Merge pull request #2037 from fredrik-johansson/matmul

Matrix multiplication for nfloat_complex
flintlib · Jul 19, 2024 · 91f0ece · 91f0ece
2 parents 886c317 + 60d5397
commit 91f0ece
Show file tree

Hide file tree

Showing 7 changed files with 952 additions and 33 deletions.
diff --git a/doc/source/nfloat.rst b/doc/source/nfloat.rst
@@ -412,4 +412,8 @@ real pairs.
               int _nfloat_complex_vec_set(nfloat_complex_ptr res, nfloat_complex_srcptr x, slong len, gr_ctx_t ctx)
               int _nfloat_complex_vec_add(nfloat_complex_ptr res, nfloat_complex_srcptr x, nfloat_complex_srcptr y, slong len, gr_ctx_t ctx)
               int _nfloat_complex_vec_sub(nfloat_complex_ptr res, nfloat_complex_srcptr x, nfloat_complex_srcptr y, slong len, gr_ctx_t ctx)
-
+              int nfloat_complex_mat_mul_fixed_classical(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx)
+              int nfloat_complex_mat_mul_waksman(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx)
+              int nfloat_complex_mat_mul_block(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, slong min_block_size, gr_ctx_t ctx)
+              int nfloat_complex_mat_mul_reorder(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx)
+              int nfloat_complex_mat_mul(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx)
diff --git a/src/gr_mat/test_approx_mul.c b/src/gr_mat/test_approx_mul.c
@@ -57,9 +57,6 @@ void gr_mat_test_approx_mul_max_norm(gr_method_mat_binary_op mul_impl, gr_srcptr
 
         status |= gr_mat_randtest(A, state, ctx);
         status |= gr_mat_randtest(B, state, ctx);
-        status |= gr_mat_entrywise_unary_op(A, (gr_method_unary_op) gr_abs, A, ctx);
-        status |= gr_mat_entrywise_unary_op(B, (gr_method_unary_op) gr_abs, B, ctx);
-
         status |= gr_mat_randtest(C, state, ctx);
         status |= gr_mat_randtest(D, state, ctx);
 

diff --git a/src/nfloat.h b/src/nfloat.h
@@ -563,6 +563,12 @@ int _nfloat_complex_vec_sub(nfloat_complex_ptr res, nfloat_complex_srcptr x, nfl
 int _nfloat_complex_vec_dot(nfloat_complex_ptr res, nfloat_complex_srcptr initial, int subtract, nfloat_complex_srcptr x, nfloat_complex_srcptr y, slong len, gr_ctx_t ctx);
 int _nfloat_complex_vec_dot_rev(nfloat_complex_ptr res, nfloat_complex_srcptr initial, int subtract, nfloat_complex_srcptr x, nfloat_complex_srcptr y, slong len, gr_ctx_t ctx);
 
+int nfloat_complex_mat_mul_fixed_classical(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx);
+int nfloat_complex_mat_mul_waksman(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx);
+int nfloat_complex_mat_mul_block(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, slong min_block_size, gr_ctx_t ctx);
+int nfloat_complex_mat_mul_reorder(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx);
+int nfloat_complex_mat_mul(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t ctx);
+
 #ifdef __cplusplus
 }
 #endif

diff --git a/src/nfloat/complex.c b/src/nfloat/complex.c
@@ -1792,8 +1792,8 @@ gr_method_tab_input _nfloat_complex_methods_input[] =
 /*
     {GR_METHOD_POLY_MULLOW,     (gr_funcptr) nfloat_complex_poly_mullow},
     {GR_METHOD_POLY_ROOTS_OTHER,(gr_funcptr) nfloat_complex_poly_roots_other},
-    {GR_METHOD_MAT_MUL,         (gr_funcptr) nfloat_complex_mat_mul},
 */
+    {GR_METHOD_MAT_MUL,         (gr_funcptr) nfloat_complex_mat_mul},
     {GR_METHOD_MAT_DET,         (gr_funcptr) gr_mat_det_generic_field},
     {GR_METHOD_MAT_FIND_NONZERO_PIVOT,     (gr_funcptr) gr_mat_find_nonzero_pivot_large_abs},