Rollup merge of rust-lang#130432 - azhogin:azhogin/regparm, r=workingjubilee,pnkfelix

rust_for_linux: -Zregparm=<N> commandline flag for X86 (rust-lang#116972)

Command line flag `-Zregparm=<N>` for X86 (32-bit) for rust-for-linux: rust-lang#116972
Implemented in the similar way as fastcall/vectorcall support (args are marked InReg if fit).

Author: workingjubilee

Diff for: compiler/rustc_codegen_gcc/src/builder.rs

@@ -30,7 +30,7 @@ use rustc_middle::ty::{Instance, ParamEnv, Ty, TyCtxt};
 use rustc_span::Span;
 use rustc_span::def_id::DefId;
 use rustc_target::abi::call::FnAbi;
-use rustc_target::spec::{HasTargetSpec, HasWasmCAbiOpt, Target, WasmCAbi};
+use rustc_target::spec::{HasTargetSpec, HasWasmCAbiOpt, HasX86AbiOpt, Target, WasmCAbi, X86Abi};
 
 use crate::common::{SignType, TypeReflection, type_is_pointer};
 use crate::context::CodegenCx;
@@ -2347,6 +2347,12 @@ impl<'tcx> HasWasmCAbiOpt for Builder<'_, '_, 'tcx> {
     }
 }
 
+impl<'tcx> HasX86AbiOpt for Builder<'_, '_, 'tcx> {
+    fn x86_abi_opt(&self) -> X86Abi {
+        self.cx.x86_abi_opt()
+    }
+}
+
 pub trait ToGccComp {
     fn to_gcc_comparison(&self) -> ComparisonOp;
 }

Diff for: compiler/rustc_codegen_gcc/src/context.rs

@@ -19,7 +19,9 @@ use rustc_session::Session;
 use rustc_span::source_map::respan;
 use rustc_span::{DUMMY_SP, Span};
 use rustc_target::abi::{HasDataLayout, PointeeInfo, Size, TargetDataLayout, VariantIdx};
-use rustc_target::spec::{HasTargetSpec, HasWasmCAbiOpt, Target, TlsModel, WasmCAbi};
+use rustc_target::spec::{
+    HasTargetSpec, HasWasmCAbiOpt, HasX86AbiOpt, Target, TlsModel, WasmCAbi, X86Abi,
+};
 
 use crate::callee::get_fn;
 use crate::common::SignType;
@@ -538,6 +540,12 @@ impl<'gcc, 'tcx> HasWasmCAbiOpt for CodegenCx<'gcc, 'tcx> {
     }
 }
 
+impl<'gcc, 'tcx> HasX86AbiOpt for CodegenCx<'gcc, 'tcx> {
+    fn x86_abi_opt(&self) -> X86Abi {
+        X86Abi { regparm: self.tcx.sess.opts.unstable_opts.regparm }
+    }
+}
+
 impl<'gcc, 'tcx> LayoutOfHelpers<'tcx> for CodegenCx<'gcc, 'tcx> {
     #[inline]
     fn handle_layout_err(&self, err: LayoutError<'tcx>, span: Span, ty: Ty<'tcx>) -> ! {

Diff for: compiler/rustc_interface/src/tests.rs

@@ -836,6 +836,7 @@ fn test_unstable_options_tracking_hash() {
     tracked!(profile_emit, Some(PathBuf::from("abc")));
     tracked!(profile_sample_use, Some(PathBuf::from("abc")));
     tracked!(profiler_runtime, "abc".to_string());
+    tracked!(regparm, Some(3));
     tracked!(relax_elf_relocations, Some(true));
     tracked!(remap_cwd_prefix, Some(PathBuf::from("abc")));
     tracked!(sanitizer, SanitizerSet::ADDRESS);

Diff for: compiler/rustc_middle/src/ty/layout.rs

@@ -21,7 +21,9 @@ use rustc_span::{DUMMY_SP, ErrorGuaranteed, Span};
 use rustc_target::abi::call::FnAbi;
 use rustc_target::abi::{FieldIdx, TyAbiInterface, VariantIdx, call};
 use rustc_target::spec::abi::Abi as SpecAbi;
-use rustc_target::spec::{HasTargetSpec, HasWasmCAbiOpt, PanicStrategy, Target, WasmCAbi};
+use rustc_target::spec::{
+    HasTargetSpec, HasWasmCAbiOpt, HasX86AbiOpt, PanicStrategy, Target, WasmCAbi, X86Abi,
+};
 use tracing::debug;
 use {rustc_abi as abi, rustc_hir as hir};
 
