Skip to content

Commit ec58d0c

Browse files
authored
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).
2 parents cd0ea60 + 483bec7 commit ec58d0c

File tree

18 files changed

+1012
-862
lines changed

18 files changed

+1012
-862
lines changed

Diff for: src/etc/platform-intrinsics/aarch64.json

+2-2
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
{
2-
"platform": "aarch64",
3-
"intrinsic_prefix": "aarch64_v",
2+
"platform": "aarch64_v",
3+
"intrinsic_prefix": "",
44
"llvm_prefix": "llvm.aarch64.neon.",
55
"number_info": {
66
"signed": {

Diff for: src/etc/platform-intrinsics/arm.json

+2-2
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
{
2-
"platform": "arm",
3-
"intrinsic_prefix": "arm_v",
2+
"platform": "arm_v",
3+
"intrinsic_prefix": "",
44
"llvm_prefix": "llvm.neon.v",
55
"number_info": {
66
"signed": {

Diff for: src/etc/platform-intrinsics/generator.py

+39-8
Original file line numberDiff line numberDiff line change
@@ -26,10 +26,9 @@
2626
class PlatformInfo(object):
2727
def __init__(self, json):
2828
self._platform = json['platform']
29-
self._intrinsic_prefix = json['intrinsic_prefix']
3029

31-
def intrinsic_prefix(self):
32-
return self._intrinsic_prefix
30+
def platform_prefix(self):
31+
return self._platform
3332

3433
class IntrinsicSet(object):
3534
def __init__(self, platform, json):
@@ -38,6 +37,7 @@ def __init__(self, platform, json):
3837
self._intrinsics = json['intrinsics']
3938
self._widths = json['width_info']
4039
self._platform = platform
40+
self._intrinsic_prefix = json['intrinsic_prefix']
4141

4242
def intrinsics(self):
4343
for raw in self._intrinsics:
@@ -48,6 +48,9 @@ def intrinsics(self):
4848
def platform(self):
4949
return self._platform
5050

51+
def intrinsic_prefix(self):
52+
return self._intrinsic_prefix
53+
5154
def llvm_prefix(self):
5255
return self._llvm_prefix
5356

@@ -538,8 +541,14 @@ def intrinsic_suffix(self):
538541
*self._args,
539542
width = self._width)
540543

544+
def platform_prefix(self):
545+
return self._platform.platform().platform_prefix()
546+
547+
def intrinsic_set_name(self):
548+
return self._platform.intrinsic_prefix()
549+
541550
def intrinsic_name(self):
542-
return self._platform.platform().intrinsic_prefix() + self.intrinsic_suffix()
551+
return self._platform.intrinsic_prefix() + self.intrinsic_suffix()
543552

544553
def compiler_args(self):
545554
return ', '.join(arg.compiler_ctor_ref() for arg in self._args_raw)
@@ -561,6 +570,27 @@ def parse_args():
561570
formatter_class = argparse.RawDescriptionHelpFormatter,
562571
description = 'Render an intrinsic definition JSON to various formats.',
563572
epilog = textwrap.dedent('''\
573+
Quick How-To:
574+
575+
There are two operating modes: single file and multiple files.
576+
577+
For example, ARM is specified as a single file. To generate the
578+
compiler-definitions for ARM just pass the script the "arm.json" file:
579+
580+
python generator.py --format compiler-defs arm.json
581+
582+
The X86 architecture is specified as multiple files (for the different
583+
instruction sets that x86 supports). To generate the compiler
584+
definitions one needs to pass the script a "platform information file"
585+
(with the -i flag) next to the files of the different intruction sets.
586+
For example, to generate the X86 compiler-definitions for SSE4.2, just:
587+
588+
python generator.py --format compiler-defs -i x86/info.json sse42.json
589+
590+
And to generate the compiler-definitions for SSE4.1 and SSE4.2, just:
591+
592+
python generator.py --format compiler-defs -i x86/info.json sse41.json sse42.json
593+
564594
An intrinsic definition consists of a map with fields:
565595
- intrinsic: pattern for the name(s) of the vendor's C intrinsic(s)
566596
- llvm: pattern for the name(s) of the internal llvm intrinsic(s)
@@ -730,8 +760,9 @@ def open(self, platform):
730760
return 'extern "platform-intrinsic" {'
731761

732762
def render(self, mono):
733-
return ' fn {}{};'.format(mono.intrinsic_name(),
734-
mono.intrinsic_signature())
763+
return ' fn {}{}{};'.format(mono.platform_prefix(),
764+
mono.intrinsic_name(),
765+
mono.intrinsic_signature())
735766

736767
def close(self):
737768
return '}'
@@ -765,15 +796,15 @@ def open(self, platform):
765796
#[inline(never)]
766797
pub fn find(name: &str) -> Option<Intrinsic> {{
767798
if !name.starts_with("{0}") {{ return None }}
768-
Some(match &name["{0}".len()..] {{'''.format(platform.intrinsic_prefix())
799+
Some(match &name["{0}".len()..] {{'''.format(platform.platform_prefix())
769800

