merge the (rest of) texmem branch
diff --git a/progs/demos/Makefile b/progs/demos/Makefile
index 43d0f17..feb2abd 100644
--- a/progs/demos/Makefile
+++ b/progs/demos/Makefile
@@ -47,6 +47,7 @@
renormal \
shadowtex \
singlebuffer \
+ streaming_rect \
spectex \
spriteblast \
stex3d \
diff --git a/progs/demos/streaming_rect.c b/progs/demos/streaming_rect.c
new file mode 100644
index 0000000..86e0080
--- /dev/null
+++ b/progs/demos/streaming_rect.c
@@ -0,0 +1,322 @@
+
+/*
+ * GL_ARB_multitexture demo
+ *
+ * Command line options:
+ * -info print GL implementation information
+ *
+ *
+ * Brian Paul November 1998 This program is in the public domain.
+ * Modified on 12 Feb 2002 for > 2 texture units.
+ */
+
+#define GL_GLEXT_PROTOTYPES
+
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <GL/glut.h>
+
+#include "readtex.h"
+
+
+#define ANIMATE 10
+#define PBO 11
+#define QUIT 100
+
+static GLboolean Animate = GL_TRUE;
+static GLboolean use_pbo = 1;
+static GLboolean whole_rect = 1;
+
+static GLfloat Drift = 0.0;
+static GLfloat drift_increment = 1/255.0;
+static GLfloat Xrot = 20.0, Yrot = 30.0;
+
+static GLuint Width = 1024;
+static GLuint Height = 512;
+
+
+static void Idle( void )
+{
+ if (Animate) {
+
+ Drift += drift_increment;
+ if (Drift >= 1.0)
+ Drift = 0.0;
+
+ glutPostRedisplay();
+ }
+}
+
+static int max( int a, int b ) { return a > b ? a : b; }
+static int min( int a, int b ) { return a < b ? a : b; }
+
+static void DrawObject()
+{
+ GLint size = Width * Height * 4;
+
+ if (use_pbo) {
+ /* XXX: This is extremely important - semantically makes the buffer
+ * contents undefined, but in practice means that the driver can
+ * release the old copy of the texture and allocate a new one
+ * without waiting for outstanding rendering to complete.
+ */
+ glBufferDataARB(GL_PIXEL_UNPACK_BUFFER_EXT, size, NULL, GL_STREAM_DRAW_ARB);
+
+ {
+ char *image = glMapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, GL_WRITE_ONLY_ARB);
+
+ printf("char %d\n", (unsigned char)(Drift * 255));
+
+ memset(image, size, (unsigned char)(Drift * 255));
+
+ glUnmapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT);
+ }
+
+
+ /* BGRA is required for most hardware paths:
+ */
+ glTexImage2D(GL_TEXTURE_RECTANGLE_ARB, 0, GL_RGBA, Width, Height, 0,
+ GL_BGRA, GL_UNSIGNED_BYTE, NULL);
+ }
+ else {
+ static char *image = NULL;
+
+ if (image == NULL)
+ image = malloc(size);
+
+ memset(image, size, (unsigned char)(Drift * 255));
+
+ /* BGRA should be the fast path for regular uploads as well.
