Added settings for benchmarking batched routines

2 files changed, 63 insertions, 1 deletions

diff --git a/scripts/benchmark/benchmark.py b/scripts/benchmark/benchmark.py
index 46f2c503..c4054669 100644
--- a/scripts/benchmark/benchmark.py
+++ b/scripts/benchmark/benchmark.py
@@ -17,9 +17,11 @@ import utils
 
 EXPERIMENTS = {
     "axpy": settings.AXPY,
+    "axpybatched": settings.AXPYBATCHED,
     "gemv": settings.GEMV,
     "gemm": settings.GEMM,
     "gemm_small": settings.GEMM_SMALL,
+    "gemmbatched": settings.GEMMBATCHED,
     "symm": settings.SYMM,
     "syrk": settings.SYRK,
     "summary": settings.SUMMARY,
@@ -62,7 +64,7 @@ def run_benchmark(name, arguments_list, precision, num_runs, platform, device):
 
 def parse_arguments(argv):
     parser = argparse.ArgumentParser(description="Runs a full benchmark for a specific routine on a specific device")
-    parser.add_argument("-b", "--benchmark", required=True, help="The benchmark to perform (choose from %s)" % EXPERIMENTS.keys())
+    parser.add_argument("-b", "--benchmark", required=True, help="The benchmark to perform (choose from %s)" % sorted(EXPERIMENTS.keys()))
     parser.add_argument("-p", "--platform", required=True, type=int, help="The ID of the OpenCL platform to test on")
     parser.add_argument("-d", "--device", required=True, type=int, help="The ID of the OpenCL device to test on")
     parser.add_argument("-n", "--num_runs", type=int, default=None, help="Overrides the default number of benchmark repeats for averaging")
diff --git a/scripts/benchmark/settings.py b/scripts/benchmark/settings.py
index 77f5fc21..13b7c359 100644
--- a/scripts/benchmark/settings.py
+++ b/scripts/benchmark/settings.py
@@ -60,6 +60,35 @@ AXPY = {
     ]
 }
 
+AXPYBATCHED = {
+    "label_names": ["CLBlast", "clBLAS (non batched)"],
+    "num_rows": 1, "num_cols": 3,
+    "benchmarks": [
+        {
+            "name": "axpybatched", "num_runs": 40,
+            "title": "10 batches",
+            "x_label": "sizes (n)", "x_keys": ["n"],
+            "y_label": "GB/s (higher is better)", "y_keys": ["GBs_1", "GBs_2"],
+            "arguments": [{"batch_num": 10, "n": utils.k(32), "incx": 1, "incy": 1, "step": utils.k(16), "num_steps": 16}],
+        },
+        {
+            "name": "axpybatched", "num_runs": 20,
+            "title": "50 batches",
+            "x_label": "sizes (m=n=k)", "x_keys": ["n"],
+            "y_label": "GB/s (higher is better)", "y_keys": ["GBs_1", "GBs_2"],
+            "arguments": [{"batch_num": 50, "n": utils.k(32), "incx": 1, "incy": 1, "step": utils.k(16), "num_steps": 16}],
+        },
+        {
+            "name": "axpybatched", "num_runs": 40,
+            "title": "n=128K",
+            "x_label": "number of batches", "x_keys": ["batch_num"],
+            "y_label": "GB/s (higher is better)", "y_keys": ["GBs_1", "GBs_2"],
+            "arguments": [{"batch_num": b, "n": utils.k(128), "incx": 1, "incy": 1, "step": 1, "num_steps": 1}
+                          for b in utils.powers_of_2(1, 128)],
+        }
+    ]
+}
+
 GEMV = {
     "label_names": ["CLBlast", "clBLAS"],
     "num_rows": 2, "num_cols": 3,
@@ -191,6 +220,37 @@ GEMM_SMALL = {
     ]
 }
 
+GEMMBATCHED = {
+    "label_names": ["CLBlast", "clBLAS (non batched)"],
+    "num_rows": 1, "num_cols": 3,
+    "benchmarks": [
+        {
+            "name": "gemmbatched", "num_runs": 40,
+            "title": "10 batches",
+            "x_label": "sizes (m=n=k)", "x_keys": ["m"],
+            "y_label": "GFLOPS (higher is better)", "y_keys": ["GFLOPS_1", "GFLOPS_2"],
+            "arguments": [{"batch_num": 10, "m": 16, "n": 16, "k": 16, "layout": 102,
+                           "transA": 111, "transB": 111, "step": 16, "num_steps": 16}],
+        },
+        {
+            "name": "gemmbatched", "num_runs": 20,
+            "title": "50 batches",
+            "x_label": "sizes (m=n=k)", "x_keys": ["m"],
+            "y_label": "GFLOPS (higher is better)", "y_keys": ["GFLOPS_1", "GFLOPS_2"],
+            "arguments": [{"batch_num": 50, "m": 16, "n": 16, "k": 16, "layout": 102,
+                           "transA": 111, "transB": 111, "step": 16, "num_steps": 16}],
+        },
+        {
+            "name": "gemmbatched", "num_runs": 40,
+            "title": "m=n=k=32",
+            "x_label": "number of batches", "x_keys": ["batch_num"],
+            "y_label": "GFLOPS (higher is better)", "y_keys": ["GFLOPS_1", "GFLOPS_2"],
+            "arguments": [{"batch_num": b, "m": 32, "n": 32, "k": 32, "layout": 102,
+                           "transA": 111, "transB": 111} for b in utils.powers_of_2(1, 128)],
+        }
+    ]
+}
+
 SYMM = {
     "label_names": ["CLBlast", "clBLAS"],
     "num_rows": 2, "num_cols": 3,