Mercurial > hg > dmlib

changeset 239:07c4f1a7ddc6

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Make DMVector 16 bytes in size (4 floats), add SSE optimizations, move some functions to dmvecmat.h as static inline for speed, etc.
author Matti Hamalainen <ccr@tnsp.org>
date Tue, 09 Oct 2012 12:59:09 +0300
parents 9643c517967d
children 61b9cd67cd55
files dmvecmat.c dmvecmat.h vecmattest.c
diffstat 3 files changed, 154 insertions(+), 133 deletions(-) [+]
line wrap: on
line diff
--- a/dmvecmat.c	Tue Oct 09 06:06:08 2012 +0300
+++ b/dmvecmat.c	Tue Oct 09 12:59:09 2012 +0300
@@ -5,109 +5,6 @@
  * (C) Copyright 2011 Tecnic Software productions (TNSP)
  */
 #include "dmvecmat.h"
-#include <math.h>
-
-
-void dm_vector_copy(DMVector *vd, DMVector *vs)
-{
-    memcpy(vd, vs, sizeof(DMVector));
-}
-
-
-/* Basic vector operations
- */
-void dm_vector_add(DMVector *vr, const DMVector *v2)
-{
-    vr->x += v2->x;
-    vr->y += v2->y;
-    vr->z += v2->z;
-}
-
-
-void dm_vector_add_r(DMVector *vr, const DMVector *v1, const DMVector *v2)
-{
-    vr->x = v1->x + v2->x;
-    vr->y = v1->y + v2->y;
-    vr->z = v1->z + v2->z;
-}
-
-
-void dm_vector_sub(DMVector *vr, const DMVector *v2)
-{
-    vr->x -= v2->x;
-    vr->y -= v2->y;
-    vr->z -= v2->z;
-}
-
-
-void dm_vector_sub_r(DMVector *vr, const DMVector *v1, const DMVector *v2)
-{
-    vr->x = v1->x - v2->x;
-    vr->y = v1->y - v2->y;
-    vr->z = v1->z - v2->z;
-}
-
-
-/* Return vector length
- */
-DMFloat dm_vector_length(const DMVector *vs)
-{
-    return sqrt((vs->x * vs->x) + (vs->y * vs->y) + (vs->z * vs->z));
-}
-
-
-/* Normalize vector
- */
-void dm_vector_normalize(DMVector *vec)
-{
-    DMFloat l = dm_vector_length(vec);
-
-    if (l > 0.0f)
-    {
-        l = 1.0f / l;
-        vec->x *= l;
-        vec->y *= l;
-        vec->z *= l;
-    }
-}
-
-
-/* Scale given vector
- */
-void dm_vector_scale(DMVector * vec, const DMFloat k)
-{
-    vec->x *= k;
-    vec->y *= k;
-    vec->z *= k;
-}
-
-
-/* Returns dot-product of two given vectors
- */
-DMFloat dm_vector_dot(const DMVector *v1, const DMVector *v2)
-{
-    return (v1->x * v2->x) + (v1->y * v2->y) + (v1->z * v2->z);
-}
-
-
-/* Returns cross-product of two given vectors
- */
-void dm_vector_cross(DMVector *vr, const DMVector *v1, const DMVector *v2)
-{
-    vr->x = (v1->y * v2->z) - (v1->z * v2->y);
-    vr->y = (v1->z * v2->x) - (v1->x * v2->z);
-    vr->z = (v1->x * v2->y) - (v1->y * v2->x);
-}
-
-
-/* Multiply given vector with a matrix
- */
-void dm_vector_mul_by_mat(DMVector *vd, const DMVector *vs, const DMMatrix *mat)
-{
-    vd->x = (vs->x * mat->m[0][0]) + (vs->y * mat->m[1][0]) + (vs->z * mat->m[2][0]);
-    vd->y = (vs->x * mat->m[0][1]) + (vs->y * mat->m[1][1]) + (vs->z * mat->m[2][1]);
-    vd->z = (vs->x * mat->m[0][2]) + (vs->y * mat->m[1][2]) + (vs->z * mat->m[2][2]);
-}
 
