Auto merge of #34412 - gnzlbg:document_platform_intrinsics_generate, r=alexcrichton

Add x86 intrinsics for bit manipulation (BMI 1.0, BMI 2.0, and TBM).

This PR adds the LLVM x86 intrinsics for the bit manipulation instruction sets (BMI 1.0, BMI 2.0, and TBM).

The objective of this pull-request is to allow building a library that implements all the algorithms offered by those instruction sets, using compiler intrinsics for the targets that support them (by means of `target_feature`).

The target features added are:

- `bmi`: Bit Manipulation Instruction Set 1.0, available in Intel >= Haswell and AMD's >= Jaguar/Piledriver,
- `bmi2`: Bit Manipulation Instruction Set 2.0, available in Intel >= Haswell and AMD's >= Excavator,
- `tbm`: Trailing Bit Manipulation, available only in AMD's Piledriver (won't be available in newer CPUs).

The intrinsics added are:

- BMI 1.0:
  - `bextr`: Bit field extract (with register).
- BMI 2.0:
  - `bzhi`: Zero high bits starting with specified bit position.
  - `pdep`: Parallel bits deposit.
  - `pext`: Parallel bits extract.
- TBM:
 - `bextri`: Bit field extract (with immediate).

Author: bors

src/etc/platform-intrinsics/aarch64.json

@@ -1,6 +1,6 @@
 {
-    "platform": "aarch64",
-    "intrinsic_prefix": "aarch64_v",
+    "platform": "aarch64_v",
+    "intrinsic_prefix": "",
     "llvm_prefix": "llvm.aarch64.neon.",
     "number_info": {
         "signed": {

src/etc/platform-intrinsics/arm.json

@@ -1,6 +1,6 @@
 {
-    "platform": "arm",
-    "intrinsic_prefix": "arm_v",
+    "platform": "arm_v",
+    "intrinsic_prefix": "",
     "llvm_prefix": "llvm.neon.v",
     "number_info": {
         "signed": {

src/etc/platform-intrinsics/generator.py

@@ -26,10 +26,9 @@
 class PlatformInfo(object):
     def __init__(self, json):
         self._platform = json['platform']
-        self._intrinsic_prefix = json['intrinsic_prefix']
 
-    def intrinsic_prefix(self):
-        return self._intrinsic_prefix
+    def platform_prefix(self):
+        return self._platform
 
 class IntrinsicSet(object):
     def __init__(self, platform, json):
@@ -38,6 +37,7 @@ def __init__(self, platform, json):
         self._intrinsics = json['intrinsics']
         self._widths = json['width_info']
         self._platform = platform
+        self._intrinsic_prefix = json['intrinsic_prefix']
 
     def intrinsics(self):
         for raw in self._intrinsics:
@@ -48,6 +48,9 @@ def intrinsics(self):
     def platform(self):
         return self._platform
 
+    def intrinsic_prefix(self):
+        return self._intrinsic_prefix
+
     def llvm_prefix(self):
         return self._llvm_prefix
 
@@ -538,8 +541,14 @@ def intrinsic_suffix(self):
                                       *self._args,
                                       width = self._width)
 
+    def platform_prefix(self):
+        return self._platform.platform().platform_prefix()
+
+    def intrinsic_set_name(self):
+        return self._platform.intrinsic_prefix()
+
     def intrinsic_name(self):
-        return self._platform.platform().intrinsic_prefix() + self.intrinsic_suffix()
+        return self._platform.intrinsic_prefix() + self.intrinsic_suffix()
 
     def compiler_args(self):
         return ', '.join(arg.compiler_ctor_ref() for arg in self._args_raw)
@@ -561,6 +570,27 @@ def parse_args():
         formatter_class = argparse.RawDescriptionHelpFormatter,
         description = 'Render an intrinsic definition JSON to various formats.',
         epilog = textwrap.dedent('''\
+        Quick How-To:
+
+        There are two operating modes: single file and multiple files.
+
+        For example, ARM is specified as a single file. To generate the
+        compiler-definitions for ARM just pass the script the "arm.json" file:
+
+        python generator.py --format compiler-defs arm.json
+
+        The X86 architecture is specified as multiple files (for the different
+        instruction sets that x86 supports). To generate the compiler
+        definitions one needs to pass the script a "platform information file"
+        (with the -i flag) next to the files of the different intruction sets.
+        For example, to generate the X86 compiler-definitions for SSE4.2, just:
+
+        python generator.py --format compiler-defs -i x86/info.json sse42.json
+
+        And to generate the compiler-definitions for SSE4.1 and SSE4.2, just:
+
+        python generator.py --format compiler-defs -i x86/info.json sse41.json sse42.json
+
         An intrinsic definition consists of a map with fields:
         - intrinsic: pattern for the name(s) of the vendor's C intrinsic(s)
         - llvm: pattern for the name(s) of the internal llvm intrinsic(s)
@@ -730,8 +760,9 @@ def open(self, platform):
         return 'extern "platform-intrinsic" {'
 
     def render(self, mono):
-        return '    fn {}{};'.format(mono.intrinsic_name(),
-                                     mono.intrinsic_signature())
+        return '    fn {}{}{};'.format(mono.platform_prefix(),
+                                       mono.intrinsic_name(),
+                                       mono.intrinsic_signature())
 
