From e4aac6993f6ed6da7d28a56562953581b08dd42b Mon Sep 17 00:00:00 2001
From: Steven Robertson <steven@strobe.cc>
Date: Mon, 13 Sep 2010 00:20:15 -0400
Subject: [PATCH] A few touchups

---
 cuburn/device_code.py | 12 ++++++------
 cuburn/render.py      |  4 ++--
 main.py               |  2 +-
 3 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/cuburn/device_code.py b/cuburn/device_code.py
index 3028deb..4bf3e16 100644
--- a/cuburn/device_code.py
+++ b/cuburn/device_code.py
@@ -68,12 +68,12 @@ class IterThread(PTXEntryPoint):
         with block("Claim a CP"):
             std.set_is_first_thread(reg.pred('p_is_first'))
             op.atom.add.u32(cp_idx, addr(g_num_cps_started), 1, ifp=p_is_first)
-            op.st.shared.u32(addr(s_cp_idx), cp_idx, ifp=p_is_first)
-            op.st.shared.s32(addr(s_num_samples), 0)
+            op.st.volatile.shared.u32(addr(s_cp_idx), cp_idx, ifp=p_is_first)
+            op.st.volatile.shared.s32(addr(s_num_samples), 0)
 
         comment("Load the CP index in all threads")
         op.bar.sync(0)
-        op.ld.shared.u32(cp_idx, addr(s_cp_idx))
+        op.ld.volatile.shared.u32(cp_idx, addr(s_cp_idx))
 
         with block("Check to see if this CP is valid (if not, we're done)"):
             reg.u32('num_cps')
@@ -178,7 +178,7 @@ class IterThread(PTXEntryPoint):
             reg.s32('num_samples num_samples_needed')
             comment('Sync before making decision to prevent divergence')
             op.bar.sync(3)
-            op.ld.shared.s32(num_samples, addr(s_num_samples))
+            op.ld.volatile.shared.s32(num_samples, addr(s_num_samples))
             cp.get(cpA, num_samples_needed, 'cp.nsamples')
             op.setp.ge.s32(p_cp_done, num_samples, num_samples_needed)
             op.bra.uni(cp_loop_start, ifp=p_cp_done)
@@ -519,9 +519,9 @@ class ShufflePoints(PTXFragment):
                 op.mad.lo.u32(shuf_read, shuf_off, 4, shuf_read)
             for var in args:
                 op.bar.sync(bar)
-                op.st.shared.b32(addr(shuf_write), var)
+                op.st.volatile.shared.b32(addr(shuf_write), var)
                 op.bar.sync(bar)
-                op.ld.shared.b32(var, addr(shuf_read))
+                op.ld.volatile.shared.b32(var, addr(shuf_read))
 
 class MWCRNG(PTXFragment):
     shortname = "mwc"
diff --git a/cuburn/render.py b/cuburn/render.py
index 27cd00d..c25b4a9 100644
--- a/cuburn/render.py
+++ b/cuburn/render.py
@@ -63,8 +63,8 @@ class _Frame(pyflam3.Frame):
                     sizeof(BaseGenome))
         self.ngenomes = len(genomes)
 
-        # TODO: do this here?
-        self.pixel_aspect_ratio = float(genomes[0].height) / genomes[0].width
+        # TODO: allow user to override this
+        self.pixel_aspect_ratio = 1.0
 
     def interpolate(self, time, stagger=0, cp=None):
         cp = cp or BaseGenome()
diff --git a/main.py b/main.py
index 9c55a00..e908407 100644
--- a/main.py
+++ b/main.py
@@ -59,7 +59,7 @@ def main(args):
     print anim.features.hist_width
     print anim.features.hist_height
     print anim.features.hist_stride
-    window = pyglet.window.Window(800, 600)
+    window = pyglet.window.Window(1600, 900)
     image = pyglet.image.ImageData(anim.features.hist_width,
                                    anim.features.hist_height,
                                    'RGBA',