new matrix allocation mechanism

This commit is contained in:
Florian Stecker 2019-02-08 13:35:02 +01:00
parent 3d8378aa16
commit 870ae7d2d2
6 changed files with 152 additions and 154 deletions

24
draw.c
View File

@ -13,22 +13,22 @@ vector_t cross(vector_t a, vector_t b)
vector_t apply(DrawingContext *ctx, gsl_matrix *m, vector_t x)
{
gsl_vector *tmp = getTempVector(ctx);
gsl_vector *tmp2 = getTempVector(ctx);
gsl_vector *tmp = getTempVector(ctx->ws);
gsl_vector *tmp2 = getTempVector(ctx->ws);
vector_t out;
LOOP(i) gsl_vector_set(tmp, i, x.x[i]);
gsl_blas_dgemv(CblasNoTrans, 1.0, m, tmp, 0.0, tmp2);
LOOP(i) out.x[i] = gsl_vector_get(tmp2, i);
releaseTempVectors(ctx, 2);
releaseTempVectors(ctx->ws, 2);
}
int fixedPoints(DrawingContext *ctx, const char *word, vector_t *out)
{
gsl_matrix *tmp = getTempMatrix(ctx);
gsl_matrix *ev = getTempMatrix(ctx);
gsl_matrix **gen = getTempMatrices(ctx, 3);
gsl_matrix *tmp = getTempMatrix(ctx->ws);
gsl_matrix *ev = getTempMatrix(ctx->ws);
gsl_matrix **gen = getTempMatrices(ctx->ws, 3);
initializeTriangleGenerators(gen, ctx->cartan);
@ -42,16 +42,16 @@ int fixedPoints(DrawingContext *ctx, const char *word, vector_t *out)
LOOP(i) LOOP(j) out[i].x[j] = gsl_matrix_get(ev, j, i);
releaseTempMatrices(ctx, 5);
releaseTempMatrices(ctx->ws, 5);
return count;
}
void transformFrameStd(DrawingContext *ctx, vector_t *x, gsl_matrix *out)
{
gsl_matrix *tmp = getTempMatrix(ctx);
gsl_vector *fourth = getTempVector(ctx);
gsl_vector *lambda = getTempVector(ctx);
gsl_matrix *tmp = getTempMatrix(ctx->ws);
gsl_vector *fourth = getTempVector(ctx->ws);
gsl_vector *lambda = getTempVector(ctx->ws);
int s;
LOOP(i) LOOP(j) gsl_matrix_set(out, j, i, x[i].x[j]);
@ -63,8 +63,8 @@ void transformFrameStd(DrawingContext *ctx, vector_t *x, gsl_matrix *out)
gsl_matrix_fprintf(stdout, out, "%f");
releaseTempMatrices(ctx, 1);
releaseTempVectors(ctx, 2);
releaseTempMatrices(ctx->ws, 1);
releaseTempVectors(ctx->ws, 2);
}
// level 1: the elementary drawing functions, drawPoint, drawSegment2d

