diff --git a/stats/s1/s1_runtime_loceye.txt b/stats/s1/s1_runtime_loceye.txt new file mode 100644 index 0000000..8312aa3 --- /dev/null +++ b/stats/s1/s1_runtime_loceye.txt @@ -0,0 +1,57 @@ +Device chosen is "GeForce GTX 1070" +Device has 15 multi processors and compute capability 6.1 +Max threads per block supported are 1024 + +Reading dataset and labels... +Done. + +Device memory allocation wall clock time = 0.083752 + +calculate_kernel_matrix_kernel called with: +dimBlock.x = 32, dimBlock.y = 32 +dimGrid.x = 157, dimGrid.y = 157 + +calculate_denominator called with: +dimBlock.x = 1024, dimBlock.y = 1 +dimGrid.x = 5, dimGrid.y = 1 + +shift_points_kernel called with: +dimBlock.x = 512, dimBlock.y = 2 +dimGrid.x = 10, dimGrid.y = 1 + +Recursion n. 0, error 1433009.094419 +Recursion n. 1, error 846076.669706 +Recursion n. 2, error 457323.896842 +Recursion n. 3, error 232981.679496 +Recursion n. 4, error 129695.421325 +Recursion n. 5, error 73386.379913 +Recursion n. 6, error 42859.404834 +Recursion n. 7, error 34613.230704 +Recursion n. 8, error 31166.226384 +Recursion n. 9, error 25075.599825 +Recursion n. 10, error 14788.867230 +Recursion n. 11, error 6526.169908 +Recursion n. 12, error 2538.871384 +Recursion n. 13, error 953.135636 +Recursion n. 14, error 354.381780 +Recursion n. 15, error 131.434483 +Recursion n. 16, error 48.740960 +Recursion n. 17, error 18.090348 +Recursion n. 18, error 6.723606 +Recursion n. 19, error 2.503479 +Recursion n. 20, error 0.934231 +Recursion n. 21, error 0.349569 +Recursion n. 22, error 0.131220 +Recursion n. 23, error 0.049442 +Recursion n. 24, error 0.018711 +Recursion n. 25, error 0.007116 +Recursion n. 26, error 0.002722 +Recursion n. 27, error 0.001047 +Recursion n. 28, error 0.000406 +Recursion n. 29, error 0.000158 +Recursion n. 30, error 0.000062 + +Copying between device and host wall clock time = 1.291885 + +Total number of recursions = 30 +Mean Shift wall clock time = 2.356798 diff --git a/stats/s1/s1_runtime_loceye_serial.txt b/stats/s1/s1_runtime_loceye_serial.txt new file mode 100644 index 0000000..219a9d5 --- /dev/null +++ b/stats/s1/s1_runtime_loceye_serial.txt @@ -0,0 +1,33 @@ +Iteration n. 0, error 1434684.624217 +Iteration n. 1, error 846466.140776 +Iteration n. 2, error 456756.809962 +Iteration n. 3, error 232102.841892 +Iteration n. 4, error 128970.916818 +Iteration n. 5, error 73131.136038 +Iteration n. 6, error 42959.583463 +Iteration n. 7, error 34479.873709 +Iteration n. 8, error 30952.428863 +Iteration n. 9, error 24978.110418 +Iteration n. 10, error 14758.750104 +Iteration n. 11, error 6515.798691 +Iteration n. 12, error 2534.949856 +Iteration n. 13, error 951.600403 +Iteration n. 14, error 353.770499 +Iteration n. 15, error 131.188064 +Iteration n. 16, error 48.640549 +Iteration n. 17, error 18.049030 +Iteration n. 18, error 6.706455 +Iteration n. 19, error 2.496305 +Iteration n. 20, error 0.931212 +Iteration n. 21, error 0.348291 +Iteration n. 22, error 0.130677 +Iteration n. 23, error 0.049212 +Iteration n. 24, error 0.018613 +Iteration n. 25, error 0.007074 +Iteration n. 26, error 0.002704 +Iteration n. 27, error 0.001040 +Iteration n. 28, error 0.000403 +Iteration n. 29, error 0.000157 +Iteration n. 30, error 0.000062 +Total iterations = 30 +Mean Shift wall clock time = 21.889184 diff --git a/stats/s1/s1_runtime_shared_loceye.txt b/stats/s1/s1_runtime_shared_loceye.txt new file mode 100644 index 0000000..771e9bb --- /dev/null +++ b/stats/s1/s1_runtime_shared_loceye.txt @@ -0,0 +1,57 @@ +Device chosen is "GeForce GTX 1070" +Device has 15 multi processors and compute capability 6.1 +Max threads per block supported are 1024 + +Reading dataset and labels... +Done. + +Device memory allocation wall clock time = 0.000946 + +calculate_kernel_matrix_kernel called with: +dimBlock.x = 32, dimBlock.y = 32 +dimGrid.x = 157, dimGrid.y = 157 + +calculate_denominator called with: +dimBlock.x = 1024, dimBlock.y = 1 +dimGrid.x = 5, dimGrid.y = 1 + +shift_points_kernel called with: +dimBlock.x = 240, dimBlock.y = 2 +dimGrid.x = 21, dimGrid.y = 1 + +Recursion n. 0, error 1433009.094419 +Recursion n. 1, error 846076.669706 +Recursion n. 2, error 457323.896842 +Recursion n. 3, error 232981.679496 +Recursion n. 4, error 129695.421325 +Recursion n. 5, error 73386.379913 +Recursion n. 6, error 42859.404834 +Recursion n. 7, error 34613.230704 +Recursion n. 8, error 31166.226384 +Recursion n. 9, error 25075.599825 +Recursion n. 10, error 14788.867230 +Recursion n. 11, error 6526.169908 +Recursion n. 12, error 2538.871384 +Recursion n. 13, error 953.135636 +Recursion n. 14, error 354.381780 +Recursion n. 15, error 131.434483 +Recursion n. 16, error 48.740960 +Recursion n. 17, error 18.090348 +Recursion n. 18, error 6.723606 +Recursion n. 19, error 2.503479 +Recursion n. 20, error 0.934231 +Recursion n. 21, error 0.349569 +Recursion n. 22, error 0.131220 +Recursion n. 23, error 0.049442 +Recursion n. 24, error 0.018711 +Recursion n. 25, error 0.007116 +Recursion n. 26, error 0.002722 +Recursion n. 27, error 0.001047 +Recursion n. 28, error 0.000406 +Recursion n. 29, error 0.000158 +Recursion n. 30, error 0.000062 + +Copying between device and host wall clock time = 1.294858 + +Total number of recursions = 30 +Mean Shift wall clock time = 2.270383