Merge pull request #24 from flagos-ai/cuda_v1

merge

Author: zyq1105331849

README.md

@@ -128,9 +128,9 @@ python tests/test_spsv.py <dir/> --csv-coo out.csv      # same CSV columns as CS
 **test_spsm.py** - SpSM (triangular matrix-matrix solve; **square** matrices only):
 
 ```bash
-python tests/test_spsm.py --synthetic --n 512 --rhs 32
-python tests/test_spsm.py <dir/> --csv-csr spsm_csr.csv --rhs 32
-python tests/test_spsm.py <dir/> --csv-coo spsm_coo.csv --rhs 32
+python tests/test_spsm.py --synthetic --n 512 --rhs 1024
+python tests/test_spsm.py <dir/> --csv-csr spsm_csr.csv --rhs 1024
+python tests/test_spsm.py <dir/> --csv-coo spsm_coo.csv --rhs 1024
 ```
 
 **test_gather.py** / **test_scatter.py** - gather/scatter benchmarks (pytest or `python tests/test_gather.py`).

README_cn.md

@@ -126,9 +126,9 @@ python tests/test_spsv.py <目录/> --csv-coo out.csv     # 列与 CSR 相同
 **test_spsm.py** - SpSM（三角矩阵-稠密矩阵求解；**仅方阵**）：
 
 ```bash
-python tests/test_spsm.py --synthetic --n 512 --rhs 32
-python tests/test_spsm.py <目录/> --csv-csr spsm_csr.csv --rhs 32
-python tests/test_spsm.py <目录/> --csv-coo spsm_coo.csv --rhs 32
+python tests/test_spsm.py --synthetic --n 512 --rhs 1024
+python tests/test_spsm.py <目录/> --csv-csr spsm_csr.csv --rhs 1024
+python tests/test_spsm.py <目录/> --csv-coo spsm_coo.csv --rhs 1024
 ```
 
 **test_gather.py** / **test_scatter.py** - gather/scatter 基准（pytest 或 `python tests/test_gather.py`）。

benchmark/benchmark_spsm.py

@@ -1,4 +1,4 @@
-"""Run SpSM synthetic benchmark. From project root: python benchmark/benchmark_spsm.py [--n 512 --rhs 32]."""
+"""Run SpSM synthetic benchmark. From project root: python benchmark/benchmark_spsm.py [--n 512 --rhs 1024]."""
 
 import sys
 from pathlib import Path

conf/operators.yaml

@@ -261,6 +261,38 @@ ops:
     stages:
       - beta: "1.0"
 
+  - id: spsv_descriptor_api
+    description: |
+      Provides descriptor, buffer-size, analysis, preprocess, and solve APIs for SpSV workflows.
+    for:
+      - flagsparse_create_spsv_handle
+      - flagsparse_create_dnvec
+      - flagsparse_create_spmat_csr
+      - flagsparse_create_spmat_coo
+      - flagsparse_spsv_buffer_size
+      - flagsparse_spsv_buffer_size_ex
+      - flagsparse_spsv_create_workspace
+      - flagsparse_spsv_analysis_csr
+      - flagsparse_spsv_analysis_coo
+      - flagsparse_spsv_analysis_ex
+      - flagsparse_spsv_preprocess_csr
+      - flagsparse_spsv_preprocess_coo
+      - flagsparse_spsv_solve_csr
+      - flagsparse_spsv_solve_coo
+      - flagsparse_spsv_solve_ex
+    labels:
+      - flagsparse
+      - sparse
+      - csr
+      - coo
+      - descriptor
+      - public-api
+    kind:
+      - SparseSolver
+      - Utility
+    stages:
+      - beta: "1.0"
+
   - id: spsm_csr
     description: |
       Solves sparse triangular systems with dense matrix right-hand sides in CSR format.
@@ -292,3 +324,31 @@ ops:
       - SparseSolver
     stages:
       - beta: "1.0"
+
+  - id: sparse_format_constructors
+    description: |
+      Creates sparse matrix containers and converts COO data into supported sparse formats.
+    for:
+      - create_csr_matrix
+      - create_coo_matrix
+      - create_csc_matrix
+      - create_bsr_matrix
+      - create_sell_matrix
+      - create_blocked_ell_matrix
+      - coo_to_csr
+      - coo_to_csc
+      - coo_to_bsr
+      - coo_to_sell
+      - coo_to_blocked_ell
+      - generate_random_sparse_matrix
+      - read_mtx_file
+    labels:
+      - flagsparse
+      - sparse
+      - format
+      - public-api
+    kind:
+      - SparseFormat
+      - Utility
+    stages:
+      - beta: "1.0"