View File

@ -29,16 +29,16 @@ void initializeTriangleGenerators(gsl_matrix **gen, gsl_matrix *cartan)
int computeLimitCurve(DrawingContext *ctx)
{
workspace_t *ws = ctx->ws;
gsl_matrix *cartan_pos = getTempMatrix(ctx);
gsl_matrix *cob_pos = getTempMatrix(ctx);
gsl_matrix *coxeter_pos = getTempMatrix(ctx);
gsl_matrix *coxeter_fixedpoints_pos = getTempMatrix(ctx);
gsl_matrix *fixedpoints_pos = getTempMatrix(ctx);
gsl_matrix *coxeter = getTempMatrix(ctx);
gsl_matrix *coxeter_fixedpoints = getTempMatrix(ctx);
gsl_matrix *fixedpoints = getTempMatrix(ctx);
gsl_matrix **gen = getTempMatrices(ctx, 3);
gsl_matrix **elements = getTempMatrices(ctx, ctx->n_group_elements);
gsl_matrix *cartan_pos = getTempMatrix(ctx->ws);
gsl_matrix *cob_pos = getTempMatrix(ctx->ws);
gsl_matrix *coxeter_pos = getTempMatrix(ctx->ws);
gsl_matrix *coxeter_fixedpoints_pos = getTempMatrix(ctx->ws);
gsl_matrix *fixedpoints_pos = getTempMatrix(ctx->ws);
gsl_matrix *coxeter = getTempMatrix(ctx->ws);
gsl_matrix *coxeter_fixedpoints = getTempMatrix(ctx->ws);
gsl_matrix *fixedpoints = getTempMatrix(ctx->ws);
gsl_matrix **gen = getTempMatrices(ctx->ws, 3);
gsl_matrix **elements = getTempMatrices(ctx->ws, ctx->n_group_elements);
groupelement_t *group = ctx->group;
int success = 0;
@ -92,7 +92,7 @@ int computeLimitCurve(DrawingContext *ctx)
success = 1;
error_out:
releaseTempMatrices(ctx, 11+ctx->n_group_elements);
releaseTempMatrices(ctx->ws, 11+ctx->n_group_elements);
return success;
}

