Merge pull request rust-lang#14 from japaric/platform-intrinsics

NVPTX: platform intrinsics + tests
japaric · Jul 12, 2016 · 5e7c419 · 5e7c419
2 parents 1bf24cd + e8149cc
commit 5e7c419
Show file tree

Hide file tree

Showing 13 changed files with 316 additions and 53 deletions.
diff --git a/src/etc/platform-intrinsics/nvptx/cuda.json b/src/etc/platform-intrinsics/nvptx/cuda.json
@@ -0,0 +1,13 @@
+{
+    "intrinsic_prefix": "_",
+    "llvm_prefix": "llvm.cuda.",
+    "intrinsics": [
+        {
+            "intrinsic": "syncthreads",
+            "width": ["0"],
+            "llvm": "syncthreads",
+            "ret": "V",
+            "args": []
+        }
+    ]
+}
diff --git a/src/etc/platform-intrinsics/nvptx/info.json b/src/etc/platform-intrinsics/nvptx/info.json
@@ -0,0 +1,7 @@
+{
+  "platform": "nvptx",
+  "number_info": {
+    "signed": {}
+  },
+  "width_info": {}
+}
diff --git a/src/etc/platform-intrinsics/nvptx/sreg.json b/src/etc/platform-intrinsics/nvptx/sreg.json
@@ -0,0 +1,90 @@
+{
+    "intrinsic_prefix": "_",
+    "llvm_prefix": "llvm.nvvm.read.ptx.sreg.",
+    "intrinsics": [
+        {
+            "intrinsic": "block_dim_x",
+            "width": ["0"],
+            "llvm": "ntid.x",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "block_dim_y",
+            "width": ["0"],
+            "llvm": "ntid.y",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "block_dim_z",
+            "width": ["0"],
+            "llvm": "ntid.z",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "block_idx_x",
+            "width": ["0"],
+            "llvm": "ctaid.x",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "block_idx_y",
+            "width": ["0"],
+            "llvm": "ctaid.y",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "block_idx_z",
+            "width": ["0"],
+            "llvm": "ctaid.z",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "grid_dim_x",
+            "width": ["0"],
+            "llvm": "nctaid.x",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "grid_dim_y",
+            "width": ["0"],
+            "llvm": "nctaid.y",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "grid_dim_z",
+            "width": ["0"],
+            "llvm": "nctaid.z",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "thread_idx_x",
+            "width": ["0"],
+            "llvm": "tid.x",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "thread_idx_y",
+            "width": ["0"],
+            "llvm": "tid.y",
+            "ret": "S32",
+            "args": []
+        },
+        {
+            "intrinsic": "thread_idx_z",
+            "width": ["0"],
+            "llvm": "tid.z",
+            "ret": "S32",
+            "args": []
+        }
+    ]
+}
diff --git a/src/libcore/intrinsics.rs b/src/libcore/intrinsics.rs
@@ -598,21 +598,3 @@ extern "rust-intrinsic" {
     pub fn try(f: fn(*mut u8), data: *mut u8, local_ptr: *mut u8) -> i32;
 
 }
-
-#[cfg(not(stage0))]
-#[cfg(arch = "nvptx")]
-extern "rust-intrinsic" {
-    pub fn thread_idx_x() -> i32;
-    pub fn thread_idx_y() -> i32;
-    pub fn thread_idx_z() -> i32;
-    pub fn block_idx_x() -> i32;
-    pub fn block_idx_y() -> i32;
-    pub fn block_idx_z() -> i32;
-    pub fn block_dim_x() -> i32;
-    pub fn block_dim_y() -> i32;
-    pub fn block_dim_z() -> i32;
-    pub fn grid_dim_x() -> i32;
-    pub fn grid_dim_y() -> i32;
-    pub fn grid_dim_z() -> i32;
-    pub fn syncthreads();
-}
diff --git a/src/librustc_platform_intrinsics/lib.rs b/src/librustc_platform_intrinsics/lib.rs
@@ -95,6 +95,7 @@ static VOID: Type = Type::Void;
 mod x86;
 mod arm;
 mod aarch64;
+mod nvptx;
 
 impl Intrinsic {
     pub fn find(name: &str) -> Option<Intrinsic> {
@@ -104,6 +105,8 @@ impl Intrinsic {
             arm::find(name)
         } else if name.starts_with("aarch64_") {
             aarch64::find(name)
+        } else if name.starts_with("nvptx_") {
+            nvptx::find(name)
         } else {
             None
         }

