Support different scalar integer widths in Rust v. LLVM.

Some x86 C intrinsics are declared to take `int ...` (i.e. exposed in
Rust as `i32`), but LLVM implements them by taking `i8` instead.

5 files changed, 70 insertions, 28 deletions

diff --git a/src/etc/platform-intrinsics/generator.py b/src/etc/platform-intrinsics/generator.py
index 7f5a737e33e..97b2f57010b 100644
--- a/src/etc/platform-intrinsics/generator.py
+++ b/src/etc/platform-intrinsics/generator.py
@@ -16,7 +16,8 @@ import re
 import textwrap
 
 SPEC = re.compile(
-    r'^(?:(?P<id>[iusfIUSF])(?:\((?P<start>\d*)-(?P<end>\d*)\)|(?P<width>\d*))' +
+    r'^(?:(?P<id>[iusfIUSF])(?:\((?P<start>\d+)-(?P<end>\d+)\)|'
+    r'(?P<width>\d+)(:?/(?P<llvm_width>\d+))?)'
     r'|(?P<reference>\d+)(?P<modifiers>[vShdnwus]*)(?P<force_width>x\d+)?)$'
 )
 
@@ -111,27 +112,39 @@ class Number(Type):
         return platform_info.number_type_info(self)
 
 class Signed(Number):
-    def __init__(self, bitwidth):
+    def __init__(self, bitwidth, llvm_bitwidth = None):
         Number.__init__(self, bitwidth)
+        self._llvm_bitwidth = llvm_bitwidth
 
     def compiler_ctor(self):
-        return 'i({})'.format(self.bitwidth())
+        if self._llvm_bitwidth is None:
+            return 'i({})'.format(self.bitwidth())
+        else:
+            return 'i_({}, {})'.format(self.bitwidth(),
+                                       self._llvm_bitwidth)
 
     def llvm_name(self):
-        return 'i{}'.format(self.bitwidth())
+        bw = self._llvm_bitwidth or self.bitwidth()
+        return 'i{}'.format(bw)
 
     def rust_name(self):
         return 'i{}'.format(self.bitwidth())
 
 class Unsigned(Number):
-    def __init__(self, bitwidth):
+    def __init__(self, bitwidth, llvm_bitwidth = None):
         Number.__init__(self, bitwidth)
+        self._llvm_bitwidth = llvm_bitwidth
 
     def compiler_ctor(self):
-        return 'u({})'.format(self.bitwidth())
+        if self._llvm_bitwidth is None:
+            return 'u({})'.format(self.bitwidth())
+        else:
+            return 'u_({}, {})'.format(self.bitwidth(),
+                                       self._llvm_bitwidth)
 
     def llvm_name(self):
-        return 'i{}'.format(self.bitwidth())
+        bw = self._llvm_bitwidth or self.bitwidth()
+        return 'i{}'.format(bw)
 
     def rust_name(self):
         return 'u{}'.format(self.bitwidth())
@@ -220,18 +233,28 @@ class TypeSpec(object):
                 id = match.group('id')
                 is_vector = id.islower()
                 type_ctors = TYPE_ID_LOOKUP[id.lower()]
+
                 start = match.group('start')
                 if start is not None:
                     end = match.group('end')
+                    llvm_width = None
                 else:
                     start = end = match.group('width')
+                    llvm_width = match.group('llvm_width')
                 start = int(start)
                 end = int(end)
 
                 bitwidth = start
                 while bitwidth <= end:
                     for ctor in type_ctors:
-                        scalar = ctor(bitwidth)
+                        if llvm_width is not None:
+                            assert not is_vector
+                            llvm_width = int(llvm_width)
+                            assert llvm_width < bitwidth
+                            scalar = ctor(bitwidth, llvm_width)
+                        else:
+                            scalar = ctor(bitwidth)
+
                         if is_vector:
                             yield Vector(scalar, width // bitwidth)
                         else:
@@ -351,8 +374,9 @@ def parse_args():
         vector := vector_elem width |
         vector_elem := 'i' | 'u' | 's' | 'f'
 
-        scalar := scalar_type number
+        scalar := scalar_type number llvm_width?
         scalar_type := 'U' | 'S' | 'F'
+        llvm_width := '/' number
 
         aggregate := '(' (type),* ')' 'f'?
 
@@ -387,7 +411,11 @@ def parse_args():
         ## Scalars
 
         Similar to vectors, but these describe a single concrete type,
-        not a range. The number is the bitwidth.
+        not a range. The number is the bitwidth. The optional
+        `llvm_width` is the bitwidth of the integer that should be
+        passed to LLVM (by truncating the Rust argument): this only
+        works with scalar integers and the LLVM width must be smaller
+        than the Rust width.
 
