From be66f80641138b980de0f1d3231576dd42d4c658 Mon Sep 17 00:00:00 2001
From: Steven Robertson <steven@strobe.cc>
Date: Wed, 4 May 2011 08:13:39 -0400
Subject: [PATCH] Final xforms

---
 cuburn/code/iter.py       | 19 ++++++++++++++-----
 cuburn/code/util.py       | 18 +++++++++++++-----
 cuburn/code/variations.py | 14 ++++++++++++--
 cuburn/render.py          | 10 ++++++++--
 4 files changed, 47 insertions(+), 14 deletions(-)

diff --git a/cuburn/code/iter.py b/cuburn/code/iter.py
index 4e4ff22..055fe77 100644
--- a/cuburn/code/iter.py
+++ b/cuburn/code/iter.py
@@ -14,8 +14,6 @@ from cuburn.code import mwc, variations, filter
 from cuburn.code.util import *
 from cuburn.render import Genome
 
-import tempita
-
 class IterCode(HunkOCode):
     def __init__(self, features):
         self.features = features
@@ -34,7 +32,7 @@ texture<uchar4, cudaTextureType2D, cudaReadModeNormalizedFloat> palTex;
         px = self.packer.view('info', 'xf%d_' % xfid)
         px.sub('xf', 'cp.xforms[%d]' % xfid)
 
-        tmpl = tempita.Template("""
+        tmpl = Template("""
 __device__
 void apply_xf{{xfid}}(float *ix, float *iy, float *icolor,
                       const iter_info *info, mwc_st *rctx) {
@@ -63,7 +61,7 @@ void apply_xf{{xfid}}(float *ix, float *iy, float *icolor,
         return tmpl.substitute(g)
 
     def _iterbody(self):
-        tmpl = tempita.Template("""
+        tmpl = Template("""
 __global__
 void iter(mwc_st *msts, iter_info *infos, float *accbuf, float *denbuf) {
     mwc_st rctx = msts[gtid()];
@@ -82,14 +80,20 @@ void iter(mwc_st *msts, iter_info *infos, float *accbuf, float *denbuf) {
         float xfsel = mwc_next_01(&rctx);
 
         {{for xfid, xform in enumerate(features.xforms)}}
+        {{if xfid != features.final_xform_index}}
         if (xfsel <= {{packer.get('cp.norm_density[%d]' % xfid)}}) {
             apply_xf{{xfid}}(&x, &y, &color, info, &rctx);
         } else
+        {{endif}}
         {{endfor}}
         {
             denbuf[0] = xfsel;
             break; // TODO: fail here
         }
+        {{if features.final_xform_index}}
+        float fx = x, fy = y, fcolor;
+        apply_xf{{features.final_xform_index}}(&fx, &fy, &fcolor, info, &rctx);
+        {{endif}}
 
         if (consec_bad < 0) {
             consec_bad++;
@@ -101,8 +105,13 @@ void iter(mwc_st *msts, iter_info *infos, float *accbuf, float *denbuf) {
         // TODO: this may not optimize well, verify.
 
         float cx, cy;
+        {{if features.final_xform_index}}
+        {{apply_affine('fx', 'fy', 'cx', 'cy', packer,
+                       'cp.camera_transform', 'cam')}}
+        {{else}}
         {{apply_affine('x', 'y', 'cx', 'cy', packer,
                        'cp.camera_transform', 'cam')}}
+        {{endif}}
 
         float ditherwidth = {{packer.get('0.5 * cp.spatial_filter_radius')}};
         float ditherx = mwc_next_11(&rctx) * ditherwidth;
@@ -225,7 +234,7 @@ class MemBench(HunkOCode):
 __shared__ uint32_t coord[512];
 """
 
-    defs_tmpl = tempita.Template("""
+    defs_tmpl = Template("""
 __global__
 void iter{{W}}(mwc_st *mwcs, uint32_t *buf) {
     mwc_st rctx = mwcs[gtid()];
diff --git a/cuburn/code/util.py b/cuburn/code/util.py
index 902ab40..0cfaf3a 100644
--- a/cuburn/code/util.py
+++ b/cuburn/code/util.py
@@ -5,6 +5,10 @@ Provides tools and miscellaneous functions for building device code.
 import numpy as np
 import tempita
 