770801
def render(self, mono):
771802
return '''\
772803
"{}" => Intrinsic {{
773804
inputs: {{ static INPUTS: [&'static Type; {}] = [{}]; &INPUTS }},
774805
output: {},
775806
definition: Named("{}")
776-
}},'''.format(mono.intrinsic_suffix(),
807+
}},'''.format(mono.intrinsic_set_name() + mono.intrinsic_suffix(),
777808
len(mono._args_raw),
778809
mono.compiler_args(),
779810
mono.compiler_ret(),

Diff for: src/etc/platform-intrinsics/x86/avx.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.avx.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/avx2.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.avx2.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/bmi.json

+13
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,13 @@
1+
{
2+
"intrinsic_prefix": "_bmi",
3+
"llvm_prefix": "llvm.x86.bmi.",
4+
"intrinsics": [
5+
{
6+
"intrinsic": "_bextr_{0.bitwidth}",
7+
"width": ["0"],
8+
"llvm": "bextr.{0.bitwidth}",
9+
"ret": "S(32-64)u",
10+
"args": ["0", "0"]
11+
}
12+
]
13+
}

Diff for: src/etc/platform-intrinsics/x86/bmi2.json

+27
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,27 @@
1+
{
2+
"intrinsic_prefix": "_bmi2",
3+
"llvm_prefix": "llvm.x86.bmi.",
4+
"intrinsics": [
5+
{
6+
"intrinsic": "_bzhi_{0.bitwidth}",
7+
"width": ["0"],
8+
"llvm": "bzhi.{0.bitwidth}",
9+
"ret": "S(32-64)u",
10+
"args": ["0", "0"]
11+
},
12+
{
13+
"intrinsic": "_pdep_{0.bitwidth}",
14+
"width": ["0"],
15+
"llvm": "pdep.{0.bitwidth}",
16+
"ret": "S(32-64)u",
17+
"args": ["0", "0"]
18+
},
19+
{
20+
"intrinsic": "_pext_{0.bitwidth}",
21+
"width": ["0"],
22+
"llvm": "pext.{0.bitwidth}",
23+
"ret": "S(32-64)u",
24+
"args": ["0", "0"]
25+
}
26+
]
27+
}

Diff for: src/etc/platform-intrinsics/x86/fma.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.fma.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/info.json

+5-1
Original file line numberDiff line numberDiff line change
@@ -1,26 +1,30 @@
11
{
22
"platform": "x86",
3-
"intrinsic_prefix": "x86_mm",
43
"number_info": {
54
"signed": {
65
"kind": "s",
76
"kind_short": "",
87
"data_type": { "pattern": "epi{bitwidth}" },
8+
"bitwidth": { "pattern": "{bitwidth}" },
99
"data_type_short": { "8": "b", "16": "w", "32": "d", "64": "q" }
1010
},
1111
"unsigned": {
1212
"kind": "u",
1313
"kind_short": "u",
1414
"data_type": { "pattern": "epu{bitwidth}" },
15+
"bitwidth": { "pattern": "{bitwidth}" },
1516
"data_type_short": { "8": "b", "16": "w", "32": "d", "64": "q" }
1617
},
1718
"float": {
1819
"kind": "f",
1920
"data_type": { "32": "ps", "64": "pd" },
21+
"bitwidth": { "pattern": "{bitwidth}" },
2022
"data_type_short": { "32": "ps", "64": "pd" }
2123
}
2224
},
2325
"width_info": {
26+
"32": { "width_mm": "32", "width_suffix": "" },
27+
"64": { "width_mm": "64", "width_suffix": "" },
2428
"128": { "width_mm": "", "width_suffix": "" },
2529
"256": { "width_mm": "256", "width_suffix": ".256" },
2630
"512": { "width_mm": "512", "width_suffix": ".512" }

Diff for: src/etc/platform-intrinsics/x86/sse.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.sse.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/sse2.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.sse2.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/sse3.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.sse3.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/sse41.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.sse41.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/sse42.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.sse42.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/ssse3.json

+1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
{
2+
"intrinsic_prefix": "_mm",
23
"llvm_prefix": "llvm.x86.ssse3.",
34
"intrinsics": [
45
{

Diff for: src/etc/platform-intrinsics/x86/tbm.json

+13
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,13 @@
1+
{
2+
"intrinsic_prefix": "_tbm",
3+
"llvm_prefix": "llvm.x86.tbm.",
4+
"intrinsics": [
5+
{
6+
"intrinsic": "_bextri_u{0.bitwidth}",
7+
"width": ["0"],
8+
"llvm": "bextri.u{0.bitwidth}",
9+
"ret": "S(32-64)u",
10+
"args": ["0", "0"]
11+
}
12+
]
13+
}

Diff for: src/librustc_driver/target_features.rs

+3
Original file line numberDiff line numberDiff line change
@@ -30,12 +30,15 @@ const ARM_WHITELIST: &'static [&'static str] = &[
3030
const X86_WHITELIST: &'static [&'static str] = &[
3131
"avx\0",
3232
"avx2\0",
33+
"bmi\0",
34+
"bmi2\0",
3335
"sse\0",
3436
"sse2\0",
3537
"sse3\0",
3638
"sse4.1\0",
3739
"sse4.2\0",
3840
"ssse3\0",
41+
"tbm\0",
3942
];
4043

4144
/// Add `target_feature = "..."` cfgs for a variety of platform

0 commit comments

Comments
 (0)