+ */
+ glTexImage2D(GL_TEXTURE_RECTANGLE_ARB, 0, GL_RGBA, Width, Height, 0,
+ GL_BGRA, GL_UNSIGNED_BYTE, image);
+ }
+
+ {
+ int x,y,w,h;
+
+ if (whole_rect) {
+ x = y = 0;
+ w = Width;
+ h = Height;
+ }
+ else {
+ x = y = 0;
+ w = min(10, Width);
+ h = min(10, Height);
+ }
+
+ glBegin(GL_QUADS);
+
+ glTexCoord2f( x, y);
+ glVertex2f( x, y );
+
+ glTexCoord2f( x, y + h);
+ glVertex2f( x, y + h);
+
+ glTexCoord2f( x + w + .5, y + h);
+ glVertex2f( x + w, y + h );
+
+ glTexCoord2f( x + w, y + .5);
+ glVertex2f( x + w, y );
+
+ glEnd();
+ }
+}
+
+
+
+static void Display( void )
+{
+ static GLint T0 = 0;
+ static GLint Frames = 0;
+ GLint t;
+
+ glClear( GL_COLOR_BUFFER_BIT );
+
+ glPushMatrix();
+ DrawObject();
+ glPopMatrix();
+
+ glutSwapBuffers();
+
+ Frames++;
+
+ t = glutGet(GLUT_ELAPSED_TIME);
+ if (t - T0 >= 1000) {
+ GLfloat seconds = (t - T0) / 1000.0;
+
+ GLfloat fps = Frames / seconds;
+ printf("%d frames in %6.3f seconds = %6.3f FPS\n", Frames, seconds, fps);
+
+ drift_increment = 2.2 * seconds / Frames;
+ T0 = t;
+ Frames = 0;
+ }
+}
+
+
+static void Reshape( int width, int height )
+{
+ glViewport( 0, 0, width, height );
+ glMatrixMode( GL_PROJECTION );
+ glLoadIdentity();
+/* glFrustum( -1.0, 1.0, -1.0, 1.0, 10.0, 100.0 ); */
+ gluOrtho2D( 0, width, height, 0 );
+ glMatrixMode( GL_MODELVIEW );
+ glLoadIdentity();
+ glTranslatef(0.375, 0.375, 0);
+}
+
+
+static void ModeMenu(int entry)
+{
+ if (entry==ANIMATE) {
+ Animate = !Animate;
+ }
+ else if (entry==PBO) {
+ use_pbo = !use_pbo;
+ }
+ else if (entry==QUIT) {
+ exit(0);
+ }
+
+ glutPostRedisplay();
+}
+
+
+static void Key( unsigned char key, int x, int y )
+{
+ (void) x;
+ (void) y;
+ switch (key) {
+ case 27:
+ exit(0);
+ break;
+ }
+ glutPostRedisplay();
+}
+
+
+static void SpecialKey( int key, int x, int y )
+{
+ float step = 3.0;
+ (void) x;
+ (void) y;
+
+ switch (key) {
+ case GLUT_KEY_UP:
+ Xrot += step;
+ break;
+ case GLUT_KEY_DOWN:
+ Xrot -= step;
+ break;
+ case GLUT_KEY_LEFT:
+ Yrot += step;
+ break;
+ case GLUT_KEY_RIGHT:
+ Yrot -= step;
+ break;
+ }
+ glutPostRedisplay();
+}
+
+
+static void Init( int argc, char *argv[] )
+{
+ const char *exten = (const char *) glGetString(GL_EXTENSIONS);
+ GLuint texObj, DrawPBO;
+ GLint size;
+
+
+ if (!strstr(exten, "GL_ARB_multitexture")) {
+ printf("Sorry, GL_ARB_multitexture not supported by this renderer.\n");
+ exit(1);
+ }
+
+ glGetIntegerv(GL_MAX_TEXTURE_SIZE, &size);
+ printf("%d x %d max texture size\n", size, size);
+
+ glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+
+ /* allocate two texture objects */
+ glGenTextures(1, &texObj);
+
+ /* setup the texture objects */
+ glActiveTextureARB(GL_TEXTURE0_ARB);
+ glBindTexture(GL_TEXTURE_RECTANGLE_ARB, texObj);
+
+ glTexParameteri(GL_TEXTURE_RECTANGLE_ARB, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+ glTexParameteri(GL_TEXTURE_RECTANGLE_ARB, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+ glGenBuffersARB(1, &DrawPBO);
+
+ glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, DrawPBO);