116
linalg.c
View File

@ -24,11 +24,16 @@ workspace_t *workspace_alloc(int n)
result->evec_complex = gsl_matrix_complex_alloc(n, n);
result->eval_real = gsl_vector_alloc(n);
result->evec_real = gsl_matrix_alloc(n, n);
result->tmp = gsl_matrix_alloc(n, n);
result->permutation = gsl_permutation_alloc(n);
for(int i = 0; i < 20; i++)
result->stack[i] = gsl_matrix_alloc(n, n);
result->tmp_mat = malloc(MAX_TEMP_MATRICES*sizeof(gsl_matrix*));
for(int i = 0; i < MAX_TEMP_MATRICES; i++)
result->tmp_mat[i] = gsl_matrix_alloc(3, 3);
result->tmp_mat_used = 0;
result->tmp_vec = malloc(MAX_TEMP_MATRICES*sizeof(gsl_vector*));
for(int i = 0; i < MAX_TEMP_MATRICES; i++)
result->tmp_vec[i] = gsl_vector_alloc(3);
result->tmp_vec_used = 0;
return result;
}
@ -41,10 +46,14 @@ void workspace_free(workspace_t *workspace)
gsl_matrix_complex_free(workspace->evec_complex);
gsl_vector_free(workspace->eval_real);
gsl_matrix_free(workspace->evec_real);
gsl_matrix_free(workspace->tmp);
gsl_permutation_free(workspace->permutation);
for(int i = 0; i < 20; i++)
gsl_matrix_free(workspace->stack[i]);
for(int i = 0; i < MAX_TEMP_MATRICES; i++)
gsl_matrix_free(workspace->tmp_mat[i]);
free(workspace->tmp_mat);
for(int i = 0; i < MAX_TEMP_VECTORS; i++)
gsl_vector_free(workspace->tmp_vec[i]);
free(workspace->tmp_vec);
}
/************************************************** basic operations ********************************************************/
@ -52,16 +61,24 @@ void workspace_free(workspace_t *workspace)
void invert(gsl_matrix *in, gsl_matrix *out, workspace_t *ws)
{
int s;
gsl_matrix_memcpy(ws->tmp, in);
gsl_linalg_LU_decomp(ws->tmp, ws->permutation, &s);
gsl_linalg_LU_invert(ws->tmp, ws->permutation, out);
gsl_matrix *tmp = getTempMatrix(ws);
gsl_matrix_memcpy(tmp, in);
gsl_linalg_LU_decomp(tmp, ws->permutation, &s);
gsl_linalg_LU_invert(tmp, ws->permutation, out);
releaseTempMatrices(ws, 1);
}
void conjugate(gsl_matrix *in, gsl_matrix *conjugator, gsl_matrix *out, workspace_t *ws)
{
gsl_matrix *tmp = getTempMatrix(ws);
invert(conjugator, out, ws); // use out to temporarily store inverse conjugator
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, in, out, 0.0, ws->tmp); // in * conjugator^{-1}
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, conjugator, ws->tmp, 0.0, out);
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, in, out, 0.0, tmp); // in * conjugator^{-1}
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, conjugator, tmp, 0.0, out);
releaseTempMatrices(ws, 1);
}
void multiply(gsl_matrix *a, gsl_matrix *b, gsl_matrix *out)
@ -71,14 +88,18 @@ void multiply(gsl_matrix *a, gsl_matrix *b, gsl_matrix *out)
void multiply_right(gsl_matrix *a, gsl_matrix *b, workspace_t *ws)
{
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, a, b, 0.0, ws->stack[0]);
gsl_matrix_memcpy(a, ws->stack[0]);
gsl_matrix *tmp = getTempMatrix(ws);
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, a, b, 0.0, tmp);
gsl_matrix_memcpy(a, tmp);
releaseTempMatrices(ws, 1);
}
void multiply_left(gsl_matrix *a, gsl_matrix *b, workspace_t *ws)
{
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, a, b, 0.0, ws->stack[0]);
gsl_matrix_memcpy(b, ws->stack[0]);
gsl_matrix *tmp = getTempMatrix(ws);
gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, a, b, 0.0, tmp);
gsl_matrix_memcpy(b, tmp);
releaseTempMatrices(ws, 1);
}
void multiply_many(workspace_t *ws, gsl_matrix *out, int n, ...)
@ -98,11 +119,15 @@ void multiply_many(workspace_t *ws, gsl_matrix *out, int n, ...)
void cartan_calc(gsl_matrix *g, double *mu, workspace_t *ws)
{
gsl_matrix_memcpy(ws->tmp, g);
gsl_linalg_SV_decomp(ws->tmp, ws->evec_real, ws->eval_real, ws->work_sv);
gsl_matrix *tmp = getTempMatrix(ws);
for(int i = 0; i < ws->n - 1; i++)
mu[i] = log(gsl_vector_get(ws->eval_real, i) / gsl_vector_get(ws->eval_real, i+1));
gsl_matrix_memcpy(tmp, g);