@@ -544,6 +546,12 @@ impl<'tcx> HasWasmCAbiOpt for TyCtxt<'tcx> {
     }
 }
 
+impl<'tcx> HasX86AbiOpt for TyCtxt<'tcx> {
+    fn x86_abi_opt(&self) -> X86Abi {
+        X86Abi { regparm: self.sess.opts.unstable_opts.regparm }
+    }
+}
+
 impl<'tcx> HasTyCtxt<'tcx> for TyCtxt<'tcx> {
     #[inline]
     fn tcx(&self) -> TyCtxt<'tcx> {
@@ -595,6 +603,12 @@ impl<'tcx> HasWasmCAbiOpt for LayoutCx<'tcx> {
     }
 }
 
+impl<'tcx> HasX86AbiOpt for LayoutCx<'tcx> {
+    fn x86_abi_opt(&self) -> X86Abi {
+        self.calc.cx.x86_abi_opt()
+    }
+}
+
 impl<'tcx> HasTyCtxt<'tcx> for LayoutCx<'tcx> {
     fn tcx(&self) -> TyCtxt<'tcx> {
         self.calc.cx

Diff for: compiler/rustc_session/messages.ftl

@@ -136,3 +136,6 @@ session_unsupported_crate_type_for_target =
     dropping unsupported crate type `{$crate_type}` for target `{$target_triple}`
 
 session_unsupported_dwarf_version = requested DWARF version {$dwarf_version} is greater than 5
+
+session_unsupported_regparm = `-Zregparm={$regparm}` is unsupported (valid values 0-3)
+session_unsupported_regparm_arch = `-Zregparm=N` is only supported on x86

Diff for: compiler/rustc_session/src/errors.rs

@@ -485,6 +485,16 @@ pub(crate) struct FunctionReturnRequiresX86OrX8664;
 #[diag(session_function_return_thunk_extern_requires_non_large_code_model)]
 pub(crate) struct FunctionReturnThunkExternRequiresNonLargeCodeModel;
 
+#[derive(Diagnostic)]
+#[diag(session_unsupported_regparm)]
+pub(crate) struct UnsupportedRegparm {
+    pub(crate) regparm: u32,
+}
+
+#[derive(Diagnostic)]
+#[diag(session_unsupported_regparm_arch)]
+pub(crate) struct UnsupportedRegparmArch;
+
 #[derive(Diagnostic)]
 #[diag(session_failed_to_create_profiler)]
 pub(crate) struct FailedToCreateProfiler {

Diff for: compiler/rustc_session/src/options.rs

@@ -2000,6 +2000,11 @@ options! {
         "enable queries of the dependency graph for regression testing (default: no)"),
     randomize_layout: bool = (false, parse_bool, [TRACKED],
         "randomize the layout of types (default: no)"),
+    regparm: Option<u32> = (None, parse_opt_number, [TRACKED],
+        "On x86-32 targets, setting this to N causes the compiler to pass N arguments \
+        in registers EAX, EDX, and ECX instead of on the stack for\
+        \"C\", \"cdecl\", and \"stdcall\" fn.\
+        It is UNSOUND to link together crates that use different values for this flag!"),
     relax_elf_relocations: Option<bool> = (None, parse_opt_bool, [TRACKED],
         "whether ELF relocations can be relaxed"),
     remap_cwd_prefix: Option<PathBuf> = (None, parse_opt_pathbuf, [TRACKED],

Diff for: compiler/rustc_session/src/session.rs

@@ -1337,6 +1337,15 @@ fn validate_commandline_args_with_session_available(sess: &Session) {
         }
     }
 
+    if let Some(regparm) = sess.opts.unstable_opts.regparm {
+        if regparm > 3 {
+            sess.dcx().emit_err(errors::UnsupportedRegparm { regparm });
+        }
+        if sess.target.arch != "x86" {
+            sess.dcx().emit_err(errors::UnsupportedRegparmArch);
+        }
+    }
+
     // The code model check applies to `thunk` and `thunk-extern`, but not `thunk-inline`, so it is
     // kept as a `match` to force a change if new ones are added, even if we currently only support
     // `thunk-extern` like Clang.

