sw: Add more complex Snitch application tests

.gitlab-ci.yml

@@ -87,15 +87,16 @@ sim-vsim:
     - vsim-compile
   script:
     # Run the simulation
-    - make vsim-run-batch
+    - make vsim-run-batch-verify
     # Check either success or failure for non-zero exit codes
     - 'if [ -z "${NZ_EXIT_CODE}" ]; then grep "] SUCCESS" transcript || (exit 1); else grep "] FAILED: return code ${NZ_EXIT_CODE}" transcript || (exit 1); fi'
     # Check for UART output
     - 'if [ ! -z "${USTR}" ]; then (grep " \[UART\] ${USTR}" transcript); fi'
     # Check for any fatal errors
     - 'if grep "Fatal:" transcript; then exit 1; fi'
-    # Check for any errors (except one for non-zero exit codes)
-    - 'if [ ! -z "${NZ_EXIT_CODE}" ]; then count=$(grep -c "Error:" transcript); if [ "$count" -ne 1 ]; then exit 1; fi; else if grep -q "Error:" transcript; then exit 1; fi; fi'
+    # Check for any non-fatal errors. One and only one error is expected with a non-zero exit code.
+    # Ignore all errors when using a separate verification script.
+    - 'if [ -z "${VERIFY_PY}" ]; then if [ ! -z "${NZ_EXIT_CODE}" ]; then count=$(grep -c "Error:" transcript); if [ "$count" -ne 1 ]; then exit 1; fi; else if grep -q "Error:" transcript; then exit 1; fi; fi; fi'
   artifacts:
     paths:
       - transcript

.gitlab/common.yml

@@ -4,7 +4,7 @@
 
 variables:
   PIP_CACHE_DIR: "$CI_PROJECT_DIR/.cache/pip"
-  PD_COMMIT: "c22baf1fcce6b0cf76b85d2009e146ba3bbb6807"
+  PD_COMMIT: "245dd3d4a7c33f6052a913691382871bd2c53fc0"
   SPU_COMMIT: "c2e8815487bd713624d74ef3e3e0465196b6d67f"
 
 # Check the cache for bender and python dependencies
@@ -66,11 +66,13 @@ variables:
     - .cache-deps
     - .init-env
   script:
-    - make sn-tests
+    - make sn-tests DEBUG=ON
+    - make sn-apps DEBUG=ON
     - make pb-sn-tests
   artifacts:
     paths:
       - sw/snitch/tests/build/*.elf
+      - sw/snitch/apps/**/build/*.elf
     expire_in: 1 day
 
 # Compile the cheshire software tests

.gitlab/sw-tests.yml

@@ -7,6 +7,7 @@
   variables:
     CHS_BUILD_DIR: sw/cheshire/tests
     SN_BUILD_DIR: sw/snitch/tests/build
+    SN_ROOT: .deps/snitch_cluster
   parallel:
     matrix:
       - { CHS_BINARY: $CHS_BUILD_DIR/sanity.spm.elf, PRELMODE: 0 }
@@ -29,3 +30,8 @@
       - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: $SN_BUILD_DIR/redmule.elf }
       - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: $SN_BUILD_DIR/redmule_quant.elf }
       - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: $SN_BUILD_DIR/datamover.elf }
+      - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: sw/snitch/apps/gemm_2d/build/gemm_2d.elf, VERIFY_PY: $SN_ROOT/sw/blas/gemm/scripts/verify.py, PRELMODE: 3 }
+      - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: sw/snitch/apps/fused_concat_linear/build/fused_concat_linear.elf, VERIFY_PY: $SN_ROOT/sw/dnn/fused_concat_linear/scripts/verify.py, PRELMODE: 3 }
+      - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: sw/snitch/apps/mha/build/mha.elf, VERIFY_PY: $SN_ROOT/sw/dnn/mha/scripts/verify.py, PRELMODE: 3 }
+      - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: sw/snitch/apps/gemm/build/gemm.elf, VERIFY_PY: $SN_ROOT/sw/blas/gemm/scripts/verify.py, PRELMODE: 3 }
+      - { CHS_BINARY: $CHS_BUILD_DIR/simple_offload.spm.elf, SN_BINARY: sw/snitch/apps/axpy/build/axpy.elf, VERIFY_PY: $SN_ROOT/sw/blas/axpy/scripts/verify.py, PRELMODE: 3 }

