snax_KUL_cluster and snax_gemmX_xdma_cluster (#10)

* Modify hardware cfg * chore: update cluster configuration file with cluster base offset * Fix integration problems * Update new probe file
KULeuven-MICAS · Aug 17, 2024 · 53707c4 · 53707c4
1 parent b651eca
commit 53707c4
Show file tree

Hide file tree

Showing 6 changed files with 55 additions and 3,322 deletions.
diff --git a/target/fpga/probes.ltx b/target/fpga/probes.ltx
diff --git a/target/rtl/Makefile b/target/rtl/Makefile
@@ -282,7 +282,7 @@ $(TARGET_CLINT_DIR)/clint.%: $(SOURCE_CLINT_DIR)/data/clint.%.tpl $(CFG) | $(TAR
 # SNAX Accelerator + Wrapper Generations #
 ###########################################
 # TODO: This SNAX_CFGS should not be fixed! It should be created dynamically instead. 
-SNAX_CFGS += cfg/cluster_cfg/snax_streamer_gemm_add_c_cluster.hjson
+SNAX_CFGS += cfg/cluster_cfg/snax_KUL_cluster.hjson
 SNAX_CFGS += cfg/cluster_cfg/snax_streamer_gemmX_xdma_cluster.hjson
 
 #######################
@@ -314,7 +314,7 @@ ifeq ($(findstring snax_xdma_cluster,$(SNAX_CFGS)),snax_xdma_cluster)
  $(eval BENDER_TARGETS += -t snax_xdma -t snax_xdma_cluster)
 endif
 
-ifeq ($(findstring snax_kul_cluster_all_narrow,$(SNAX_CFGS)),snax_kul_cluster_all_narrow)
+ifeq ($(findstring snax_KUL_cluster,$(SNAX_CFGS)),snax_KUL_cluster)
  $(eval BENDER_TARGETS += -t snax_KUL_cluster)
 endif
 

diff --git a/...ter_cfg/snax_kul_cluster_all_narrow.hjson → ...tl/cfg/cluster_cfg/snax_KUL_cluster.hjson b/...ter_cfg/snax_kul_cluster_all_narrow.hjson → ...tl/cfg/cluster_cfg/snax_KUL_cluster.hjson
@@ -10,11 +10,11 @@
  },
 
  cluster: {
- name: "snax_kul_cluster_all_narrow",
+ name: "snax_KUL_cluster",
  boot_addr: 4096, // 0x1000
  cluster_base_addr: 268435456, // 0x1000_0000
- cluster_base_offset: 0, // 0x0
- cluster_base_hartid: 0,
+ cluster_base_offset: 262144, // 256KB
+ cluster_base_hartid: 1,
  addr_width: 48,
  data_width: 64,
  tcdm: {
@@ -24,11 +24,28 @@
  cluster_periph_size: 64, // kB
  zero_mem_size: 64, // kB
  dma_data_width: 512,
- dma_axi_req_fifo_depth: 3,
- dma_req_fifo_depth: 3,
+ dma_axi_req_fifo_depth: 16,
+ dma_req_fifo_depth: 8,
  // AXI bandwidth switcher
- use_ax_bw_converter: true,
+ use_ax_bw_converter: false,
  converted_axi_bandwidth: 256,
+
+ // Additional parameters for Occamy Integration
+ narrow_trans: 4,
+ wide_trans: 32,
+ dma_user_width: 1,
+ // We don't need Snitch debugging in Occamy
+ enable_debug: false,
+ // We don't need Snitch (core-internal) virtual memory support
+ vm_support: false,
+ // Memory configuration inputs
+ sram_cfg_expose: true,
+ sram_cfg_fields: {
+ ema: 3,
+ emaw: 2,
+ emas: 1
+ },
+
  // Timing parameters
  timing: {
  lat_comp_fp32: 3,

diff --git a/target/rtl/cfg/occamy_cfg/snax_two_clusters.hjson b/target/rtl/cfg/occamy_cfg/snax_two_clusters.hjson
@@ -115,7 +115,7 @@
  cfg_base_offset: 65536 // 0x10000
  },
  clusters:[
- "snax_streamer_gemm_add_c_cluster",
+ "snax_KUL_cluster",
  "snax_streamer_gemmX_xdma_cluster",
  ],
 

diff --git a/target/rtl/test/bootrom.elf b/target/rtl/test/bootrom.elf
diff --git a/target/rtl/test/bootrom.ld b/target/rtl/test/bootrom.ld
@@ -18,6 +18,7 @@ SECTIONS
  .end ALIGN(16): { *(end) } > bootrom
 
  /DISCARD/ : { *(.riscv.attributes) *(.comment) }
+
  /* Global and stack pointer */
  __global_pointer$ = ADDR(.misc) + SIZEOF(.misc) / 2;
  __stack_pointer$ = ORIGIN(spm) + LENGTH(spm) - 8;