Add Om1 lowering with no optimizations

crosstest/crosstest.py

Index: crosstest/crosstest.py
diff --git a/crosstest/crosstest.py b/crosstest/crosstest.py
new file mode 100755
index 0000000000000000000000000000000000000000..ab241cf5c94f5f7a2e2e3e6988f0d1e033d01703
--- /dev/null
+++ b/crosstest/crosstest.py
@@ -0,0 +1,66 @@
+#!/usr/bin/env python2
+
+import argparse
+import os, sys
+import subprocess
+import tempfile
+
+sys.path.insert(0, '../pydir')
+from utils import shellcmd
+
+if __name__ == '__main__':
+    argparser = argparse.ArgumentParser()
+    argparser.add_argument('--test', required=True, action='append',
+                           help='List of C/C++/.ll files with test functions')
+    argparser.add_argument('--driver', required=True,
+                           help='Driver program')
+    argparser.add_argument('--target', required=False, default='X8632',

[JF, 2014/05/01 00:16:55]

X8632 isn't the naming you followed in the previous CL?

[Jim Stichnoth, 2014/05/05 07:03:55]

Looks like I was inconsistent in the previous CL.

I changed llvm2ice.cpp to allow strings x8632, x86-32, x86_32, x8664, x86-64,
x86_64, arm, arm32, and arm64, to meet the best-in-breed standards of the rest
of the PNaCl toolchain.

(But I limited this script to the choices x8632, x8664, and arm.)

+                           help='Translation target string')
+    argparser.add_argument('-O', required=False, default='2', dest='optlevel',
+                           choices=['m1', '-1', '0', '1', '2'],

[JF, 2014/05/01 00:16:55]

Are m1 and -1 the same?

[Jim Stichnoth, 2014/05/05 07:03:55]

Yes.  Added to the help string.

+                           help='Optimization level');
+    argparser.add_argument('--prefix', required=True,
+                           help='String prepended to Subzero symbol names')
+    argparser.add_argument('--output', '-o', required=True,
+                           help='Executable to produce')
+    argparser.add_argument('--dir', required=False, default='.',
+                           help='Output directory for all files')
+    args = argparser.parse_args()
+
+    objs = []
+    for arg in args.test:
+        base, ext = os.path.splitext(arg)
+        if ext == '.ll':
+            bitcode = arg
+        else:
+            bitcode = os.path.join(args.dir, base + '.pnacl.ll')
+            shellcmd(['../pydir/build-pnacl-ir.py', '--dir', args.dir, arg])
+            shellcmd('sed -i "s/^define internal /define /" ' + bitcode)
+            shellcmd('sed -i "s/le32-unknown-nacl/i686-pc-linux-gnu/" ' + bitcode)

[JF, 2014/05/01 00:16:55]

sed?

[Jim Stichnoth, 2014/05/05 07:03:55]

okok, I redid this with python regex.

+
+        asm_sz = os.path.join(args.dir, base + '.sz.s')
+        obj_sz = os.path.join(args.dir, base + '.sz.o')
+        obj_llc = os.path.join(args.dir, base + '.llc.o')
+        shellcmd(['../llvm2ice',
+                  '-O' + args.optlevel,
+                  '--target=' + args.target,
+                  '--prefix=' + args.prefix,
+                  '-o=' + asm_sz,
+                  bitcode])
+        shellcmd(['$LLVM_BIN_PATH/llvm-mc',

[JF, 2014/05/01 00:16:55]

Shouldn't LLVM_BIN_PATH be an argument instead?

[Jim Stichnoth, 2014/05/05 07:03:55]

Done. Added --llvm-bin-path arg, and changed runtests.sh to use it.

+                  '-arch=x86',

[JF, 2014/05/01 00:16:55]

x86 always?

[Jim Stichnoth, 2014/05/05 07:03:55]

Added the arch_map dictionary to map Subzero --target strings to llvm-mc --arch
strings.

+                  '-x86-asm-syntax=intel',
+                  '-filetype=obj',
+                  '-o=' + obj_sz,
+                  asm_sz])
+        shellcmd(['$LLVM_BIN_PATH/llc',
+                  '-filetype=obj',
+                  '-o=' + obj_llc,
+                  bitcode])
+        objs.append(obj_sz)
+        objs.append(bitcode)
+        #objs.append(obj_llc)

[JF, 2014/05/01 00:16:55]

Commented out.

[Jim Stichnoth, 2014/05/05 07:03:55]

This was left in intentionally, but I cleaned it up and documented it.  For the
interested/lazy:

# Each original bitcode file needs to be translated by the
# LLVM toolchain and have its object file linked in.  There
# are two ways to do this: explicitly use llc, or include the
# .ll file in the link command.  It turns out that these two
# approaches can produce different semantics on some undefined
# bitcode behavior.  Specifically, LLVM produces different
# results for overflowing fptoui instructions for i32 and i64
# on x86-32.  As it turns out, Subzero lowering was based on
# inspecting the object code produced by the direct llc
# command, so we need to directly run llc on the bitcode, even
# though it makes this script longer, to avoid spurious
# failures.  This behavior can be inspected by switching
# use_llc between True and False.

+
+    shellcmd(['$LLVM_BIN_PATH/clang', '-g', '-m32', args.driver] +
+             objs +
+             ['-lm', '-o', os.path.join(args.dir, args.output)])

[JF, 2014/05/01 00:16:55]

Doesn't this sometimes need to be clang++ to find the C++ libraries?

[Jim Stichnoth, 2014/05/05 07:03:55]

You are right.  'clang' doesn't work as the linker if the driver uses e.g.
std::cout.  'clang++' seems to compile the driver as C++, which messes up the
external linkage.

It's now changed to select between clang and clang++ based on the driver's
extension.