+class Template(tempita.Template):
+    default_namespace = tempita.Template.default_namespace.copy()
+Template.default_namespace.update({'np': np})
+
 class HunkOCode(object):
     """An apparently passive container for device code."""
     # Use property objects to make these dynamic
@@ -13,11 +17,11 @@ class HunkOCode(object):
     defs = ''
 
 def assemble_code(*sections):
-    return '\n'.join(['\n'.join([getattr(sect, kind) for sect in sections])
-                      for kind in ['headers', 'decls', 'defs']])
+    return ''.join([''.join([getattr(sect, kind) for sect in sections])
+                    for kind in ['headers', 'decls', 'defs']])
 
 def apply_affine(x, y, xo, yo, packer, base_accessor, base_name):
-    return tempita.Template("""
+    return Template("""
     {{xo}} = {{packer.get(ba + '[0,0]', bn + '_xx')}} * {{x}}
            + {{packer.get(ba + '[0,1]', bn + '_xy')}} * {{y}}
            + {{packer.get(ba + '[0,2]', bn + '_xo')}};
@@ -112,6 +116,8 @@ class DataPacker(HunkOCode):
     future, but for now it's incredibly barebones.
     """
 
+    default_namespace = {'np': np}
+
     def __init__(self, tname, clsize=128):
         """
         Create a new DataPacker.
@@ -148,13 +154,15 @@ class DataPacker(HunkOCode):
         return (4 * len(self) + self.clsize - 1) / self.clsize * self.clsize
 
     def pack(self, **kwargs):
+        base_ns = self.default_namespace.copy()
+        base_ns.update(kwargs)
         out = np.zeros(self.align/4, dtype=np.float32)
         subbed_nses = {}
 
         for i, name in enumerate(self.packed_order):
             view, accessor = self.packed[name]
             if view not in subbed_nses:
-                subbed_nses[view] = view._apply_subs(dict(kwargs))
+                subbed_nses[view] = view._apply_subs(dict(base_ns))
             try:
                 val = eval(accessor, subbed_nses[view])
             except Exception, e:
@@ -165,7 +173,7 @@ class DataPacker(HunkOCode):
 
     @property
     def decls(self):
-        tmpl = tempita.Template("""
+        tmpl = Template("""
 typedef struct {
 
 {{for name, accessor in values}}
diff --git a/cuburn/code/variations.py b/cuburn/code/variations.py
index 30b82c1..898538c 100644
--- a/cuburn/code/variations.py
+++ b/cuburn/code/variations.py
@@ -65,7 +65,7 @@ var(8, 'disc', """
     float a = w * atan2f(tx, ty) * M_1_PI;
     float r = M_PI * sqrtf(tx*tx + ty*ty);
     ox += sinf(r) * a
-    oy += cosf(r) * a 
+    oy += cosf(r) * a
     """)
 
 var(9, 'spiral', """
@@ -208,7 +208,7 @@ var(25, 'fan2', """
         a -= dx2;
     else
         a += dx2;
-    
+
     ox += r * sinf(a);
     oy += r * cosf(a);
     """)
@@ -238,6 +238,16 @@ var(29, 'cylinder', """
     ox += w * sinf(tx);
     oy += w * ty;
     """)
+var(30, 'perspective', """
+    float pdist = {{px.get('xf.perspective_dist')}};
+    float pvsin = {{px.get('np.sin(xf.perspective_angle*np.pi/2)', 'pvsin')}};
+    float pvfcos = {{px.get(
+        'xf.perspective_dist*np.cos(xf.perspective_angle*np.pi/2)', 'pvfcos')}};
+
+    float t = 1.0 / (pdist - ty * pvsin);
+    ox += w * pdist * tx * t;
+    oy += w * pvfcos * ty * t;
+    """)
 
 var(33, 'juliascope', """
     float ang = atan2f(ty, tx);
diff --git a/cuburn/render.py b/cuburn/render.py
index 6ef86ac..9285b92 100644
--- a/cuburn/render.py
+++ b/cuburn/render.py
@@ -21,7 +21,8 @@ class Genome(pyflam3.Genome):
 
     def _init(self):
         self.xforms = [self.xform[i] for i in range(self.num_xforms)]
-        dens = np.array([x.density for x in self.xforms])
+        dens = np.array([x.density for i, x in enumerate(self.xforms)
+                         if i != self.final_xform_index])
         dens /= np.sum(dens)
         self.norm_density = [np.sum(dens[:i+1]) for i in range(len(dens))]
 
@@ -102,7 +103,12 @@ class Features(object):
         self.xforms = [XFormFeatures([cp.xforms[i] for cp in genomes], i)
                        for i in range(self.nxforms)]
         if any(lambda cp: cp.final_xform_enable):
-            raise NotImplementedError("Final xform")
+            if not reduce(lambda a, b: a == b,
+                          [cp.final_xform_index for cp in genomes]):
+                raise ValueError("Differing final xform indexes")
+            self.final_xform_index = genomes[0].final_xform_index
+        else:
+            self.final_xform_index = None
 
         self.width = genomes[0].width
         self.height = genomes[0].height