gsl_linalg_SV_decomp(tmp, ws->evec_real, ws->eval_real, ws->work_sv);
for(int i = 0; i < ws->n - 1; i++)
mu[i] = log(gsl_vector_get(ws->eval_real, i) / gsl_vector_get(ws->eval_real, i+1));
releaseTempMatrices(ws, 1);
}
void initialize(gsl_matrix *g, double *data, int x, int y)
@ -140,9 +165,15 @@ double trace(gsl_matrix *g)
double determinant(gsl_matrix *g, workspace_t *ws)
{
int s;
gsl_matrix_memcpy(ws->tmp, g);
gsl_linalg_LU_decomp(ws->tmp, ws->permutation, &s);
return gsl_linalg_LU_det(ws->tmp, s);
double result;
gsl_matrix *tmp = getTempMatrix(ws);
gsl_matrix_memcpy(tmp, g);
gsl_linalg_LU_decomp(tmp, ws->permutation, &s);
result = gsl_linalg_LU_det(tmp, s);
releaseTempMatrices(ws, 1);
return result;
}
int jordan_calc(gsl_matrix *g, double *mu, workspace_t *ws)
@ -168,9 +199,12 @@ int jordan_calc(gsl_matrix *g, double *mu, workspace_t *ws)
int eigenvectors(gsl_matrix *g, gsl_matrix *evec_real, workspace_t *ws)
{
gsl_matrix_memcpy(ws->stack[0], g);
gsl_matrix *g_ = getTempMatrix(ws);
int success = 0;
gsl_matrix_memcpy(g_, g);
gsl_eigen_nonsymmv_params(0, ws->work_nonsymmv);
int r = gsl_eigen_nonsymmv(ws->stack[0], ws->eval_complex, ws->evec_complex, ws->work_nonsymmv);
int r = gsl_eigen_nonsymmv(g_, ws->eval_complex, ws->evec_complex, ws->work_nonsymmv);
ERROR(r, "gsl_eigen_nonsymmv failed!\n");
gsl_eigen_nonsymmv_sort(ws->eval_complex, ws->evec_complex, GSL_EIGEN_SORT_ABS_DESC);
@ -181,21 +215,27 @@ int eigenvectors(gsl_matrix *g, gsl_matrix *evec_real, workspace_t *ws)
real = 0;
if(!real)
return 0; // non-real eigenvalues!
goto eigenvectors_out;
for(int i = 0; i < ws->n; i++)
for(int j = 0; j < ws->n; j++)
gsl_matrix_set(evec_real, i, j, GSL_REAL(gsl_matrix_complex_get(ws->evec_complex, i, j)));
return 1;
success = 1;
eigenvectors_out:
releaseTempMatrices(ws, 1);
return success;
}
// only fills in the real eigenvectors and returns their count
int real_eigenvectors(gsl_matrix *g, gsl_matrix *evec_real, workspace_t *ws)
{
gsl_matrix_memcpy(ws->stack[0], g);
gsl_matrix *g_ = getTempMatrix(ws);
gsl_matrix_memcpy(g_, g);
gsl_eigen_nonsymmv_params(0, ws->work_nonsymmv);
int r = gsl_eigen_nonsymmv(ws->stack[0], ws->eval_complex, ws->evec_complex, ws->work_nonsymmv);
int r = gsl_eigen_nonsymmv(g_, ws->eval_complex, ws->evec_complex, ws->work_nonsymmv);
ERROR(r, "gsl_eigen_nonsymmv failed!\n");
gsl_eigen_nonsymmv_sort(ws->eval_complex, ws->evec_complex, GSL_EIGEN_SORT_ABS_DESC);
@ -210,23 +250,30 @@ int real_eigenvectors(gsl_matrix *g, gsl_matrix *evec_real, workspace_t *ws)
}
}
releaseTempMatrices(ws, 1);
return real;
}
void eigensystem_symm(gsl_matrix *g, gsl_vector *eval, gsl_matrix *evec, workspace_t *ws)
{
gsl_matrix_memcpy(ws->stack[0], g);
int r = gsl_eigen_symmv (ws->stack[0], eval, evec, ws->work_symmv);
gsl_matrix *g_ = getTempMatrix(ws);
gsl_matrix_memcpy(g_, g);
int r = gsl_eigen_symmv (g_, eval, evec, ws->work_symmv);
ERROR(r, "gsl_eigen_symmv failed!\n");
gsl_eigen_symmv_sort(eval, evec, GSL_EIGEN_SORT_ABS_DESC);
releaseTempMatrices(ws, 1);
}
// returns number of positive directions and matrix transforming TO diagonal basis
int diagonalize_symmetric_form(gsl_matrix *A, gsl_matrix *cob, workspace_t *ws)
{
gsl_matrix_memcpy(ws->stack[0], A);
int r = gsl_eigen_symmv (ws->stack[0], ws->eval_real, cob, ws->work_symmv);
gsl_matrix *A_ = getTempMatrix(ws);
gsl_matrix_memcpy(A_, A);
int r = gsl_eigen_symmv (A_, ws->eval_real, cob, ws->work_symmv);
ERROR(r, "gsl_eigen_symmv failed!\n");
gsl_eigen_symmv_sort(ws->eval_real, cob, GSL_EIGEN_SORT_VAL_ASC);
@ -243,9 +290,6 @@ int diagonalize_symmetric_form(gsl_matrix *A, gsl_matrix *cob, workspace_t *ws)
*gsl_matrix_ptr(cob, i, j) *= sqrt(fabs(gsl_vector_get(ws->eval_real, i)));
}
releaseTempMatrices(ws, 1);
return positive;
// printf("Eigenvalues: %.10f, %.10f, %.10f\n", gsl_vector_get(ws->eval_real, 0), gsl_vector_get(ws->eval_real, 1), gsl_vector_get(ws->eval_real, 2));
// return 0;
}

