adjust unit tests for introduced nthreads max

Author: omlins

test/test_parallel.jl

@@ -25,13 +25,13 @@ import ParallelStencil.@gorgeousexpand
             @require !@is_initialized()
             @init_parallel_stencil($package, Float64, 3)
             @require @is_initialized()
-            @testset "@parallel <kernelcall>" begin # NOTE: calls must go to ParallelStencil.ParallelKernel.parallel and must therefore give the same result as in ParallelKernel (tests copied 1-to-1 from there).
+            @testset "@parallel <kernelcall>" begin # NOTE: calls must go to ParallelStencil.ParallelKernel.parallel and must therefore give the same result as in ParallelKernel, except for memopt tests (tests copied 1-to-1 from there).
                 @static if $package == $PKG_CUDA
                     call = @prettystring(1, @parallel f(A))
-                    @test occursin("CUDA.@cuda blocks = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))))) threads = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))) stream = CUDA.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[3])))", call)
+                    @test occursin("CUDA.@cuda blocks = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))); nthreads_x_max = 32)) threads = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))); nthreads_x_max = 32) stream = CUDA.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[3])))", call)
                     @test occursin("CUDA.synchronize(CUDA.stream())", call)
                     call = @prettystring(1, @parallel ranges f(A))
-                    @test occursin("CUDA.@cuda blocks = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)))) threads = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges))) stream = CUDA.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ranges), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[3])))", call)
+                    @test occursin("CUDA.@cuda blocks = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)); nthreads_x_max = 32)) threads = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)); nthreads_x_max = 32) stream = CUDA.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ranges), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[3])))", call)
                     call = @prettystring(1, @parallel nblocks nthreads f(A))
                     @test occursin("CUDA.@cuda blocks = nblocks threads = nthreads stream = CUDA.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)))[3])))", call)
                     call = @prettystring(1, @parallel ranges nblocks nthreads f(A))
@@ -44,10 +44,10 @@ import ParallelStencil.@gorgeousexpand
                     @test occursin("CUDA.@cuda blocks = ParallelStencil.ParallelKernel.compute_nblocks(cld.(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)),", call) # NOTE: now it is a very long multi line expression; before it continued as follows: (1, 1, 16)), ParallelStencil.compute_nthreads_memopt(cld.(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), (1, 1, 16)), 3, (-1:1, -1:1, -1:1))) threads = ParallelStencil.compute_nthreads_memopt(cld.(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), (1, 1, 16)), 3, (-1:1, -1:1, -1:1)) stream = CUDA.stream() shmem = ((ParallelStencil.compute_nthreads_memopt(cld.(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), (1, 1, 16)), 3, (-1:1, -1:1, -1:1)))[1] + 3) * ((ParallelStencil.compute_nthreads_memopt(cld.(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), (1, 1, 16)), 3, (-1:1, -1:1, -1:1)))[2] + 3) * sizeof(Float64) f(A, ParallelStencil.ParallelKernel.promote_ranges(ranges), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[3])))", call)
                 elseif $package == $PKG_AMDGPU
                     call = @prettystring(1, @parallel f(A))
-                    @test occursin("AMDGPU.@roc gridsize = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))))) groupsize = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))) stream = AMDGPU.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[3])))", call)
+                    @test occursin("AMDGPU.@roc gridsize = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))); nthreads_x_max = 64)) groupsize = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A))); nthreads_x_max = 64) stream = AMDGPU.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.get_ranges(A)))[3])))", call)
                     @test occursin("AMDGPU.synchronize(AMDGPU.stream())", call)
                     call = @prettystring(1, @parallel ranges f(A))
-                    @test occursin("AMDGPU.@roc gridsize = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)))) groupsize = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges))) stream = AMDGPU.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ranges), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[3])))", call)
+                    @test occursin("AMDGPU.@roc gridsize = ParallelStencil.ParallelKernel.compute_nblocks(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)), ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)); nthreads_x_max = 64)) groupsize = ParallelStencil.ParallelKernel.compute_nthreads(length.(ParallelStencil.ParallelKernel.promote_ranges(ranges)); nthreads_x_max = 64) stream = AMDGPU.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ranges), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ranges))[3])))", call)
                     call = @prettystring(1, @parallel nblocks nthreads f(A))
                     @test occursin("AMDGPU.@roc gridsize = nblocks groupsize = nthreads stream = AMDGPU.stream() f(A, ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)))[1])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)))[2])), (Int64)(length((ParallelStencil.ParallelKernel.promote_ranges(ParallelStencil.ParallelKernel.compute_ranges(nblocks .* nthreads)))[3])))", call)
                     call = @prettystring(1, @parallel ranges nblocks nthreads f(A))