#5480: Add test_prefetcher.cpp tests back to CI and update test numbers

- Also, improve some BW/iterations logging in test_dispatcher.cpp now that tests will use device iterations instead of host iterations
tenstorrent · Apr 2, 2024 · 4bed077 · 4bed077
1 parent b14d41a
commit 4bed077
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 11 deletions.
diff --git a/tests/scripts/run_cpp_fd2_tests.sh b/tests/scripts/run_cpp_fd2_tests.sh
@@ -20,11 +20,12 @@ fi
 #############################################
 echo "Running test_prefetcher tests now...";
 
-# FIXME Hangs sometimes TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 0 -i 3  # Smoke Test
-# FIXME comment out just in case TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 1 -i 3  # Random Test
-# FIXME occasional hang/mismatch TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 2 -i 3  # PCIE Test
-# FIXME comment out just in case  TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 3 -i 3  # Paged DRAM Read Test
-# FIXME Failing TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 4 -i 3  # Paged DRAM Write + Read Test
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 0 -i 3  # TrueSmoke Test
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 1 -i 3  # Smoke Test
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 2 -i 3  # Random Test
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 3 -i 3  # PCIE Test
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 4 -i 3  # Paged DRAM Read Test
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 5 -i 3  # Paged DRAM Write + Read Test
 
 # Testcase: Paged Write Cmd to DRAM. 256 pages, 224b size.
 TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_prefetcher -t 4 -i 1 -dpgs 224 -dpgr 256

diff --git a/tests/tt_metal/tt_metal/perf_microbenchmark/dispatch/test_dispatcher.cpp b/tests/tt_metal/tt_metal/perf_microbenchmark/dispatch/test_dispatcher.cpp
@@ -514,8 +514,9 @@ int main(int argc, char **argv) {
         }
 
         std::chrono::duration<double> elapsed_seconds = (end-start);
-        log_info(LogTest, "Ran in {}us", elapsed_seconds.count() * 1000 * 1000);
-        log_info(LogTest, "Ran in {}us per iteration", elapsed_seconds.count() * 1000 * 1000 / iterations_g);
+        uint32_t total_iterations = iterations_g * prefetcher_iterations_g;
+        log_info(LogTest, "Ran in {}us (for total iterations: {})", elapsed_seconds.count() * 1000 * 1000, total_iterations);
+        log_info(LogTest, "Ran in {}us per iteration", elapsed_seconds.count() * 1000 * 1000 /total_iterations);
         if (iterations_g > 0) {
             float total_words = 0;
             if (min_xfer_size_bytes_g != max_xfer_size_bytes_g) {
@@ -543,12 +544,12 @@ int main(int argc, char **argv) {
                 break;
             }
 
-            total_words *= iterations_g;
-            float bw = total_words * sizeof(uint32_t) * prefetcher_iterations_g / (elapsed_seconds.count() * 1024.0 * 1024.0 * 1024.0);
+            total_words *= total_iterations;
+            float bw = total_words * sizeof(uint32_t) / (elapsed_seconds.count() * 1024.0 * 1024.0 * 1024.0);
             std::stringstream ss;
             ss << std::fixed << std::setprecision(3) << bw;
-            log_info(LogTest, "BW: {} GB/s (from total_words: {} size: {} MB via host_iter: {} for num_cores: {})",
-                ss.str(), total_words, total_words * sizeof(uint32_t) / (1024.0 * 1024.0), iterations_g, all_workers_g.size());
+            log_info(LogTest, "BW: {} GB/s (from total_words: {} size: {} MB via host_iter: {} prefetcher_iter: {} for num_cores: {})",
+                ss.str(), total_words, total_words * sizeof(uint32_t) / (1024.0 * 1024.0), iterations_g, prefetcher_iterations_g, all_workers_g.size());
         } else {
             log_info(LogTest, "BW: -- GB/s (use -i 1 to report bandwidth)");
         }