Diff for: compiler/rustc_target/src/callconv/mod.rs

@@ -6,7 +6,7 @@ use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 
 use crate::abi::{self, Abi, Align, HasDataLayout, Size, TyAbiInterface, TyAndLayout};
-use crate::spec::{self, HasTargetSpec, HasWasmCAbiOpt, WasmCAbi};
+use crate::spec::{self, HasTargetSpec, HasWasmCAbiOpt, HasX86AbiOpt, WasmCAbi};
 
 mod aarch64;
 mod amdgpu;
@@ -631,7 +631,7 @@ impl<'a, Ty> FnAbi<'a, Ty> {
     ) -> Result<(), AdjustForForeignAbiError>
     where
         Ty: TyAbiInterface<'a, C> + Copy,
-        C: HasDataLayout + HasTargetSpec + HasWasmCAbiOpt,
+        C: HasDataLayout + HasTargetSpec + HasWasmCAbiOpt + HasX86AbiOpt,
     {
         if abi == spec::abi::Abi::X86Interrupt {
             if let Some(arg) = self.args.first_mut() {
@@ -643,14 +643,18 @@ impl<'a, Ty> FnAbi<'a, Ty> {
         let spec = cx.target_spec();
         match &spec.arch[..] {
             "x86" => {
-                let flavor = if let spec::abi::Abi::Fastcall { .. }
-                | spec::abi::Abi::Vectorcall { .. } = abi
-                {
-                    x86::Flavor::FastcallOrVectorcall
-                } else {
-                    x86::Flavor::General
+                let (flavor, regparm) = match abi {
+                    spec::abi::Abi::Fastcall { .. } | spec::abi::Abi::Vectorcall { .. } => {
+                        (x86::Flavor::FastcallOrVectorcall, None)
+                    }
+                    spec::abi::Abi::C { .. }
+                    | spec::abi::Abi::Cdecl { .. }
+                    | spec::abi::Abi::Stdcall { .. } => {
+                        (x86::Flavor::General, cx.x86_abi_opt().regparm)
+                    }
+                    _ => (x86::Flavor::General, None),
                 };
-                x86::compute_abi_info(cx, self, flavor);
+                x86::compute_abi_info(cx, self, x86::X86Options { flavor, regparm });
             }
             "x86_64" => match abi {
                 spec::abi::Abi::SysV64 { .. } => x86_64::compute_abi_info(cx, self),

Diff for: compiler/rustc_target/src/callconv/x86.rs

@@ -8,7 +8,12 @@ pub(crate) enum Flavor {
     FastcallOrVectorcall,
 }
 
-pub(crate) fn compute_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>, flavor: Flavor)
+pub(crate) struct X86Options {
+    pub flavor: Flavor,
+    pub regparm: Option<u32>,
+}
+
+pub(crate) fn compute_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>, opts: X86Options)
 where
     Ty: TyAbiInterface<'a, C> + Copy,
     C: HasDataLayout + HasTargetSpec,
@@ -128,58 +133,77 @@ where
         }
     }
 
-    if flavor == Flavor::FastcallOrVectorcall {
-        // Mark arguments as InReg like clang does it,
-        // so our fastcall/vectorcall is compatible with C/C++ fastcall/vectorcall.
+    fill_inregs(cx, fn_abi, opts, false);
+}
 
-        // Clang reference: lib/CodeGen/TargetInfo.cpp
-        // See X86_32ABIInfo::shouldPrimitiveUseInReg(), X86_32ABIInfo::updateFreeRegs()
+pub(crate) fn fill_inregs<'a, Ty, C>(
+    cx: &C,
+    fn_abi: &mut FnAbi<'a, Ty>,
+    opts: X86Options,
+    rust_abi: bool,
+) where
+    Ty: TyAbiInterface<'a, C> + Copy,
+{
+    if opts.flavor != Flavor::FastcallOrVectorcall && opts.regparm.is_none_or(|x| x == 0) {
+        return;
+    }
+    // Mark arguments as InReg like clang does it,
+    // so our fastcall/vectorcall is compatible with C/C++ fastcall/vectorcall.
 
-        // IsSoftFloatABI is only set to true on ARM platforms,
-        // which in turn can't be x86?
+    // Clang reference: lib/CodeGen/TargetInfo.cpp
+    // See X86_32ABIInfo::shouldPrimitiveUseInReg(), X86_32ABIInfo::updateFreeRegs()
 
-        let mut free_regs = 2;
+    // IsSoftFloatABI is only set to true on ARM platforms,
+    // which in turn can't be x86?
 
