Add a quick pass to optimize sign / zero extension instructions. For targets where...
[oota-llvm.git] / lib / Target / CellSPU / README.txt
index 1d90f2a77c3da98a5abdb8c3785b7cccc34fe12c..4783dd5d24eb22fd150b5a09bfacdfd59978233b 100644 (file)
@@ -8,6 +8,7 @@ Department in The Aerospace Corporation:
 - Mark Thomas (floating point instructions)
 - Michael AuYeung (intrinsics)
 - Chandler Carruth (LLVM expertise)
+- Nehal Desai (debugging, i32 operations, RoadRunner SPU expertise)
 
 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
@@ -18,7 +19,7 @@ OUT OF OR IN CONNECTION WITH THE USE OF THE SOFTWARE INCLUDING, WITHOUT
 LIMITATION, DAMAGES RESULTING FROM LOST OR CONTAMINATED DATA, LOST PROFITS OR
 REVENUE, COMPUTER MALFUNCTION, OR FOR ANY SPECIAL, INCIDENTAL, CONSEQUENTIAL,
 OR PUNITIVE  DAMAGES, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES OR
-SUCH DAMAGES ARE FORESEEABLE. 
+SUCH DAMAGES ARE FORESEEABLE.
 
 ---------------------------------------------------------------------------
 --WARNING--:
@@ -34,17 +35,56 @@ to add 'spu' to configure's --enable-targets option, e.g.:
 ---------------------------------------------------------------------------
 
 TODO:
-* Finish branch instructions, branch prediction
+* Create a machine pass for performing dual-pipeline scheduling specifically
+  for CellSPU, and insert branch prediction instructions as needed.
 
-  These instructions were started, but only insofar as to get llvm-gcc-4.2's
-  crtbegin.ll working (which doesn't.)
+* i32 instructions:
 
-* Double floating point support
+  * i32 division (work-in-progress)
 
-  This was started. "What's missing?" to be filled in.
+* i64 support (see i64operations.c test harness):
+
+  * shifts and comparison operators: done
+  * sign and zero extension: done
+  * addition: done
+  * subtraction: needed
+  * multiplication: done
+
+* i128 support:
+
+  * zero extension, any extension: done
+  * sign extension: needed
+  * arithmetic operators (add, sub, mul, div): needed
+  * logical operations (and, or, shl, srl, sra, xor, nor, nand): needed
+
+    * or: done
+
+* f64 support
+
+  * Comparison operators:
+    SETOEQ              unimplemented
+    SETOGT              unimplemented
+    SETOGE              unimplemented
+    SETOLT              unimplemented
+    SETOLE              unimplemented
+    SETONE              unimplemented
+    SETO                done (lowered)
+    SETUO               done (lowered)
+    SETUEQ              unimplemented
+    SETUGT              unimplemented
+    SETUGE              unimplemented
+    SETULT              unimplemented
+    SETULE              unimplemented
+    SETUNE              unimplemented
+
+* LLVM vector suport
+
+  * VSETCC needs to be implemented. It's pretty straightforward to code, but
+    needs implementation.
 
 * Intrinsics
 
-  Lots of progress. "What's missing/incomplete?" to be filled in.
+  * spu.h instrinsics added but not tested. Need to have an operational
+    llvm-spu-gcc in order to write a unit test harness.
 
 ===-------------------------------------------------------------------------===