+ glBufferDataARB(GL_PIXEL_UNPACK_BUFFER_EXT,
+ Width * Height * 4, NULL, GL_STREAM_DRAW);
+
+ glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
+
+ glEnable(GL_TEXTURE_RECTANGLE_ARB);
+
+ glShadeModel(GL_SMOOTH);
+ glClearColor(0.3, 0.3, 0.4, 1.0);
+
+ if (argc > 1 && strcmp(argv[1], "-info")==0) {
+ printf("GL_RENDERER = %s\n", (char *) glGetString(GL_RENDERER));
+ printf("GL_VERSION = %s\n", (char *) glGetString(GL_VERSION));
+ printf("GL_VENDOR = %s\n", (char *) glGetString(GL_VENDOR));
+ printf("GL_EXTENSIONS = %s\n", (char *) glGetString(GL_EXTENSIONS));
+ }
+}
+
+
+int main( int argc, char *argv[] )
+{
+ GLint i;
+
+ glutInit( &argc, argv );
+
+ for (i = 1; i < argc; i++) {
+ if (strcmp(argv[i], "-w") == 0) {
+ Width = atoi(argv[i+1]);
+ if (Width <= 0) {
+ printf("Error, bad width\n");
+ exit(1);
+ }
+ i++;
+ }
+ else if (strcmp(argv[i], "-h") == 0) {
+ Height = atoi(argv[i+1]);
+ if (Height <= 0) {
+ printf("Error, bad height\n");
+ exit(1);
+ }
+ i++;
+ }
+ }
+
+ glutInitWindowSize( Width, Height );
+ glutInitWindowPosition( 0, 0 );
+ glutInitDisplayMode( GLUT_RGB | GLUT_DOUBLE );
+ glutCreateWindow(argv[0] );
+
+ Init( argc, argv );
+
+ glutReshapeFunc( Reshape );
+ glutKeyboardFunc( Key );
+ glutSpecialFunc( SpecialKey );
+ glutDisplayFunc( Display );
+ glutIdleFunc( Idle );
+
+ glutCreateMenu(ModeMenu);
+ glutAddMenuEntry("Toggle Animation", ANIMATE);
+ glutAddMenuEntry("Toggle PBO", PBO);
+ glutAddMenuEntry("Quit", QUIT);
+ glutAttachMenu(GLUT_RIGHT_BUTTON);
+
+ glutMainLoop();
+ return 0;
+}
diff --git a/progs/demos/texdown.c b/progs/demos/texdown.c
index 79525a0..fc98fdd 100644
--- a/progs/demos/texdown.c
+++ b/progs/demos/texdown.c
@@ -38,8 +38,8 @@
#include <GL/glut.h>
-static GLsizei MaxSize = 1024;
-static GLsizei TexWidth = 256, TexHeight = 256, TexBorder = 0;
+static GLsizei MaxSize = 2048;
+static GLsizei TexWidth = 1024, TexHeight = 1024, TexBorder = 0;
static GLboolean ScaleAndBias = GL_FALSE;
static GLboolean SubImage = GL_FALSE;
static GLdouble DownloadRate = 0.0; /* texels/sec */
@@ -47,6 +47,32 @@
static GLuint Mode = 0;
+/* Try and avoid L2 cache effects by cycling through a small number of
+ * textures.
+ *
+ * At the initial size of 1024x1024x4 == 4mbyte, say 8 textures will
+ * keep us out of most caches at 32mb total.
+ *
+ * This turns into a fairly interesting question of what exactly you
+ * expect to be in cache in normal usage, and what you think should be
+ * outside. There's no rules for this, no reason to favour one usage
+ * over another except what the application you care about happens to
+ * resemble most closely.
+ *
+ * - Should the client texture image be in L2 cache? Has it just been
+ * generated or read from disk?
+ * - Does the application really use >1 texture, or is it constantly
+ * updating one image in-place?
+ *
+ * Different answers will favour different texture upload mechanisms.
+ * To upload an image that is purely outside of cache, a DMA-based
+ * upload will probably win, whereas for small, in-cache textures,
+ * copying looks good.