diff --git a/src/librustc_platform_intrinsics/nvptx.rs b/src/librustc_platform_intrinsics/nvptx.rs
@@ -0,0 +1,92 @@
+// Copyright 2015 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// DO NOT EDIT: autogenerated by etc/platform-intrinsics/generator.py
+// ignore-tidy-linelength
+
+#![allow(unused_imports)]
+
+use {Intrinsic, Type};
+use IntrinsicDef::Named;
+
+// The default inlining settings trigger a pathological behaviour in
+// LLVM, which causes makes compilation very slow. See #28273.
+#[inline(never)]
+pub fn find(name: &str) -> Option<Intrinsic> {
+    if !name.starts_with("nvptx") { return None }
+    Some(match &name["nvptx".len()..] {
+        "_syncthreads" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::VOID,
+            definition: Named("llvm.cuda.syncthreads")
+        },
+        "_block_dim_x" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.ntid.x")
+        },
+        "_block_dim_y" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.ntid.y")
+        },
+        "_block_dim_z" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.ntid.z")
+        },
+        "_block_idx_x" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.ctaid.x")
+        },
+        "_block_idx_y" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.ctaid.y")
+        },
+        "_block_idx_z" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.ctaid.z")
+        },
+        "_grid_dim_x" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.nctaid.x")
+        },
+        "_grid_dim_y" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.nctaid.y")
+        },
+        "_grid_dim_z" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.nctaid.z")
+        },
+        "_thread_idx_x" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.tid.x")
+        },
+        "_thread_idx_y" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.tid.y")
+        },
+        "_thread_idx_z" => Intrinsic {
+            inputs: { static INPUTS: [&'static Type; 0] = []; &INPUTS },
+            output: &::I32,
+            definition: Named("llvm.nvvm.read.ptx.sreg.tid.z")
+        },
+        _ => return None,
+    })
+}
diff --git a/src/librustc_trans/context.rs b/src/librustc_trans/context.rs
@@ -1084,19 +1084,6 @@ fn declare_intrinsic(ccx: &CrateContext, key: &str) -> Option<ValueRef> {
     ifn!("llvm.localrecover", fn(i8p, i8p, t_i32) -> i8p);
     ifn!("llvm.x86.seh.recoverfp", fn(i8p, i8p) -> i8p);
 
-    ifn!("llvm.cuda.syncthreads", fn() -> void);
-    ifn!("llvm.nvvm.read.ptx.sreg.tid.x", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.tid.y", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.tid.z", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.ctaid.x", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.ctaid.y", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.ctaid.z", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.ntid.x", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.ntid.y", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.ntid.z", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.nctaid.x", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.nctaid.y", fn() -> t_i32);
-    ifn!("llvm.nvvm.read.ptx.sreg.nctaid.z", fn() -> t_i32);
     ifn!("llvm.assume", fn(i1) -> void);
 
     if ccx.sess().opts.debuginfo != NoDebugInfo {

diff --git a/src/librustc_trans/intrinsic.rs b/src/librustc_trans/intrinsic.rs
@@ -89,19 +89,6 @@ fn get_simple_intrinsic(ccx: &CrateContext, name: &str) -> Option<ValueRef> {
         "roundf32" => "llvm.round.f32",
         "roundf64" => "llvm.round.f64",
         "assume" => "llvm.assume",
-        "thread_idx_x" => "llvm.nvvm.read.ptx.sreg.tid.x",
-        "thread_idx_y" => "llvm.nvvm.read.ptx.sreg.tid.y",
-        "thread_idx_z" => "llvm.nvvm.read.ptx.sreg.tid.z",
-        "block_idx_x" => "llvm.nvvm.read.ptx.sreg.ctaid.x",
-        "block_idx_y" => "llvm.nvvm.read.ptx.sreg.ctaid.y",
-        "block_idx_z" => "llvm.nvvm.read.ptx.sreg.ctaid.z",
-        "block_dim_x" => "llvm.nvvm.read.ptx.sreg.ntid.x",
-        "block_dim_y" => "llvm.nvvm.read.ptx.sreg.ntid.y",
-        "block_dim_z" => "llvm.nvvm.read.ptx.sreg.ntid.z",
-        "grid_dim_x" => "llvm.nvvm.read.ptx.sreg.nctaid.x",
-        "grid_dim_y" => "llvm.nvvm.read.ptx.sreg.nctaid.y",
-        "grid_dim_z" => "llvm.nvvm.read.ptx.sreg.nctaid.z",
-        "syncthreads" => "llvm.cuda.syncthreads",
         _ => return None
     };
     Some(ccx.get_intrinsic(&llvm_name))

