From 062e9bb49d3235eb776f72dcadc6973baba610b7 Mon Sep 17 00:00:00 2001
From: Hans Pabst <hans.pabst@intel.com>
Date: Mon, 27 Jan 2025 08:49:47 +0100
Subject: [PATCH] DBM: MAX_BATCH_SIZE 10000 -> 30000 (#3890)

- MAX_BATCH_SIZE tuned on CPUs assuming GPUs benefit even more.
- Driver/Makefile/HIP: missing object file/compilation.
---
 src/dbm/Makefile          | 6 +++---
 src/dbm/dbm_hyperparams.h | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/dbm/Makefile b/src/dbm/Makefile
index f3bb458412..81e7932776 100644
--- a/src/dbm/Makefile
+++ b/src/dbm/Makefile
@@ -91,10 +91,10 @@ endif
 all: dbm_miniapp.x
 
 clean:
-	rm -fv *.o ../offload/*.o $(OPENCL_GENKRNL)
+	rm -fv $(ALL_OBJECTS) $(OPENCL_GENKRNL)
 
 realclean: clean
-	rm -fv *.x
+	rm -fv dbm_miniapp.x
 
 
 # Enable OpenCL when DBCSR library was prebuilt (assume __DBCSR_ACC).
@@ -169,7 +169,7 @@ endif
 
 # Enable HIP/ROCm when hipcc compiler is present.
 else ifneq ($(HIPCC),)
-ALL_OBJECTS += dbm_multiply_gpu.o
+ALL_OBJECTS += dbm_multiply_gpu.o dbm_multiply_gpu_kernel.o
 CFLAGS += -I${ROCM_PATH}/include -D__OFFLOAD_HIP -D__HIP_PLATFORM_AMD__
 LIBS += -L${ROCM_PATH}/lib -lamdhip64 -lhipfft -lhipblas -lrocblas
 
diff --git a/src/dbm/dbm_hyperparams.h b/src/dbm/dbm_hyperparams.h
index b642013ed0..8e0e408dca 100644
--- a/src/dbm/dbm_hyperparams.h
+++ b/src/dbm/dbm_hyperparams.h
@@ -13,7 +13,7 @@
 #define HASHTABLE_FACTOR 3.0
 #define ALLOCATION_FACTOR 1.5
 #define SHARDS_PER_THREAD 1.0
-#define MAX_BATCH_SIZE 10000
+#define MAX_BATCH_SIZE 30000
 #define BATCH_NUM_BUCKETS 1000
 #define INITIAL_NBLOCKS_ALLOCATED 100
 #define INITIAL_DATA_ALLOCATED 1024