Bender.lock

@@ -146,7 +146,7 @@ packages:
     dependencies:
     - common_cells
   common_cells:
-    revision: bef3d3c5ed0e2cc211e69a6dbd81c4fe3a97025c
+    revision: e1c09c75775c5f03eb45906d5145dbd2f5bcfb95
     version: null
     source:
       Git: https://github.com/pulp-platform/common_cells.git
@@ -383,7 +383,7 @@ packages:
     - common_cells
     - register_interface
   snitch_cluster:
-    revision: e4eaa0fb64767bb8f6b7d1f5fa705928171092b2
+    revision: ef3ece6c9e119fbfc25b26bb89a429ccdaacb5c6
     version: null
     source:
       Git: https://github.com/pulp-platform/snitch_cluster.git

Bender.yml

@@ -12,7 +12,7 @@ dependencies:
   axi:                { git: "https://github.com/pulp-platform/axi.git",                version: "0.39.6"                               }
   common_cells:       { git: "https://github.com/pulp-platform/common_cells.git",       rev: "snitch"                                   }
   cheshire:           { git: "https://github.com/pulp-platform/cheshire.git",           rev: "picobello"                                }
-  snitch_cluster:     { git: "https://github.com/pulp-platform/snitch_cluster.git",     rev: "e4eaa0fb64767bb8f6b7d1f5fa705928171092b2" }
+  snitch_cluster:     { git: "https://github.com/pulp-platform/snitch_cluster.git",     rev: "ef3ece6c9e119fbfc25b26bb89a429ccdaacb5c6" }
   floo_noc:           { git: "https://github.com/pulp-platform/FlooNoC.git",            rev: "develop"                                  }
   obi:                { git: "https://github.com/pulp-platform/obi.git",                rev: "acfcd0f80c7539aa8da7821a66d9acf2074a5b4e" }
   redmule:            { git: "https://github.com/pulp-platform/redmule.git",            rev: "picobello"                                }

Makefile

@@ -218,7 +218,7 @@ python-venv: .venv
 	python -m pip install --upgrade pip setuptools && \
 	python -m pip install --cache-dir $(PIP_CACHE_DIR) -r requirements.txt && \
 	python -m pip install --cache-dir $(PIP_CACHE_DIR) $(shell $(BENDER) path floo_noc) --no-deps && \
-	python -m pip install --cache-dir $(PIP_CACHE_DIR) $(shell $(BENDER) path snitch_cluster)
+	python -m pip install --cache-dir $(PIP_CACHE_DIR) "$(shell $(BENDER) path snitch_cluster)[kernels]"
 
 python-venv-clean:
 	rm -rf .venv

README.md

@@ -112,6 +112,14 @@ Use the `vsim-run-batch` command to run tests in batch mode with RTL optimizatio
 
 Use the `PRELMODE=3` flag to enable fast preload of the Snitch binary, and speed up the simulation.
 
+Some applications produce a lot of output data, which would be time-consuming to check in simulation.
+Said applications usually come with a Python verification script that can check the results from a dump of the memory contents at the end of the simulation.
+For example, a verification script for the GEMM kernel can be found under `$(bender path snitch_cluster)/sw/blas/gemm/scripts/verify.py`
+To run an application on Snitch and verify its results, do:
+```bash
+make vsim-run-batch-verify VERIFY_PY=$(bender path snitch_cluster)/sw/blas/gemm/scripts/verify.py PRELMODE=3 CHS_BINARY=sw/cheshire/tests/simple_offload.spm.elf SN_BINARY=sw/snitch/apps/blas/gemm/build/gemm.elf
+```
+
 ### Additional help
 
 Additionally, you can run the following command to get a list of all available commands:

hw/mem_tile.sv

@@ -16,7 +16,8 @@ module mem_tile
 #(
   parameter bit          AxiUserAtop    = 1'b1,
   parameter int unsigned AxiUserAtopMsb = 3,
-  parameter int unsigned AxiUserAtopLsb = 0
+  parameter int unsigned AxiUserAtopLsb = 0,
+  parameter int unsigned MemTileId      = 0
 ) (
   input  logic                    clk_i,
   input  logic                    rst_ni,
@@ -275,6 +276,25 @@ module mem_tile
     end
   end
 
