Update benchmarks problem size

CliMA · Oct 3, 2024 · 62913a2 · 62913a2
1 parent f3064ba
commit 62913a2
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 36 deletions.
diff --git a/test/execution/bm_fused_reads_vs_hard_coded.jl b/test/execution/bm_fused_reads_vs_hard_coded.jl
@@ -42,7 +42,9 @@ use_cuda = @isdefined(CUDA) && CUDA.has_cuda() # will be true if you first run `
     function knl_multi_copyto_hard_coded!(X, Y, ::Val{nitems}) where {nitems}
         (; x1, x2, x3, x4, x5, x6, x7, x8, x9, x10) = X
         (; y1, y2, y3, y4, y5, y6, y7, y8, y9, y10) = Y
-        idx = CUDA.threadIdx().x + (CUDA.blockIdx().x - 1) * CUDA.blockDim().x
+        idx =
+            CUDA.threadIdx().x +
+            (CUDA.blockIdx().x - Int32(1)) * CUDA.blockDim().x
         @inbounds begin
             if idx ≤ nitems
                 y1[idx] = x1[idx] + x2[idx] + x3[idx] + x4[idx]
@@ -95,7 +97,7 @@ AType = use_cuda ? CUDA.CuArray : Array
 device_name = use_cuda ? CUDA.name(CUDA.device()) : "CPU"
 bm = Benchmark(; device_name, float_type = Float32)
 
-problem_size = (50, 5, 5, 6, 50)
+problem_size = (50, 5, 5, 6, 5400)
 
 array_size = problem_size # array
 X = get_arrays(:x, AType, bm.float_type, array_size)

diff --git a/test/execution/bm_fused_shared_reads.jl b/test/execution/bm_fused_shared_reads.jl
@@ -39,7 +39,7 @@ use_cuda = @isdefined(CUDA) && CUDA.has_cuda() # will be true if you first run `
 AType = use_cuda ? CUDA.CuArray : Array
 device_name = use_cuda ? CUDA.name(CUDA.device()) : "CPU"
 bm = Benchmark(; device_name, float_type = Float32)
-problem_size = (50, 5, 5, 6, 50)
+problem_size = (50, 5, 5, 6, 5400)
 
 array_size = problem_size # array
 X = get_arrays(:x, AType, bm.float_type, array_size)

diff --git a/test/execution/bm_fused_shared_reads_writes.jl b/test/execution/bm_fused_shared_reads_writes.jl
@@ -27,6 +27,7 @@ end
 function perf_kernel_shared_reads_writes_fused!(X, Y)
     (; x1, x2, x3, x4, x5, x6, x7, x8, x9, x10) = X
     (; y1, y2, y3, y4, y5, y6, y7, y8, y9, y10) = Y
+    # Totoal: 10 writes, 10 reads, and 5 read/write overlaps
     MBF.@fused_direct begin
         @. y1 = x1 + x6
         @. y2 = x2 + x7
@@ -45,12 +46,8 @@ end
 use_cuda = @isdefined(CUDA) && CUDA.has_cuda() # will be true if you first run `using CUDA`
 AType = use_cuda ? CUDA.CuArray : Array
 device_name = use_cuda ? CUDA.name(CUDA.device()) : "CPU"
-bm = Benchmark(;
-    problem_size = (prod((50, 5, 5, 6, 50)),),
-    device_name,
-    float_type = Float32,
-)
-problem_size = (50, 5, 5, 6, 50)
+bm = Benchmark(; device_name, float_type = Float32)
+problem_size = (50, 5, 5, 6, 5400)
 
 array_size = problem_size # array
 X = get_arrays(:x, AType, bm.float_type, array_size)
@@ -68,7 +65,7 @@ push_benchmark!(
     perf_kernel_shared_reads_writes_unfused!,
     X,
     Y;
-    n_reads_writes = 10 + 15,
+    n_reads_writes = 10 + 10,
     problem_size = array_size,
 )
 push_benchmark!(
@@ -77,7 +74,7 @@ push_benchmark!(
     perf_kernel_shared_reads_writes_fused!,
     X,
     Y;
-    n_reads_writes = 10 + 15,
+    n_reads_writes = 10 + 10,
     problem_size = array_size,
 )
 
@@ -97,7 +94,7 @@ push_benchmark!(
     perf_kernel_shared_reads_writes_unfused!,
     X,
     Y;
-    n_reads_writes = 10 + 15,
+    n_reads_writes = 10 + 10,
     problem_size = array_size,
 )
 push_benchmark!(
@@ -106,7 +103,7 @@ push_benchmark!(
     perf_kernel_shared_reads_writes_fused!,
     X,
     Y;
-    n_reads_writes = 10 + 15,
+    n_reads_writes = 10 + 10,
     problem_size = array_size,
 )
 

diff --git a/test/execution/utils_test.jl b/test/execution/utils_test.jl
@@ -90,34 +90,13 @@ function test_kernel!(; fused!, unfused!, X, Y)
         y .= map(_ -> rand(), y)
     end
     X_fused = deepcopy(X)
-    X_unfused = deepcopy(X)
+    X_unfused = X
     Y_fused = deepcopy(Y)
-    Y_unfused = deepcopy(Y)
+    Y_unfused = Y
     fused!(X_fused, Y_fused)
     unfused!(X_unfused, Y_unfused)
     @testset "Test correctness of $(nameof(typeof(fused!)))" begin
         test_compare(X_fused, X_unfused)
         test_compare(Y_fused, Y_unfused)
     end
 end
-function test_kernel_args!(; fused!, unfused!, args)
-    (; X, Y) = args
-    for x in X
-        x .= rand(size(x)...)
-    end
-    for y in Y
-        y .= rand(size(y)...)
-    end
-    X_fused = deepcopy(X)
-    X_unfused = deepcopy(X)
-    Y_fused = deepcopy(Y)
-    Y_unfused = deepcopy(Y)
-    args_fused = (; X = X_fused, Y = Y_fused)
-    args_unfused = (; X = X_unfused, Y = Y_unfused)
-    fused!(args_fused)
-    unfused!(args_unfused)
-    @testset "Test correctness of $(nameof(typeof(fused!)))" begin
-        test_compare(X_fused, X_unfused)
-        test_compare(Y_fused, Y_unfused)
-    end
-end