Merge branch 'rbarton/gemm' of https://github.com/pulp-platform/snitc…

…h_cluster into rbarton/gemm

.github/workflows/build-docker.yml

@@ -6,7 +6,7 @@
 name: build-docker
 on:
   push:
-    branches: [main]
+    branches: [tracer/dma]
   workflow_dispatch:
 jobs:
   build-docker:
@@ -27,6 +27,6 @@ jobs:
           context: .
           file: util/container/Dockerfile
           push: true
-          tags: ghcr.io/pulp-platform/snitch_cluster:${{ github.ref_name }}
+          tags: ghcr.io/pulp-platform/snitch_cluster:tracer-dma
           build-args: |-
             SNITCH_LLVM_VERSION=latest

.github/workflows/ci.yml

@@ -15,12 +15,26 @@ jobs:
     name: Build documentation
     runs-on: ubuntu-22.04
     container:
-      image: ghcr.io/pulp-platform/snitch_cluster:main
+      image: ghcr.io/pulp-platform/snitch_cluster:tracer-dma
     steps:
       - uses: actions/checkout@v2
       - name: Build docs
         run: make docs
 
+  #####################
+  # Python unit tests #
+  #####################
+
+  pytest:
+    name: Python unit tests
+    runs-on: ubuntu-22.04
+    container:
+      image: ghcr.io/pulp-platform/snitch_cluster:tracer-dma
+    steps:
+      - uses: actions/checkout@v2
+      - name: Run pytest
+        run: pytest
+
   ##############################################
   # Simulate SW on Snitch Cluster w/ Verilator #
   ##############################################
@@ -29,7 +43,7 @@ jobs:
     name: Simulate SW on Snitch Cluster w/ Verilator
     runs-on: ubuntu-22.04
     container:
-      image: ghcr.io/pulp-platform/snitch_cluster:main
+      image: ghcr.io/pulp-platform/snitch_cluster:tracer-dma
     steps:
       - uses: actions/checkout@v2
         with:
@@ -54,7 +68,7 @@ jobs:
     name: Simulate SW on Snitch Cluster w/ Banshee
     runs-on: ubuntu-22.04
     container:
-      image: ghcr.io/pulp-platform/snitch_cluster:main
+      image: ghcr.io/pulp-platform/snitch_cluster:tracer-dma
     steps:
       - uses: actions/checkout@v2
         with:

.gitlab-ci.yml

@@ -40,6 +40,14 @@ docs:
   script:
     - make docs
 
+#####################
+# Python unit tests #
+#####################
+
+pytest:
+  script:
+    - pytest
+
 #################################
 # Build Snitch cluster software #
 #################################

docs/rm/bench/join.md

@@ -0,0 +1 @@
+::: join

docs/rm/bench/roi.md

@@ -0,0 +1 @@
+::: roi

docs/rm/bench/visualize.md

@@ -0,0 +1 @@
+::: visualize

docs/rm/trace/annotate.md

@@ -0,0 +1 @@
+::: annotate

docs/rm/trace/events.md

@@ -0,0 +1 @@
+::: events

docs/rm/trace/gen_trace.md

@@ -0,0 +1 @@
+::: gen_trace

hw/future/src/dma/axi_dma_backend.sv

@@ -288,7 +288,6 @@ module axi_dma_backend #(
   //--------------------------------------
   //pragma translate_off
 `ifndef SYNTHESYS
-`ifndef VERILATOR
   generate
     if (DmaTracing) begin : gen_dma_tracer
       string fn;
@@ -595,7 +594,6 @@ module axi_dma_backend #(
       end
     end
   endgenerate
-`endif
 `endif
   //pragma translate_on
 endmodule : axi_dma_backend