ops_support.csv

@@ -163,12 +163,12 @@ spmv,CSR,int64,complex128,trans,triton,SUPPORTED
 spmv,CSR,int64,complex128,conj,triton,SUPPORTED
 spsm,COO,int32,float32,non,triton,SUPPORTED
 spsm,COO,int32,float64,non,triton,SUPPORTED
-spsm,COO,int64,float32,non,triton,SUPPORTED
-spsm,COO,int64,float64,non,triton,SUPPORTED
+spsm,COO,int32,complex64,non,triton,SUPPORTED
+spsm,COO,int32,complex128,non,triton,SUPPORTED
 spsm,CSR,int32,float32,non,triton,SUPPORTED
 spsm,CSR,int32,float64,non,triton,SUPPORTED
-spsm,CSR,int64,float32,non,triton,SUPPORTED
-spsm,CSR,int64,float64,non,triton,SUPPORTED
+spsm,CSR,int32,complex64,non,triton,SUPPORTED
+spsm,CSR,int32,complex128,non,triton,SUPPORTED
 spsv,COO,int32,float32,non,triton,SUPPORTED
 spsv,COO,int32,float32,trans,triton,SUPPORTED
 spsv,COO,int32,float64,non,triton,SUPPORTED

ops_support.py

@@ -268,8 +268,8 @@ def registry(modules: dict[str, SourceModule]) -> tuple[ApiSpec, ...]:
         ApiSpec("sddmm", "flagsparse_sddmm_csr", "sddmm_csr", "CSR", "triton", value_const="SUPPORTED_SDDMM_VALUE_DTYPES", index_const="SUPPORTED_INDEX_DTYPES", ops=("non",)),
         ApiSpec("spsv", "flagsparse_spsv_csr", "spsv", "CSR", "triton", value_const="SUPPORTED_SPSV_VALUE_DTYPES", index_const="SUPPORTED_SPSV_INDEX_DTYPES", ops=("NON_TRANS", "TRANS"), notes="TRANS support is narrower than NON_TRANS; see combo constants"),
         ApiSpec("spsv", "flagsparse_spsv_coo", "spsv", "COO", "triton", value_const="SUPPORTED_SPSV_VALUE_DTYPES", index_const="SUPPORTED_SPSV_INDEX_DTYPES", ops=("NON_TRANS", "TRANS"), notes="TRANS support is narrower than NON_TRANS; see combo constants"),
-        ApiSpec("spsm", "flagsparse_spsm_csr", "spsm", "CSR", "triton", value_const="SUPPORTED_SPSM_VALUE_DTYPES", index_const="SUPPORTED_SPSM_INDEX_DTYPES", ops=("NON_TRANS",), notes="opA/opB must both be NON_TRANS; row-major dense layout only"),
-        ApiSpec("spsm", "flagsparse_spsm_coo", "spsm", "COO", "triton", value_const="SUPPORTED_SPSM_VALUE_DTYPES", index_const="SUPPORTED_SPSM_INDEX_DTYPES", ops=("NON_TRANS",), notes="opA/opB must both be NON_TRANS; row-major dense layout only"),
+        ApiSpec("spsm", "flagsparse_spsm_csr", "spsm", "CSR", "triton", value_const="SUPPORTED_SPSM_VALUE_DTYPES", index_const="SUPPORTED_SPSM_INDEX_DTYPES", values=("float32", "float64", "complex64", "complex128"), indices=("int32",), ops=("NON_TRANS",), notes="opA/opB must both be NON_TRANS; row-major dense layout only"),
+        ApiSpec("spsm", "flagsparse_spsm_coo", "spsm", "COO", "triton", value_const="SUPPORTED_SPSM_VALUE_DTYPES", index_const="SUPPORTED_SPSM_INDEX_DTYPES", values=("float32", "float64", "complex64", "complex128"), indices=("int32",), ops=("NON_TRANS",), notes="opA/opB must both be NON_TRANS; row-major dense layout only"),
     )
 
 

pyproject.toml

@@ -10,8 +10,7 @@ readme = "README.md"
 requires-python = ">=3.10"
 dependencies = []
 authors = [{ name = "Your Name", email = "your.email@example.com" }]
-license = "Apache-2.0"
-license-files = ["LICENSE"]
+license = { file = "LICENSE" }
 classifiers = [
     "Development Status :: 4 - Beta",
     "Intended Audience :: Developers",