+`ifndef SYNTHESIS
+  // AXI Monitor dumper to improvce debiugging
+  axi_dumper #(
+    .BusName   ($sformatf("mem_tile_%d", MemTileId)),
+    .LogAW     (1'b1),
+    .LogAR     (1'b1),
+    .LogW      (1'b1),
+    .LogB      (1'b1),
+    .LogR      (1'b1),
+    .axi_req_t (axi_nw_join_req_t),
+    .axi_resp_t(axi_nw_join_rsp_t)
+  ) i_axi_monitor (
+    .clk_i,
+    .rst_ni,
+    .axi_req_i (axi_req),
+    .axi_resp_i(axi_rsp)
+  );
+`endif
+
   axi_to_obi #(
     .ObiCfg      (MgrObiCfg),
     .obi_req_t   (mgr_obi_req_t),

hw/picobello_top.sv

@@ -273,7 +273,9 @@ module picobello_top
     localparam int MemTileX = int'(MemTilePhysicalId.x);
     localparam int MemTileY = int'(MemTilePhysicalId.y);
 
-    mem_tile i_mem_tile (
+    mem_tile #(
+      .MemTileId(int'(m))
+    ) i_mem_tile (
       .clk_i,
       .rst_ni,
       .test_enable_i   (test_mode_i),

requirements.txt

@@ -25,5 +25,5 @@ hjson  # for reggen
 
 # For peakrdl
 peakrdl
-peakrdl-rawheader @ git+https://github.com/micprog/peakrdl-rawheader.git
+peakrdl-rawheader @ git+https://github.com/colluca/PeakRDL-rawheader.git@7b8dbc9ad5854dc1cdaf36d4ea024c29ffb00a4c
 peakrdl-markdown

sw/snitch/apps/axpy/app.mk

@@ -0,0 +1,15 @@
+# Copyright 2025 ETH Zurich and University of Bologna.
+# Licensed under the Apache License, Version 2.0, see LICENSE for details.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Luca Colagrande <[email protected]>
+
+APP              := axpy
+$(APP)_BUILD_DIR ?= $(PB_SNITCH_SW_DIR)/apps/$(APP)/build
+$(APP)_DATA_CFG  := $(PB_SNITCH_SW_DIR)/apps/$(APP)/data/params.json
+SRC_DIR          := $(SN_ROOT)/sw/blas/$(APP)/src
+SRCS             := $(SRC_DIR)/main.c
+$(APP)_INCDIRS   := $(SN_ROOT)/sw/blas
+
+include $(SN_ROOT)/sw/apps/common.mk
+include $(SN_ROOT)/target/snitch_cluster/sw/apps/common.mk

sw/snitch/apps/axpy/data/params.json

@@ -0,0 +1,9 @@
+// Copyright 2023 ETH Zurich and University of Bologna.
+// Licensed under the Apache License, Version 2.0, see LICENSE for details.
+// SPDX-License-Identifier: Apache-2.0
+
+{
+    "n_tiles": 5,
+    "n": 2560,
+    "funcptr": "axpy_opt"
+}

sw/snitch/apps/fused_concat_linear/app.mk

@@ -0,0 +1,15 @@
+# Copyright 2025 ETH Zurich and University of Bologna.
+# Licensed under the Apache License, Version 2.0, see LICENSE for details.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Luca Colagrande <[email protected]>
+
+APP              := fused_concat_linear
+$(APP)_BUILD_DIR ?= $(PB_SNITCH_SW_DIR)/apps/$(APP)/build
+$(APP)_DATA_CFG  := $(PB_SNITCH_SW_DIR)/apps/$(APP)/data/params.json
+SRC_DIR          := $(SN_ROOT)/sw/dnn/$(APP)/src
+SRCS             := $(SRC_DIR)/main.c
+$(APP)_INCDIRS   := $(SN_ROOT)/sw/dnn/src $(SN_ROOT)/sw/blas
+
+include $(SN_ROOT)/sw/apps/common.mk
+include $(SN_ROOT)/target/snitch_cluster/sw/apps/common.mk

sw/snitch/apps/fused_concat_linear/data/params.json

@@ -0,0 +1,11 @@
+// Copyright 2023 ETH Zurich and University of Bologna.
+// Solderpad Hardware License, Version 0.51, see LICENSE for details.
+// SPDX-License-Identifier: SHL-0.51
+
+{
+    num_inputs: 2,
+    input_shape: [16, 16],
+    output_shape: [16, 16],
+    dtype: "FP64",
+    gemm_implementation: "gemm_fp64_opt"
+}

sw/snitch/apps/gemm/app.mk

@@ -0,0 +1,15 @@
+# Copyright 2025 ETH Zurich and University of Bologna.
+# Licensed under the Apache License, Version 2.0, see LICENSE for details.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Luca Colagrande <[email protected]>
+
+APP              := gemm
+$(APP)_BUILD_DIR ?= $(PB_SNITCH_SW_DIR)/apps/$(APP)/build
+$(APP)_DATA_CFG  := $(PB_SNITCH_SW_DIR)/apps/$(APP)/data/params.json
+SRC_DIR          := $(SN_ROOT)/sw/blas/$(APP)/src
+SRCS             := $(SRC_DIR)/main.c
+$(APP)_INCDIRS   := $(SN_ROOT)/sw/blas
+
+include $(SN_ROOT)/sw/apps/common.mk
+include $(SN_ROOT)/target/snitch_cluster/sw/apps/common.mk