mkdocs.yml

@@ -25,7 +25,7 @@ plugins:
   - mkdocstrings:
       handlers:
         python:
-          paths: [util/sim]
+          paths: [util/sim, util/trace, util/bench]
   - macros:
       on_error_fail: true
 use_directory_urls: false
@@ -57,6 +57,14 @@ nav:
               - sim_utils: rm/sim/sim_utils.md
               - rm/sim/Simulation.md
               - rm/sim/Simulator.md
+          - Trace Utilities:
+              - gen_trace.py: rm/trace/gen_trace.md
+              - annotate.py: rm/trace/annotate.md
+              - events.py: rm/trace/events.md
+          - Benchmarking Utilities:
+              - join.py: rm/bench/join.md
+              - roi.py: rm/bench/roi.md
+              - visualize.py: rm/bench/visualize.md
           - Snitch Runtime:
               - Pages: runtime/Pages/index.md
               - Files: runtime/Files/index.md

python-requirements.txt

@@ -2,24 +2,28 @@
 # Licensed under the Apache License, Version 2.0, see LICENSE for details.
 # SPDX-License-Identifier: Apache-2.0
 
+# Keep sorted.
 bin2coe
 dataclasses
 editorconfig-checker==2.3.51
 flake8
 gitpython
 hjson
+json5
 jsonref
 jsonschema
 mako
+matplotlib
+pandas
 progressbar2
-tabulate
-yamllint
-pyyaml
+psutil
+pyelftools
 pytablewriter
+pytest
+pyyaml
+tabulate
 termcolor
-pandas
-pyelftools
-psutil
+yamllint
 
 -r docs/requirements.txt
 -r sw/dnn/requirements.txt

sw/snRuntime/api/sync_decls.h

@@ -9,6 +9,7 @@ typedef struct {
 
 extern volatile uint32_t _snrt_mutex;
 extern volatile snrt_barrier_t _snrt_barrier;
+extern volatile uint32_t _reduction_result;
 
 inline volatile uint32_t *snrt_mutex();
 

sw/snRuntime/src/start.c

@@ -97,10 +97,14 @@ static inline void snrt_init_libs() { snrt_alloc_init(); }
 #endif
 
 #ifdef SNRT_CRT0_EXIT
-static inline void snrt_exit(int exit_code) {
+static inline void snrt_exit_default(int exit_code) {
+    exit_code = snrt_global_all_to_all_reduction(exit_code);
     if (snrt_global_core_idx() == 0)
         *(snrt_exit_code_destination()) = (exit_code << 1) | 1;
 }
+#ifndef SNRT_CRT0_ALTERNATE_EXIT
+static inline void snrt_exit(int exit_code) { snrt_exit_default(exit_code); }
+#endif
 #endif
 
 void snrt_main() {

sw/snRuntime/src/sync.c

@@ -8,6 +8,7 @@
 
 volatile uint32_t _snrt_mutex;
 volatile snrt_barrier_t _snrt_barrier;
+volatile uint32_t _reduction_result;
 
 //================================================================================
 // Functions

sw/snRuntime/src/sync.h

@@ -81,6 +81,12 @@ inline void snrt_global_barrier() {
     snrt_cluster_hw_barrier();
 }
 
+inline uint32_t snrt_global_all_to_all_reduction(uint32_t value) {
+    __atomic_add_fetch(&_reduction_result, value, __ATOMIC_RELAXED);
+    snrt_global_barrier();
+    return _reduction_result;
+}
+
 /**
  * @brief Generic barrier
  *

sw/tests/event_unit.c

@@ -9,7 +9,6 @@ volatile static uint32_t sum = 0;
 static void task(void *arg, uint32_t argc) {
     uint32_t arg0 = ((uint32_t *)arg)[0];
     __atomic_add_fetch(&sum, arg0, __ATOMIC_RELAXED);
-    printf("work arg[0] = %d argc = %d\n", arg0, argc);
 }
 
 uint32_t run_and_verify_task(uint32_t *arg, uint32_t n_workers) {

sw/tests/fp16_comparison_scalar.c

@@ -6,9 +6,9 @@
 #include "printf.h"
 
 int main() {
-    int errs = 40;
-
     if (snrt_is_compute_core()) {
+        int errs = 40;
+
         uint32_t i8a = 0xFFFF4248;   // 3.14
         uint32_t i8an = 0xFFFFC248;  // -3.14
         uint32_t i8b = 0xFFFF3E79;   // 1.618
@@ -210,7 +210,8 @@ int main() {
 
