summaryrefslogtreecommitdiff
path: root/Events/XDC2014/XDC2014AbbotNIR.mdwn
diff options
context:
space:
mode:
authormperes <mperes@web>2014-10-05 19:06:47 -0700
committerxorg <iki-xorg@freedesktop.org>2014-10-05 19:06:47 -0700
commit39f9bfee3de00023093b7f7ae7d0cd9c5c92747c (patch)
tree050df93e2bc2680530dddad1ec24f35514c26c84 /Events/XDC2014/XDC2014AbbotNIR.mdwn
parent52d009d1d8c2852e71481348f8aad6f1ab27ae14 (diff)
Diffstat (limited to 'Events/XDC2014/XDC2014AbbotNIR.mdwn')
-rw-r--r--Events/XDC2014/XDC2014AbbotNIR.mdwn3
1 files changed, 3 insertions, 0 deletions
diff --git a/Events/XDC2014/XDC2014AbbotNIR.mdwn b/Events/XDC2014/XDC2014AbbotNIR.mdwn
new file mode 100644
index 00000000..e6d90783
--- /dev/null
+++ b/Events/XDC2014/XDC2014AbbotNIR.mdwn
@@ -0,0 +1,3 @@
+# Connor Abbott - NIR, or moving beyond GLSL IR in Mesa
+
+As the Mesa compiler stack has evolved over the past several years, we've run into fundamental issues with the current Intermediate Representation (IR), GLSL IR, that have made more powerful optimizations difficult and made our memory usage unacceptable. While at Intel over the summer, I worked on a new graphics-specific IR, NIR (New IR), that incorporates the lessons learned from GLSL IR and will allow us to share many more high-level optimizations. The talk will cover the design decisions taken, future work/areas to be improved, and a brief discussion of differences to other IR's, especially LLVM.