improved precision of glReadPixels for colorbuffers < 24bpp
diff --git a/src/mesa/main/pixel.c b/src/mesa/main/pixel.c
index efd89ca..cb0818c 100644
--- a/src/mesa/main/pixel.c
+++ b/src/mesa/main/pixel.c
@@ -1,4 +1,4 @@
-/* $Id: pixel.c,v 1.20 2000/11/28 00:07:51 brianp Exp $ */
+/* $Id: pixel.c,v 1.21 2000/12/13 00:46:21 brianp Exp $ */
/*
* Mesa 3-D graphics library
@@ -1109,3 +1109,48 @@
stencil[i] = ctx->Pixel.MapStoS[ stencil[i] & mask ];
}
}
+
+
+
+/*
+ * This function converts an array of GLchan colors to GLfloat colors.
+ * Most importantly, it undoes the non-uniform quantization of pixel
+ * values introduced when we convert shallow (< 8 bit) pixel values
+ * to GLubytes in the ctx->Driver.ReadRGBASpan() functions.
+ * This fixes a number of OpenGL conformance failures when running on
+ * 16bpp displays, for example.
+ */
+void
+_mesa_chan_to_float_span(const GLcontext *ctx, GLuint n,
+ CONST GLchan rgba[][4], GLfloat rgbaf[][4])
+{
+ const GLuint rShift = CHAN_BITS - ctx->Visual.RedBits;
+ const GLuint gShift = CHAN_BITS - ctx->Visual.GreenBits;
+ const GLuint bShift = CHAN_BITS - ctx->Visual.BlueBits;
+ GLuint aShift;
+ const GLfloat rScale = 1.0 / (GLfloat) ((1 << ctx->Visual.RedBits ) - 1);
+ const GLfloat gScale = 1.0 / (GLfloat) ((1 << ctx->Visual.GreenBits) - 1);
+ const GLfloat bScale = 1.0 / (GLfloat) ((1 << ctx->Visual.BlueBits ) - 1);
+ GLfloat aScale;
+ GLuint i;
+
+ if (ctx->Visual.AlphaBits > 0) {
+ aShift = CHAN_BITS - ctx->Visual.AlphaBits;
+ aScale = 1.0 / (GLfloat) ((1 << ctx->Visual.AlphaBits) - 1);
+ }
+ else {
+ aShift = 0;
+ aScale = 1.0F / CHAN_MAXF;
+ }
+
+ for (i = 0; i < n; i++) {
+ const GLint r = rgba[i][RCOMP] >> rShift;
+ const GLint g = rgba[i][GCOMP] >> gShift;
+ const GLint b = rgba[i][BCOMP] >> bShift;
+ const GLint a = rgba[i][ACOMP] >> aShift;
+ rgbaf[i][RCOMP] = (GLfloat) r * rScale;
+ rgbaf[i][GCOMP] = (GLfloat) g * gScale;
+ rgbaf[i][BCOMP] = (GLfloat) b * bScale;
+ rgbaf[i][ACOMP] = (GLfloat) a * aScale;
+ }
+}