+ */
+#define NR_TEXOBJ 4
+static GLuint TexObj[NR_TEXOBJ];
+
+
struct FormatRec {
GLenum Format;
GLenum Type;
@@ -116,25 +142,57 @@
}
}
+/* On x86, there is a performance cliff for memcpy to texture memory
+ * for sources below 64 byte alignment. We do our best with this in
+ * the driver, but it is better if the images are correctly aligned to
+ * start with:
+ */
+#define ALIGN (1<<12)
+
+static unsigned align(unsigned value, unsigned a)
+{
+ return (value + a - 1) & ~(a-1);
+}
+
+static int MIN2(int a, int b)
+{
+ return a < b ? a : b;
+}
static void
MeasureDownloadRate(void)
{
const int w = TexWidth + 2 * TexBorder;
const int h = TexHeight + 2 * TexBorder;
- const int bytes = w * h * BytesPerTexel(Format);
+ const int image_bytes = align(w * h * BytesPerTexel(Format), ALIGN);
+ const int bytes = image_bytes * NR_TEXOBJ;
+ GLubyte *orig_texImage, *orig_getImage;
GLubyte *texImage, *getImage;
GLdouble t0, t1, time;
int count;
int i;
+ int offset = 0;
+ GLdouble total = 0; /* ints will tend to overflow */
- texImage = (GLubyte *) malloc(bytes);
- getImage = (GLubyte *) malloc(bytes);
- if (!texImage || !getImage) {
+ printf("allocating %d bytes for %d %dx%d images\n",
+ bytes, NR_TEXOBJ, w, h);
+
+ orig_texImage = (GLubyte *) malloc(bytes + ALIGN);
+ orig_getImage = (GLubyte *) malloc(image_bytes + ALIGN);
+ if (!orig_texImage || !orig_getImage) {
DownloadRate = 0.0;
return;
}
+ printf("alloc %p %p\n", orig_texImage, orig_getImage);
+
+ texImage = (GLubyte *)align((unsigned)orig_texImage, ALIGN);
+ getImage = (GLubyte *)align((unsigned)orig_getImage, ALIGN);
+
+ for (i = 1; !(((unsigned)texImage) & i); i<<=1)
+ ;
+ printf("texture image alignment: %d bytes (%p)\n", i, texImage);
+
for (i = 0; i < bytes; i++) {
texImage[i] = i & 0xff;
}
@@ -166,16 +224,50 @@
count = 0;
t0 = glutGet(GLUT_ELAPSED_TIME) * 0.001;
do {
+ int img = count%NR_TEXOBJ;
+ GLubyte *img_ptr = texImage + img * image_bytes;
+
+ glBindTexture(GL_TEXTURE_2D, TexObj[img]);
+
if (SubImage && count > 0) {
- glTexSubImage2D(GL_TEXTURE_2D, 0, -TexBorder, -TexBorder, w, h,
+ /* Only update a portion of the image each iteration. This
+ * is presumably why you'd want to use texsubimage, otherwise
+ * you may as well just call teximage again.
+ *
+ * A bigger question is whether to use a pointer that moves
+ * with each call, ie does the incoming data come from L2
+ * cache under normal circumstances, or is it pulled from
+ * uncached memory?
+ *
+ * There's a good argument to say L2 cache, ie you'd expect
+ * the data to have been recently generated. It's possible
+ * that it could have come from a file read, which may or may
+ * not have gone through the cpu.
+ */
+ glTexSubImage2D(GL_TEXTURE_2D, 0,
+ -TexBorder,
+ -TexBorder + offset * h/8,
+ w,
+ h/8,
FormatTable[Format].Format,
- FormatTable[Format].Type, texImage);
+ FormatTable[Format].Type,
+#if 1
+ texImage /* likely in L2$ */
+#else
+ img_ptr + offset * bytes/8 /* unlikely in L2$ */
+#endif
+ );
+ offset += 1;
+ offset %= 8;
+ total += w * h / 8;
}
else {
glTexImage2D(GL_TEXTURE_2D, 0,
FormatTable[Format].IntFormat, w, h, TexBorder,
FormatTable[Format].Format,
- FormatTable[Format].Type, texImage);
+ FormatTable[Format].Type,
+ img_ptr);
+ total += w*h;
}
/* draw a tiny polygon to force texture into texram */
@@ -192,25 +284,12 @@
glDisable(GL_TEXTURE_2D);
- printf("w*h=%d count=%d time=%f\n", w*h, count, time);
- DownloadRate = w * h * count / time;
+ printf("total texels=%f time=%f\n", total, time);
+ DownloadRate = total / time;
-#if 0
- if (!ScaleAndBias) {
- /* verify texture readback */
- glGetTexImage(GL_TEXTURE_2D, 0,
- FormatTable[Format].Format,
- FormatTable[Format].Type, getImage);
- for (i = 0; i < w * h; i++) {
- if (texImage[i] != getImage[i]) {
- printf("[%d] %d != %d\n", i, texImage[i], getImage[i]);
- }
- }
- }
-#endif
- free(texImage);
- free(getImage);
+ free(orig_texImage);
+ free(orig_getImage);
{
GLint err = glGetError();