View File

@ -13,6 +13,9 @@
#define ERROR(condition, msg, ...) if(condition){fprintf(stderr, msg, ##__VA_ARGS__); exit(1);}
#define FCMP(x, y) gsl_fcmp(x, y, 1e-10)
#define MAX_TEMP_MATRICES 60000
#define MAX_TEMP_VECTORS 100
typedef struct _workspace {
int n;
gsl_eigen_nonsymmv_workspace *work_nonsymmv;
@ -22,9 +25,12 @@ typedef struct _workspace {
gsl_matrix_complex *evec_complex;
gsl_vector *eval_real;
gsl_matrix *evec_real;
gsl_matrix *tmp;
gsl_permutation *permutation;
gsl_matrix *stack[20];
gsl_matrix **tmp_mat;
int tmp_mat_used;
gsl_vector **tmp_vec;
int tmp_vec_used;
} workspace_t;
workspace_t *workspace_alloc(int n);
@ -46,4 +52,45 @@ int real_eigenvectors(gsl_matrix *g, gsl_matrix *evec, workspace_t *ws);
void eigenvectors_symm(gsl_matrix *g, gsl_vector *eval, gsl_matrix *evec, workspace_t *ws);
int diagonalize_symmetric_form(gsl_matrix *A, gsl_matrix *cob, workspace_t *ws);
// matrix allocation stuff
static gsl_matrix **getTempMatrices(workspace_t *ws, int n)
{
ERROR(ws->tmp_mat_used + n > MAX_TEMP_MATRICES, "Ran out of temporary matrices. Consider increasing MAX_TEMP_MATRICES\n");
int index = ws->tmp_mat_used;
ws->tmp_mat_used += n;
return ws->tmp_mat + index;
}
static gsl_matrix *getTempMatrix(workspace_t *ws)
{
return *getTempMatrices(ws, 1);
}
static void releaseTempMatrices(workspace_t *ws, int n)
{
ERROR(ws->tmp_mat_used - n < 0, "Released more matrices then in use\n");
ws->tmp_mat_used -= n;
}
static gsl_vector **getTempVectors(workspace_t *ws, int n)
{
ERROR(ws->tmp_vec_used + n > MAX_TEMP_VECTORS, "Ran out of temporary vectors. Consider increasing MAX_TEMP_VECTORS\n");
int index = ws->tmp_vec_used;
ws->tmp_vec_used += n;
return ws->tmp_vec + index;
}
static gsl_vector *getTempVector(workspace_t *ws)
{
return *getTempVectors(ws, 1);
}
static void releaseTempVectors(workspace_t *ws, int n)
{
ERROR(ws->tmp_vec_used - n < 0, "Released more vectors then in use\n");
ws->tmp_vec_used -= n;
}
#endif