-        for arg in fn_abi.args.iter_mut() {
-            let attrs = match arg.mode {
-                PassMode::Ignore
-                | PassMode::Indirect { attrs: _, meta_attrs: None, on_stack: _ } => {
-                    continue;
-                }
-                PassMode::Direct(ref mut attrs) => attrs,
-                PassMode::Pair(..)
-                | PassMode::Indirect { attrs: _, meta_attrs: Some(_), on_stack: _ }
-                | PassMode::Cast { .. } => {
-                    unreachable!("x86 shouldn't be passing arguments by {:?}", arg.mode)
-                }
-            };
+    // 2 for fastcall/vectorcall, regparm limited by 3 otherwise
+    let mut free_regs = opts.regparm.unwrap_or(2).into();
+
+    // For types generating PassMode::Cast, InRegs will not be set.
+    // Maybe, this is a FIXME
+    let has_casts = fn_abi.args.iter().any(|arg| matches!(arg.mode, PassMode::Cast { .. }));
+    if has_casts && rust_abi {
+        return;
+    }
 
-            // At this point we know this must be a primitive of sorts.
-            let unit = arg.layout.homogeneous_aggregate(cx).unwrap().unit().unwrap();
-            assert_eq!(unit.size, arg.layout.size);
-            if unit.kind == RegKind::Float {
+    for arg in fn_abi.args.iter_mut() {
+        let attrs = match arg.mode {
+            PassMode::Ignore | PassMode::Indirect { attrs: _, meta_attrs: None, on_stack: _ } => {
                 continue;
             }
+            PassMode::Direct(ref mut attrs) => attrs,
+            PassMode::Pair(..)
+            | PassMode::Indirect { attrs: _, meta_attrs: Some(_), on_stack: _ }
+            | PassMode::Cast { .. } => {
+                unreachable!("x86 shouldn't be passing arguments by {:?}", arg.mode)
+            }
+        };
 
-            let size_in_regs = (arg.layout.size.bits() + 31) / 32;
+        // At this point we know this must be a primitive of sorts.
+        let unit = arg.layout.homogeneous_aggregate(cx).unwrap().unit().unwrap();
+        assert_eq!(unit.size, arg.layout.size);
+        if matches!(unit.kind, RegKind::Float | RegKind::Vector) {
+            continue;
+        }
 
-            if size_in_regs == 0 {
-                continue;
-            }
+        let size_in_regs = (arg.layout.size.bits() + 31) / 32;
 
-            if size_in_regs > free_regs {
-                break;
-            }
+        if size_in_regs == 0 {
+            continue;
+        }
 
-            free_regs -= size_in_regs;
+        if size_in_regs > free_regs {
+            break;
+        }
 
-            if arg.layout.size.bits() <= 32 && unit.kind == RegKind::Integer {
-                attrs.set(ArgAttribute::InReg);
-            }
+        free_regs -= size_in_regs;
 
-            if free_regs == 0 {
-                break;
-            }
+        if arg.layout.size.bits() <= 32 && unit.kind == RegKind::Integer {
+            attrs.set(ArgAttribute::InReg);
+        }
+
+        if free_regs == 0 {
+            break;
         }
     }
 }

Diff for: compiler/rustc_target/src/spec/mod.rs

@@ -2096,6 +2096,18 @@ pub trait HasWasmCAbiOpt {
     fn wasm_c_abi_opt(&self) -> WasmCAbi;
 }
 
+/// x86 (32-bit) abi options.
+#[derive(Debug, Copy, Clone, Hash, PartialEq, Eq)]
+pub struct X86Abi {
+    /// On x86-32 targets, the regparm N causes the compiler to pass arguments
+    /// in registers EAX, EDX, and ECX instead of on the stack.
+    pub regparm: Option<u32>,
+}
+
+pub trait HasX86AbiOpt {
+    fn x86_abi_opt(&self) -> X86Abi;
+}
+
 type StaticCow<T> = Cow<'static, T>;
 
 /// Optional aspects of a target specification.

Diff for: src/doc/unstable-book/src/compiler-flags/regparm.md

@@ -0,0 +1,20 @@
+# `regparm`
+
+The tracking issue for this feature is: https://github.com/rust-lang/rust/issues/131749.
+
+------------------------
+
+Option -Zregparm=N causes the compiler to pass N arguments
+in registers EAX, EDX, and ECX instead of on the stack for "C", "cdecl", and "stdcall" fn.
+It is UNSOUND to link together crates that use different values for this flag.
+It is only supported on `x86`.
+
+It is equivalent to [Clang]'s and [GCC]'s `-mregparm`.
+
+Supported values for this option are 0-3.
+
+[Clang]: https://clang.llvm.org/docs/ClangCommandLineReference.html#cmdoption-clang-mregparm
+[GCC]: https://gcc.gnu.org/onlinedocs/gcc/x86-Function-Attributes.html#index-regparm-function-attribute_002c-x86
+
+Implementation details:
+For eligible arguments, llvm `inreg` attribute is set.