 
 /* Multiply list of given vectors with given matrix.
@@ -205,13 +102,3 @@
     mat->m[2][1] = (sy * q) - j;
     mat->m[2][2] = cx * cy;
 }
-
-
-/* Make rotation matrix from given angles (radians)
- */
-void dm_matrix_rot_a(DMMatrix *mat, const DMFloat ax, const DMFloat ay, const DMFloat az)
-{
-    dm_matrix_rot(mat, sin(ax), sin(ay), sin(az), cos(ax), cos(ay), cos(az));
-}
-
-
--- a/dmvecmat.h	Tue Oct 09 06:06:08 2012 +0300
+++ b/dmvecmat.h	Tue Oct 09 12:59:09 2012 +0300
@@ -8,15 +8,17 @@
 #define DMVECMAT_H
 #include "dmlib.h"
 
+#include <math.h>
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#define DM_MATRIX_SIZE	(3)
+#define DM_MATRIX_SIZE	(4)
 
 typedef struct
 {
-    DMFloat x, y, z;
+    DMFloat x, y, z, W;
 } DMVector;
 
 typedef	struct
@@ -24,21 +26,6 @@
     DMFloat m[DM_MATRIX_SIZE][DM_MATRIX_SIZE];
 } DMMatrix;
 
-void    dm_vector_copy(DMVector *vd, DMVector *vs);
-
-void    dm_vector_add(DMVector *vr, const DMVector *v2);
-void    dm_vector_add_r(DMVector *vr, const DMVector *v1, const DMVector *v2);
-void    dm_vector_sub(DMVector *vr, const DMVector *v2);
-void    dm_vector_sub_r(DMVector *vr, const DMVector *v1, const DMVector *v2);
-
-DMFloat dm_vector_length(const DMVector *vs);
-void    dm_vector_normalize(DMVector *vs);
-void    dm_vector_scale(DMVector * vec, const DMFloat k);
-DMFloat dm_vector_dot(const DMVector *v1, const DMVector *v2);
-void    dm_vector_cross(DMVector *vr, const DMVector *v1, const DMVector *v2);
-
-
-void    dm_vector_mul_by_mat(DMVector *vd, const DMVector *vs, const DMMatrix *mat);
 void    dm_vector_mul_by_mat_n(DMVector *list, const int nlist, const DMMatrix *mat);
 
 void    dm_matrix_unit(DMMatrix *mat);
@@ -51,8 +38,155 @@
         const DMFloat sx, const DMFloat sy, const DMFloat sz,
         const DMFloat cx, const DMFloat cy, const DMFloat cz);
 