49
main.c
View File

@ -43,15 +43,6 @@ void setupContext(DrawingContext *ctx)
ctx->cartan = gsl_matrix_alloc(3, 3);
ctx->cob = gsl_matrix_alloc(3, 3);
ctx->ws = workspace_alloc(3);
ctx->tmp = malloc(MAX_TEMP_MATRICES*sizeof(gsl_matrix*));
for(int i = 0; i < MAX_TEMP_MATRICES; i++)
ctx->tmp[i] = gsl_matrix_alloc(3, 3);
ctx->tmp_used = 0;
ctx->tmp_vec = malloc(MAX_TEMP_MATRICES*sizeof(gsl_vector*));
for(int i = 0; i < MAX_TEMP_MATRICES; i++)
ctx->tmp_vec[i] = gsl_vector_alloc(3);
ctx->tmp_vec_used = 0;
}
void destroyContext(DrawingContext *ctx)
@ -63,13 +54,6 @@ void destroyContext(DrawingContext *ctx)
gsl_matrix_free(ctx->cob);
workspace_free(ctx->ws);
for(int i = 0; i < MAX_TEMP_MATRICES; i++)
gsl_matrix_free(ctx->tmp[i]);
free(ctx->tmp);
for(int i = 0; i < MAX_TEMP_VECTORS; i++)
gsl_vector_free(ctx->tmp_vec[i]);
free(ctx->tmp_vec);
}
void updateMatrices(DrawingContext *ctx)
@ -212,39 +196,6 @@ int main()
updateMatrices(screen_context);
computeLimitCurve(screen_context);
// do stuff
/*
DrawingContext *ctx = screen_context;
gsl_matrix *tmp = getTempMatrix(ctx);
gsl_matrix *ev = getTempMatrix(ctx);
gsl_matrix *evinv = getTempMatrix(ctx);
gsl_matrix **gen = getTempMatrices(ctx, 3);
gsl_matrix *transform = getTempMatrix(ctx);
vector_t eigenvectors[3][3];
char words[3][4] = {"abc", "bca", "cab"};
initializeTriangleGenerators(gen, ctx->cartan);
gsl_matrix_set_identity(tmp);
for(int j = 0; j < strlen(words[0]); j++)
multiply_right(tmp, gen[words[0][j]-'a'], ctx->ws);
real_eigenvectors(tmp, ev, ctx->ws);
invert(ev, evinv, ctx->ws);
LOOP(i) {
gsl_matrix_set_identity(tmp);
for(int j = 0; j < strlen(words[i]); j++)
multiply_right(tmp, gen[words[i][j]-'a'], ctx->ws);
real_eigenvectors(tmp, ev, ctx->ws);
LOOP(j) LOOP(k) eigenvectors[i][j].x[k] = gsl_matrix_get(ev, k, j);
}
// solve the following: (x + y + z) (l) = lambda w
return 0;
*/
info = initCairo(0, KeyPressMask, 200, 200, "Triangle group");
if(!info)
return 1;

44
main.h
View File

@ -10,8 +10,6 @@
#define LOOP(i) for(int i = 0; i < 3; i++)
#define NUM_GROUP_ELEMENTS 50000
#define MAX_TEMP_MATRICES 60000
#define MAX_TEMP_VECTORS 100
typedef struct {
double x[3];
@ -49,50 +47,8 @@ typedef struct {
// temporary; matrices can only be freed from the top, but that's enough for us
workspace_t *ws;
gsl_matrix **tmp;
int tmp_used;
gsl_vector **tmp_vec;
int tmp_vec_used;
} DrawingContext;
static gsl_matrix **getTempMatrices(DrawingContext *ctx, int n)
{
ERROR(ctx->tmp_used + n > MAX_TEMP_MATRICES, "Ran out of temporary matrices. Consider increasing MAX_TEMP_MATRICES\n");
int index = ctx->tmp_used;
ctx->tmp_used += n;
return ctx->tmp + index;
}
static gsl_matrix *getTempMatrix(DrawingContext *ctx)
{
return *getTempMatrices(ctx, 1);
}
static void releaseTempMatrices(DrawingContext *ctx, int n)
{
ERROR(ctx->tmp_used - n < 0, "Released more matrices then in use\n");
ctx->tmp_used -= n;
}
static gsl_vector **getTempVectors(DrawingContext *ctx, int n)
{
ERROR(ctx->tmp_vec_used + n > MAX_TEMP_VECTORS, "Ran out of temporary vectors. Consider increasing MAX_TEMP_VECTORS\n");
int index = ctx->tmp_vec_used;
ctx->tmp_vec_used += n;
return ctx->tmp_vec + index;
}
static gsl_vector *getTempVector(DrawingContext *ctx)
{
return *getTempVectors(ctx, 1);
}
static void releaseTempVectors(DrawingContext *ctx, int n)
{
ERROR(ctx->tmp_vec_used - n < 0, "Released more vectors then in use\n");
ctx->tmp_vec_used -= n;
}
// implemented in limit_set.c
void cartanMatrix(gsl_matrix *cartan, double a1, double a2, double a3, double s);
void initializeTriangleGenerators(gsl_matrix **gen, gsl_matrix *cartan);