diff --git a/src/librustc_typeck/check/intrinsic.rs b/src/librustc_typeck/check/intrinsic.rs
@@ -296,15 +296,6 @@ pub fn check_intrinsic_type(ccx: &CrateCtxt, it: &hir::ForeignItem) {
                 (0, vec![tcx.mk_fn_ptr(fn_ty), mut_u8, mut_u8], tcx.types.i32)
             }
 
-            "thread_idx_x" | "thread_idx_y" | "thread_idx_z" |
-            "block_idx_x" | "block_idx_y" | "block_idx_z" |
-            "block_dim_x" | "block_dim_y" | "block_dim_z" |
-            "grid_dim_x" | "grid_dim_y" | "grid_dim_z" => {
-                (0, vec![], tcx.types.i32)
-            }
-
-            "syncthreads" => (0, vec![], tcx.mk_nil()),
-
             ref other => {
                 span_err!(tcx.sess, it.span, E0093,
                           "unrecognized intrinsic function: `{}`", *other);

diff --git a/src/test/run-make/nvptx/Makefile b/src/test/run-make/nvptx/Makefile
@@ -0,0 +1,32 @@
+-include ../tools.mk
+
+NVPTX = nvptx-unknown-unknown
+NVPTX64 = nvptx64-unknown-unknown
+
+all: smoke syncthreads thread_idx
+
+smoke:
+	$(RUSTC) --target $(NVPTX) --emit=asm $@.rs
+	grep '\.address_size *32' $(TMPDIR)/$@.s
+	$(RUSTC) --target $(NVPTX64) --emit=asm $@.rs
+	grep '\.address_size *64' $(TMPDIR)/$@.s
+
+syncthreads:
+	$(RUSTC) --target $(NVPTX) --emit=llvm-ir $@.rs
+	grep 'call.*syncthreads' $(TMPDIR)/$@.ll
+	$(RUSTC) --target $(NVPTX64) --emit=llvm-ir $@.rs
+	grep 'call.*syncthreads' $(TMPDIR)/$@.ll
+	$(RUSTC) --target $(NVPTX) --emit=asm $@.rs
+	grep 'bar\.sync' $(TMPDIR)/$@.s
+	$(RUSTC) --target $(NVPTX64) --emit=asm $@.rs
+	grep 'bar\.sync' $(TMPDIR)/$@.s
+
+thread_idx:
+	$(RUSTC) --target $(NVPTX) --emit=llvm-ir $@.rs
+	grep 'call.*tid\.x' $(TMPDIR)/$@.ll
+	$(RUSTC) --target $(NVPTX64) --emit=llvm-ir $@.rs
+	grep 'call.*tid\.x' $(TMPDIR)/$@.ll
+	$(RUSTC) --target $(NVPTX) --emit=asm $@.rs
+	grep '%tid\.x' $(TMPDIR)/$@.s
+	$(RUSTC) --target $(NVPTX64) --emit=asm $@.rs
+	grep '%tid\.x' $(TMPDIR)/$@.s
diff --git a/src/test/run-make/nvptx/smoke.rs b/src/test/run-make/nvptx/smoke.rs
@@ -0,0 +1,13 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+#![feature(no_core)]
+
+#![no_core]
diff --git a/src/test/run-make/nvptx/syncthreads.rs b/src/test/run-make/nvptx/syncthreads.rs
@@ -0,0 +1,33 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+#![feature(lang_items)]
+#![feature(no_core)]
+#![feature(platform_intrinsics)]
+
+#![no_core]
+
+#![allow(dead_code)]
+
+extern "platform-intrinsic" {
+    fn nvptx_syncthreads();
+}
+
+fn syncthreads() {
+    unsafe {
+        nvptx_syncthreads();
+    }
+}
+
+#[lang = "copy"]
+trait Copy {}
+
+#[lang = "sized"]
+trait Sized {}