-void    dm_matrix_rot_a(DMMatrix *mat,
-        const DMFloat ax, const DMFloat ay, const DMFloat az);
+
+static inline void dm_vector_copy(DMVector *vd, DMVector *vs)
+{
+    memcpy(vd, vs, sizeof(DMVector));
+}
+
+
+/* Basic vector operations
+ */
+static inline void dm_vector_add(DMVector *vr, const DMVector *v2)
+{
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "addps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (*vr)
+        : "m" (*vr), "m" (*v2)
+        : "memory", "%xmm1", "%xmm2");
+#else
+    vr->x += v2->x;
+    vr->y += v2->y;
+    vr->z += v2->z;
+#endif
+}
+
+
+static inline void dm_vector_add_r(DMVector *vr, const DMVector *v1, const DMVector *v2)
+{
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "addps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (*vr)
+        : "m" (*v1), "m" (*v2)
+        : "memory", "%xmm1", "%xmm2");
+#else
+    vr->x = v1->x + v2->x;
+    vr->y = v1->y + v2->y;
+    vr->z = v1->z + v2->z;
+#endif
+}
+
+
+static inline void dm_vector_sub(DMVector *vr, const DMVector *v2)
+{
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "subps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (*vr)
+        : "m" (*vr), "m" (*v2)
+        : "memory", "%xmm1", "%xmm2");
+#else
+    vr->x -= v2->x;
+    vr->y -= v2->y;
+    vr->z -= v2->z;
+#endif
+}
+
+
+static inline void dm_vector_sub_r(DMVector *vr, const DMVector *v1, const DMVector *v2)
+{
+#ifdef DM_USE_SIMD
+    asm("movups      %2,     %%xmm1\n"
+        "movups      %1,     %%xmm2\n"
+        "subps       %%xmm2, %%xmm1\n"
+        "movups      %%xmm1, %0\n"
+        : "=m" (*vr)
+        : "m" (*v1), "m" (*v2)
+        : "memory", "%xmm1", "%xmm2");
+#else
+    vr->x = v1->x - v2->x;
+    vr->y = v1->y - v2->y;
+    vr->z = v1->z - v2->z;
+#endif
+}
+
+
+/* Returns dot-product of two given vectors
+ */
+static inline DMFloat dm_vector_dot(const DMVector *v1, const DMVector *v2)
+{
+    return (v1->x * v2->x) + (v1->y * v2->y) + (v1->z * v2->z);
+}
+
+
+/* Return vector length
+ */
+static inline DMFloat dm_vector_length(const DMVector *vs)
+{
+    return sqrt((vs->x * vs->x) + (vs->y * vs->y) + (vs->z * vs->z));
+}
+
+
+/* Normalize vector
+ */
+static inline void dm_vector_normalize(DMVector *vec)
+{
+    DMFloat l = dm_vector_length(vec);
+
+    if (l > 0.0f)
+    {
+        l = 1.0f / l;
+        vec->x *= l;
+        vec->y *= l;
+        vec->z *= l;
+    }
+}
+
+
+/* Scale given vector
+ */
+static inline void dm_vector_scale(DMVector * vec, const DMFloat k)
+{
+    vec->x *= k;
+    vec->y *= k;
+    vec->z *= k;
+}
+
+
+/* Returns cross-product of two given vectors
+ */
+static inline void dm_vector_cross(DMVector *vr, const DMVector *v1, const DMVector *v2)
+{
+    vr->x = (v1->y * v2->z) - (v1->z * v2->y);
+    vr->y = (v1->z * v2->x) - (v1->x * v2->z);
+    vr->z = (v1->x * v2->y) - (v1->y * v2->x);
+}
+
+
+/* Multiply given vector with a matrix
+ */
+static inline void dm_vector_mul_by_mat(DMVector *vd, const DMVector *vs, const DMMatrix *mat)
+{
+    vd->x = (vs->x * mat->m[0][0]) + (vs->y * mat->m[1][0]) + (vs->z * mat->m[2][0]);
+    vd->y = (vs->x * mat->m[0][1]) + (vs->y * mat->m[1][1]) + (vs->z * mat->m[2][1]);
+    vd->z = (vs->x * mat->m[0][2]) + (vs->y * mat->m[1][2]) + (vs->z * mat->m[2][2]);
+}
+
+
+/* Make rotation matrix from given angles (radians)
+ */
+static inline void dm_matrix_rot_a(DMMatrix *mat, const DMFloat ax, const DMFloat ay, const DMFloat az)
+{
+    dm_matrix_rot(mat, sin(ax), sin(ay), sin(az), cos(ax), cos(ay), cos(az));
+}
 
 
 #ifdef __cplusplus
--- a/vecmattest.c	Tue Oct 09 06:06:08 2012 +0300
+++ b/vecmattest.c	Tue Oct 09 12:59:09 2012 +0300
@@ -64,7 +64,7 @@
 
 int main(int argc, char *argv[])
 {
-    DMVector a = { -5, 1, 17 }, b = { 1, 2, 0.5 };
+    DMVector a = { -5, 1, 17, 0 }, b = { 1, 2, 0.5, 0 };
     DMMatrix m;
 
     (void) argc;