sw/snitch/apps/gemm/data/params.json

@@ -0,0 +1,25 @@
+// Copyright 2024 ETH Zurich and University of Bologna.
+// Licensed under the Apache License, Version 2.0, see LICENSE for details.
+// SPDX-License-Identifier: Apache-2.0
+
+{
+    setup_ssr: 1,
+    parallelize_m: 1,
+    parallelize_k: 0,
+    m_tiles: 64, // number of tiles in M dimension
+    n_tiles: 1, // number of tiles in N dimension
+    k_tiles: 1, // number of tiles in K dimension
+    load_a: 1,
+    load_b: 1,
+    load_c: 1,
+    double_buffer: 1,
+    partition_banks: 0,
+    transa: false,
+    transb: false, // must be true for SIMD
+    m: 2048,
+    n: 16,
+    k: 16,
+    alpha: 1,
+    beta: 0,
+    gemm_fp: "gemm_fp64_opt"
+}

sw/snitch/apps/gemm_2d/app.mk

@@ -0,0 +1,17 @@
+# Copyright 2025 ETH Zurich and University of Bologna.
+# Licensed under the Apache License, Version 2.0, see LICENSE for details.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Lorenzo Leone <[email protected]>
+
+APP              := gemm_2d
+$(APP)_BUILD_DIR ?= $(PB_SNITCH_SW_DIR)/apps/$(APP)/build
+SRC_DIR          := $(PB_SNITCH_SW_DIR)/apps/$(APP)/src
+SRCS             := $(SRC_DIR)/gemm_2d.c
+$(APP)_INCDIRS   := $(SN_ROOT)/sw/blas $(SN_ROOT)/sw/blas/gemm/src
+
+# Refer to Snitch scripts
+$(APP)_SCRIPT_DIR :=  $(SN_ROOT)/sw/blas/gemm/scripts
+
+include $(SN_ROOT)/sw/apps/common.mk
+include $(SN_ROOT)/target/snitch_cluster/sw/apps/common.mk

sw/snitch/apps/gemm_2d/data/params.json

@@ -0,0 +1,25 @@
+// Copyright 2024 ETH Zurich and University of Bologna.
+// Licensed under the Apache License, Version 2.0, see LICENSE for details.
+// SPDX-License-Identifier: Apache-2.0
+
+{
+    setup_ssr: 1,
+    parallelize_m: 1,
+    parallelize_k: 0,
+    m_tiles: 16, // number of tiles in M dimension
+    n_tiles: 4, // number of tiles in N dimension
+    k_tiles: 1, // number of tiles in K dimension
+    load_a: 1,
+    load_b: 1,
+    load_c: 1,
+    double_buffer: 1,
+    partition_banks: 0,
+    transa: false,
+    transb: false, // must be true for SIMD
+    m: 128,
+    n: 32,
+    k: 16,
+    alpha: 1,
+    beta: 0,
+    gemm_fp: "gemm_fp64_opt"
+}

sw/snitch/apps/gemm_2d/roi.json

@@ -0,0 +1,30 @@
+[
+    <% N_TILES = 4 %>
+
+    % for cluster in range(0,16):
+        // Compute cores
+        % for j in range(0, 8):
+        {
+            "thread": "${f'hart_{cluster * 9 + j + 1}'}",
+            "roi": [
+            % for i in range(0, N_TILES):
+                {"idx": ${2 * i + 1}, "label": "${f'tile_{i}'}"},
+            % endfor
+            ]
+        },
+        % endfor
+
+        // DMA core
+        {
+            "thread": "${f'hart_{cluster * 9 + 8 + 1}'}",
+            "roi": [
+                {"idx": 1, "label": "${f'tile_in_0'}"},
+                % for i in range(0, N_TILES - 1):
+                    {"idx": ${4*i + 3}, "label": "${f'tile_in_{i+1}'}"},
+                    {"idx": ${4*i + 5}, "label": "${f'tile_out_{i}'}"},
+                % endfor
+                {"idx": ${N_TILES * 4 - 1}, "label": "${f'tile_out_{N_TILES-1}'}"},
+            ]
+        },
+    % endfor
+]