOLDER %t_all_aot.out
+// RUN: %GPU_RUN_PLACEHOLDER %t_all_aot.out
+// RUN: %ACC_RUN_PLACEHOLDER %t_all_aot.out
 
 // CPU, GPU
-// RUN: %clangxx -fsycl -fsycl-add-targets=spir64_x86_64:%t_cpu.ir,spir64_gen:%t_gen.out %t.o -o %t_cpu_gpu.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %t.o -o %t_cpu_gpu.out
 // RUN: %HOST_RUN_PLACEHOLDER %t_cpu_gpu.out
 // RUN: %CPU_RUN_PLACEHOLDER %t_cpu_gpu.out
 // RUN: %GPU_RUN_PLACEHOLDER %t_cpu_gpu.out
 
 // CPU, FPGA
-// RUN: %clangxx -fsycl -fsycl-add-targets=spir64_x86_64:%t_cpu.ir,spir64_fpga:%t_fpga.aocx %t.o -o %t_cpu_fpga.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_fpga %t.o -o %t_cpu_fpga.out
 // RUN: %HOST_RUN_PLACEHOLDER %t_cpu_fpga.out
 // RUN: %CPU_RUN_PLACEHOLDER %t_cpu_fpga.out
 // RUN: %ACC_RUN_PLACEHOLDER %t_cpu_fpga.out
 
 // GPU, FPGA
-// RUN: %clangxx -fsycl -fsycl-add-targets=spir64_gen:%t_gen.out,spir64_fpga:%t_fpga.aocx %t.o -o %t_gpu_fpga.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen,spir64_fpga -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %t.o -o %t_gpu_fpga.out
 // RUN: %HOST_RUN_PLACEHOLDER %t_gpu_fpga.out
 // RUN: %GPU_RUN_PLACEHOLDER %t_gpu_fpga.out
 // RUN: %ACC_RUN_PLACEHOLDER %t_gpu_fpga.out
 
 // No AOT-compiled image for CPU
-// RUN: %clangxx -fsycl -fsycl-add-targets=spir64:%t.spv,spir64_gen:%t_gen.out,spir64_fpga:%t_fpga.aocx %t.o -o %t_spv_gpu_fpga.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_gen,spir64_fpga -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %t.o -o %t_spv_gpu_fpga.out
 // RUN: %CPU_RUN_PLACEHOLDER %t_spv_gpu_fpga.out
 // Check that execution on AOT-compatible devices is unaffected
 // RUN: %GPU_RUN_PLACEHOLDER %t_spv_gpu_fpga.out
 // RUN: %ACC_RUN_PLACEHOLDER %t_spv_gpu_fpga.out
 
 // No AOT-compiled image for GPU
-// RUN: %clangxx -fsycl -fsycl-add-targets=spir64:%t.spv,spir64_x86_64:%t_cpu.ir,spir64_fpga:%t_fpga.aocx %t.o -o %t_spv_cpu_fpga.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_x86_64,spir64_fpga %t.o -o %t_spv_cpu_fpga.out
 // RUN: %GPU_RUN_PLACEHOLDER %t_spv_cpu_fpga.out
 // Check that execution on AOT-compatible devices is unaffected
 // RUN: %CPU_RUN_PLACEHOLDER %t_spv_cpu_fpga.out
 // RUN: %ACC_RUN_PLACEHOLDER %t_spv_cpu_fpga.out
 
 // No AOT-compiled image for FPGA
-// RUN: %clangxx -fsycl -fsycl-add-targets=spir64:%t.spv,spir64_x86_64:%t_cpu.ir,spir64_gen:%t_gen.out %t.o -o %t_spv_cpu_gpu.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %t.o -o %t_spv_cpu_gpu.out
 // RUN: %ACC_RUN_PLACEHOLDER %t_spv_cpu_gpu.out
 // Check that execution on AOT-compatible devices is unaffected
 // RUN: %CPU_RUN_PLACEHOLDER %t_spv_cpu_gpu.out

SYCL/lit.cfg.py

@@ -326,17 +326,6 @@
 if find_executable('sycl-ls'):
     config.available_features.add('sycl-ls')
 
-llvm_tools = ["llvm-spirv", "llvm-link"]
-for llvm_tool in llvm_tools:
-  llvm_tool_path = find_executable(llvm_tool)
-  if llvm_tool_path:
-    lit_config.note("Found " + llvm_tool)
-    config.available_features.add(llvm_tool)
-    config.substitutions.append( ('%' + llvm_tool.replace('-', '_'),
-                                  os.path.realpath(llvm_tool_path)) )
-  else:
-    lit_config.warning("Can't find " + llvm_tool)
-
 # Device AOT compilation tools aren't part of the SYCL project,
 # so they need to be pre-installed on the machine
 aot_tools = ["ocloc", "aoc", "opencl-aot"]