         errs -= (cmp0 == 0x1);
         errs -= (cmp1 == 0x1);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }

sw/tests/fp16_comparison_vector.c

@@ -6,9 +6,9 @@
 #include "printf.h"
 
 int main() {
-    int errs = 64;
-
     if (snrt_is_compute_core()) {
+        int errs = 64;
+
         uint32_t fa16 = 0x4048F5C3;   // 0x4248 3.14
         uint32_t fa16n = 0xC048F5C3;  // 0xC248 -3.14
         uint32_t fb16 = 0x3FCF1AA0;   // 0x3E79  1.618
@@ -287,7 +287,8 @@ int main() {
             "vfeq.h %1, ft8, ft0\n"
             : "+r"(cmp0));
         errs -= (cmp0 == 0xf);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }

sw/tests/fp16_computation_scalar.c

@@ -4,9 +4,9 @@
 #include <snrt.h>
 
 int main() {
-    int errs = 33;
-
     if (snrt_is_compute_core()) {
+        int errs = 33;
+
         uint32_t i_a = 0xFFFF4248;   // 3.14
         uint32_t i_an = 0xFFFFC248;  // -3.14
         uint32_t i_b = 0xFFFF3E79;   // 1.618
@@ -317,7 +317,8 @@ int main() {
             "feq.h %0, ft3, ft0\n"
             : "+r"(res0));
         errs -= (res0 == 0x1);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }

sw/tests/fp16_computation_vector.c

@@ -4,9 +4,9 @@
 #include "snrt.h"
 
 int main() {
-    int errs = 46;
-
     if (snrt_is_compute_core()) {
+        int errs = 46;
+
         uint32_t i_a = 0x4048F5C3;   // 3.14 0
         uint32_t i_an = 0xC048F5C3;  // -3.14
         uint32_t i_b = 0x3FCF1AA0;   // 1.618 2
@@ -563,7 +563,8 @@ int main() {
             "vfeq.h %0, ft7, ft0\n"
             : "+r"(res0));
         errs -= (res0 == 0xf);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }

sw/tests/fp16alt_comparison_scalar.c

@@ -6,9 +6,9 @@
 #include "printf.h"
 
 int main() {
-    int errs = 40;
-
     if (snrt_is_compute_core()) {
+        int errs = 40;
+
         uint32_t i8a = 0xFFFF4049;   // 3.14
         uint32_t i8an = 0xFFFFC049;  // -3.14
         uint32_t i8b = 0xFFFF3FCF;   // 1.618
@@ -213,7 +213,8 @@ int main() {
 
         errs -= (cmp0 == 0x1);
         errs -= (cmp1 == 0x1);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }

sw/tests/fp16alt_comparison_vector.c

@@ -6,9 +6,9 @@
 #include "printf.h"
 
 int main() {
-    int errs = 64;
-
     if (snrt_is_compute_core()) {
+        int errs = 64;
+
         uint32_t fa16 = 0x4048F5C3;   // 0x4248 3.14
         uint32_t fa16n = 0xC048F5C3;  // 0xC248 -3.14
         uint32_t fb16 = 0x3FCF1AA0;   // 0x3E79  1.618
@@ -289,7 +289,8 @@ int main() {
             "vfeq.ah %1, ft8, ft0\n"
             : "+r"(cmp0));
         errs -= (cmp0 == 0xf);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }

sw/tests/fp16alt_computation_scalar.c

@@ -4,9 +4,9 @@
 #include <snrt.h>
 
 int main() {
-    int errs = 33;
-
     if (snrt_is_compute_core()) {
+        int errs = 33;
+
         uint32_t i_a = 0xFFFF4049;   // 3.14
         uint32_t i_an = 0xFFFFC049;  // -3.14
         uint32_t i_b = 0xFFFF3FCF;   // 1.618
@@ -319,7 +319,8 @@ int main() {
             "feq.ah %0, ft3, ft0\n"
             : "+r"(res0));
         errs -= (res0 == 0x1);
-    }
 
-    return errs;
+        return errs;
+    }
+    return 0;
 }