         ### Types
 
@@ -474,7 +502,7 @@ class CompilerDefs(object):
 
 #![allow(unused_imports)]
 
-use {{Intrinsic, i, u, f, v, agg}};
+use {{Intrinsic, i, i_, u, u_, f, v, agg}};
 use IntrinsicDef::Named;
 use rustc::middle::ty;
 
diff --git a/src/etc/platform-intrinsics/x86/avx.json b/src/etc/platform-intrinsics/x86/avx.json
index 7d6a07f7550..4ac82fb90e9 100644
--- a/src/etc/platform-intrinsics/x86/avx.json
+++ b/src/etc/platform-intrinsics/x86/avx.json
@@ -13,7 +13,7 @@
             "width": [256],
             "llvm": "dp.ps.256",
             "ret": "f32",
-            "args": ["0", "0", "S32"]
+            "args": ["0", "0", "S32/8"]
         },
         {
             "intrinsic": "256_hadd_{0.data_type}",
diff --git a/src/etc/platform-intrinsics/x86/avx2.json b/src/etc/platform-intrinsics/x86/avx2.json
index 8b0e53e48cf..bd260ec02e9 100644
--- a/src/etc/platform-intrinsics/x86/avx2.json
+++ b/src/etc/platform-intrinsics/x86/avx2.json
@@ -79,6 +79,20 @@
             "args": ["0", "0"]
         },
         {
+            "intrinsic": "256_movemask_epi8",
+            "width": [256],
+            "llvm": "pmovmskb",
+            "ret": "S32",
+            "args": ["s8"]
+        },
+        {
+            "intrinsic": "256_mpsadbw_epu8",
+            "width": [256],
+            "llvm": "mpsadbw",
+            "ret": "u16",
+            "args": ["u8", "u8", "S32/8"]
+        },
+        {
             "intrinsic": "256_mul_{0.data_type}",
             "width": [256],
             "llvm": "pmul{0.data_type_short}.dq",
diff --git a/src/etc/platform-intrinsics/x86/sse41.json b/src/etc/platform-intrinsics/x86/sse41.json
index 8610dc83bd6..e835320e5cd 100644
--- a/src/etc/platform-intrinsics/x86/sse41.json
+++ b/src/etc/platform-intrinsics/x86/sse41.json
@@ -6,7 +6,7 @@
             "width": [128],
             "llvm": "dp{0.data_type}",
             "ret": "f(32-64)",
-            "args": ["0", "0", "S32"]
+            "args": ["0", "0", "S32/8"]
         },
         {
             "intrinsic": "_max_{0.data_type}",
@@ -34,7 +34,7 @@
             "width": [128],
             "llvm": "mpsadbw",
             "ret": "u16",
-            "args": ["u8", "u8", "S32"]
+            "args": ["u8", "u8", "S32/8"]
         },
         {
             "intrinsic": "_mul_epi32",
diff --git a/src/etc/platform-intrinsics/x86/sse42.json b/src/etc/platform-intrinsics/x86/sse42.json
index e63182d4857..c43ffef0dc5 100644
--- a/src/etc/platform-intrinsics/x86/sse42.json
+++ b/src/etc/platform-intrinsics/x86/sse42.json
@@ -6,98 +6,98 @@
             "width": [128],
             "llvm": "pcmpestria128",
             "ret": "S32",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpestrc",
             "width": [128],
             "llvm": "pcmpestric128",
             "ret": "S32",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpestri",
             "width": [128],
             "llvm": "pcmpestri128",
             "ret": "S32",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpestrm",
             "width": [128],
             "llvm": "pcmpestrm128",
             "ret": "s8",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpestro",
             "width": [128],
             "llvm": "pcmpestrio128",
             "ret": "S32",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpestrs",
             "width": [128],
             "llvm": "pcmpestris128",
             "ret": "S32",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpestrz",
             "width": [128],
             "llvm": "pcmpestriz128",
             "ret": "S32",
-            "args": ["s8", "S32", "s8", "S32", "S32"]
+            "args": ["s8", "S32", "s8", "S32", "S32/8"]
         },
         {
             "intrinsic": "_cmpistra",
             "width": [128],
             "llvm": "pcmpistria128",
             "ret": "S32",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         },
         {
             "intrinsic": "_cmpistrc",
             "width": [128],
             "llvm": "pcmpistric128",
             "ret": "S32",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         },
         {
             "intrinsic": "_cmpistri",
             "width": [128],
             "llvm": "pcmpistri128",
             "ret": "S32",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         },
         {
             "intrinsic": "_cmpistrm",
             "width": [128],
             "llvm": "pcmpistrm128",
             "ret": "s8",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         },
         {
             "intrinsic": "_cmpistro",
             "width": [128],
             "llvm": "pcmpistrio128",
             "ret": "S32",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         },
         {
             "intrinsic": "_cmpistrs",
             "width": [128],
             "llvm": "pcmpistris128",
             "ret": "S32",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         },
         {
             "intrinsic": "_cmpistrz",
             "width": [128],
             "llvm": "pcmpistriz128",
             "ret": "S32",
-            "args": ["s8", "s8", "S32"]
+            "args": ["s8", "s8", "S32/8"]
         }
     ]
 }