`main()`: matrix B always loaded from the same files as A — general A×B SpGEMM is impossible

[itsme-ranger]

Bug

src/matmul/AxBRowIP.c loads both input matrices from identical files:

// values
while (fscanf(file, "%f", &val) == 1) {
    aD[idx] = val;
    bD[idx] = val;   // ← same value as aD
    idx++;
}

// column indices
while (fscanf(file1, "%d", &number) == 1) {
    aC[idx] = number;
    bC[idx] = number;   // ← same index as aC
    idx++;
}

// row pointers
while (fscanf(file2, "%d", &number) == 1) {
    aR[idx] = number;
    bR[idx] = number;   // ← same pointer as aR
    idx++;
}

The program always computes A × A, never a general A × B with two distinct sparse matrices.

Impact

• The core library operation (SpGEMM: multiply two different sparse matrices) cannot be exercised at all.
• There is no correctness signal: A × A will produce a result, but it cannot detect bugs that only manifest when A ≠ B (e.g. dimension mismatches, wrong row-pointer indexing for B).
• All benchmark numbers reported are for the degenerate A² case, not representative of real workloads where A and B differ in structure and density.

Fix

Add a second set of input files (or accept file paths as CLI arguments) and load bD, bC, bR from them independently:

// Example: separate files for B
FILE *fileB_val = fopen("matrix_data/B_CSR_values.txt", "r");
FILE *fileB_col = fopen("matrix_data/B_CSR_colIdx.txt", "r");
FILE *fileB_row = fopen("matrix_data/B_CSR_rowPtr.txt", "r");
// ... load into bD, bC, bR independently

Or accept matrix paths as argv so any two matrices can be multiplied without editing source.

Note

This is distinct from issue #21 (CSR×CSR general support at the algorithm level). That issue addresses the kernel; this issue means even the existing kernel cannot be called with two different matrices due to the hardcoded file loading.