     def close(self):
         return '}'
@@ -765,15 +796,15 @@ def open(self, platform):
 #[inline(never)]
 pub fn find(name: &str) -> Option<Intrinsic> {{
     if !name.starts_with("{0}") {{ return None }}
-    Some(match &name["{0}".len()..] {{'''.format(platform.intrinsic_prefix())
+    Some(match &name["{0}".len()..] {{'''.format(platform.platform_prefix())
 
     def render(self, mono):
         return '''\
         "{}" => Intrinsic {{
             inputs: {{ static INPUTS: [&'static Type; {}] = [{}]; &INPUTS }},
             output: {},
             definition: Named("{}")
-        }},'''.format(mono.intrinsic_suffix(),
+        }},'''.format(mono.intrinsic_set_name() + mono.intrinsic_suffix(),
                       len(mono._args_raw),
                       mono.compiler_args(),
                       mono.compiler_ret(),

src/etc/platform-intrinsics/x86/avx.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.avx.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/avx2.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.avx2.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/bmi.json

@@ -0,0 +1,13 @@
+{
+    "intrinsic_prefix": "_bmi",
+    "llvm_prefix": "llvm.x86.bmi.",
+    "intrinsics": [
+        {
+            "intrinsic": "_bextr_{0.bitwidth}",
+            "width": ["0"],
+            "llvm": "bextr.{0.bitwidth}",
+            "ret": "S(32-64)u",
+            "args": ["0", "0"]
+        }
+    ]
+}

src/etc/platform-intrinsics/x86/bmi2.json

@@ -0,0 +1,27 @@
+{
+    "intrinsic_prefix": "_bmi2",
+    "llvm_prefix": "llvm.x86.bmi.",
+    "intrinsics": [
+        {
+            "intrinsic": "_bzhi_{0.bitwidth}",
+            "width": ["0"],
+            "llvm": "bzhi.{0.bitwidth}",
+            "ret": "S(32-64)u",
+            "args": ["0", "0"]
+        },
+        {
+            "intrinsic": "_pdep_{0.bitwidth}",
+            "width": ["0"],
+            "llvm": "pdep.{0.bitwidth}",
+            "ret": "S(32-64)u",
+            "args": ["0", "0"]
+        },
+        {
+            "intrinsic": "_pext_{0.bitwidth}",
+            "width": ["0"],
+            "llvm": "pext.{0.bitwidth}",
+            "ret": "S(32-64)u",
+            "args": ["0", "0"]
+        }
+    ]
+}

src/etc/platform-intrinsics/x86/fma.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.fma.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/info.json

@@ -1,26 +1,30 @@
 {
     "platform": "x86",
-    "intrinsic_prefix": "x86_mm",
     "number_info": {
         "signed": {
             "kind": "s",
             "kind_short": "",
             "data_type": { "pattern": "epi{bitwidth}" },
+            "bitwidth": { "pattern": "{bitwidth}" },
             "data_type_short": { "8": "b", "16": "w", "32": "d", "64": "q" }
         },
         "unsigned": {
             "kind": "u",
             "kind_short": "u",
             "data_type": { "pattern": "epu{bitwidth}" },
+            "bitwidth": { "pattern": "{bitwidth}" },
             "data_type_short": { "8": "b", "16": "w", "32": "d", "64": "q" }
         },
         "float": {
             "kind": "f",
             "data_type": { "32": "ps", "64": "pd" },
+            "bitwidth": { "pattern": "{bitwidth}" },
             "data_type_short": { "32": "ps", "64": "pd" }
         }
     },
     "width_info": {
+        "32":  { "width_mm": "32", "width_suffix": "" },
+        "64":  { "width_mm": "64", "width_suffix": "" },
         "128": { "width_mm": "", "width_suffix": "" },
         "256": { "width_mm": "256", "width_suffix": ".256" },
         "512": { "width_mm": "512", "width_suffix": ".512" }

src/etc/platform-intrinsics/x86/sse.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.sse.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/sse2.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.sse2.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/sse3.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.sse3.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/sse41.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.sse41.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/sse42.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.sse42.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/ssse3.json

@@ -1,4 +1,5 @@
 {
+    "intrinsic_prefix": "_mm",
     "llvm_prefix": "llvm.x86.ssse3.",
     "intrinsics": [
         {

src/etc/platform-intrinsics/x86/tbm.json

@@ -0,0 +1,13 @@
+{
+    "intrinsic_prefix": "_tbm",
+    "llvm_prefix": "llvm.x86.tbm.",
+    "intrinsics": [
+        {
+            "intrinsic": "_bextri_u{0.bitwidth}",
+            "width": ["0"],
+            "llvm": "bextri.u{0.bitwidth}",
+            "ret": "S(32-64)u",
+            "args": ["0", "0"]
+        }
+    ]
+}

src/librustc_driver/target_features.rs

@@ -30,12 +30,15 @@ const ARM_WHITELIST: &'static [&'static str] = &[
 const X86_WHITELIST: &'static [&'static str] = &[
     "avx\0",
     "avx2\0",
+    "bmi\0",
+    "bmi2\0",
     "sse\0",
     "sse2\0",
     "sse3\0",
     "sse4.1\0",
     "sse4.2\0",
     "ssse3\0",
+    "tbm\0",
 ];
 
 /// Add `target_feature = "..."` cfgs for a variety of platform