Auto merge of rust-lang#140256 - matthiaskrgr:rollup-8if58zf, r=matthiaskrgr

Rollup of 8 pull requests

Successful merges:

 - rust-lang#136083 (Suggest {to,from}_ne_bytes for transmutations between arrays and integers, etc)
 - rust-lang#138282 (Add `#[repr(u128)]`/`#[repr(i128)]` enums to `improper_ctypes_definitions`)
 - rust-lang#139700 (Autodiff flags)
 - rust-lang#140139 (rustc_target: Adjust RISC-V feature implication)
 - rust-lang#140141 (Move zkVM constants into `sys::env_consts`)
 - rust-lang#140150 (fix MAX_EXP and MIN_EXP docs)
 - rust-lang#140172 (Make algebraic functions into `const fn` items.)
 - rust-lang#140191 (Remove git repository from git config)

r? `@ghost`
`@rustbot` modify labels: rollup

Author: bors

compiler/rustc_codegen_cranelift/example/example.rs

@@ -1,6 +1,6 @@
 #![feature(no_core, unboxed_closures)]
 #![no_core]
-#![allow(dead_code)]
+#![allow(dead_code, unnecessary_transmutes)]
 
 extern crate mini_core;
 

compiler/rustc_codegen_gcc/example/example.rs

@@ -1,6 +1,6 @@
 #![feature(no_core, unboxed_closures)]
 #![no_core]
-#![allow(dead_code)]
+#![allow(dead_code, unnecessary_transmutes)]
 
 extern crate mini_core;
 
@@ -11,11 +11,7 @@ fn abc(a: u8) -> u8 {
 }
 
 fn bcd(b: bool, a: u8) -> u8 {
-    if b {
-        a * 2
-    } else {
-        a * 3
-    }
+    if b { a * 2 } else { a * 3 }
 }
 
 fn call() {

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -584,12 +584,10 @@ fn thin_lto(
     }
 }
 
-fn enable_autodiff_settings(ad: &[config::AutoDiff], module: &mut ModuleCodegen<ModuleLlvm>) {
+fn enable_autodiff_settings(ad: &[config::AutoDiff]) {
     for &val in ad {
+        // We intentionally don't use a wildcard, to not forget handling anything new.
         match val {
-            config::AutoDiff::PrintModBefore => {
-                unsafe { llvm::LLVMDumpModule(module.module_llvm.llmod()) };
-            }
             config::AutoDiff::PrintPerf => {
                 llvm::set_print_perf(true);
             }
@@ -603,17 +601,23 @@ fn enable_autodiff_settings(ad: &[config::AutoDiff], module: &mut ModuleCodegen<
                 llvm::set_inline(true);
             }
             config::AutoDiff::LooseTypes => {
-                llvm::set_loose_types(false);
+                llvm::set_loose_types(true);
             }
             config::AutoDiff::PrintSteps => {
                 llvm::set_print(true);
             }
-            // We handle this below
+            // We handle this in the PassWrapper.cpp
+            config::AutoDiff::PrintPasses => {}
+            // We handle this in the PassWrapper.cpp
+            config::AutoDiff::PrintModBefore => {}
+            // We handle this in the PassWrapper.cpp
             config::AutoDiff::PrintModAfter => {}
-            // We handle this below
+            // We handle this in the PassWrapper.cpp
             config::AutoDiff::PrintModFinal => {}
             // This is required and already checked
             config::AutoDiff::Enable => {}
+            // We handle this below
+            config::AutoDiff::NoPostopt => {}
         }
     }
     // This helps with handling enums for now.
@@ -647,27 +651,27 @@ pub(crate) fn run_pass_manager(
     // We then run the llvm_optimize function a second time, to optimize the code which we generated
     // in the enzyme differentiation pass.
     let enable_ad = config.autodiff.contains(&config::AutoDiff::Enable);
-    let stage =
-        if enable_ad { write::AutodiffStage::DuringAD } else { write::AutodiffStage::PostAD };
+    let stage = if thin {
+        write::AutodiffStage::PreAD
+    } else {
+        if enable_ad { write::AutodiffStage::DuringAD } else { write::AutodiffStage::PostAD }
+    };
 
     if enable_ad {
-        enable_autodiff_settings(&config.autodiff, module);
+        enable_autodiff_settings(&config.autodiff);
     }
 
     unsafe {
         write::llvm_optimize(cgcx, dcx, module, None, config, opt_level, opt_stage, stage)?;
     }
 
-    if cfg!(llvm_enzyme) && enable_ad {
-        // This is the post-autodiff IR, mainly used for testing and educational purposes.
-        if config.autodiff.contains(&config::AutoDiff::PrintModAfter) {
-            unsafe { llvm::LLVMDumpModule(module.module_llvm.llmod()) };
-        }
-
+    if cfg!(llvm_enzyme) && enable_ad && !thin {
         let opt_stage = llvm::OptStage::FatLTO;
         let stage = write::AutodiffStage::PostAD;
-        unsafe {
-            write::llvm_optimize(cgcx, dcx, module, None, config, opt_level, opt_stage, stage)?;
+        if !config.autodiff.contains(&config::AutoDiff::NoPostopt) {
+            unsafe {
+                write::llvm_optimize(cgcx, dcx, module, None, config, opt_level, opt_stage, stage)?;
+            }
         }
 
         // This is the final IR, so people should be able to inspect the optimized autodiff output,

compiler/rustc_codegen_llvm/src/back/write.rs

@@ -572,20 +572,31 @@ pub(crate) unsafe fn llvm_optimize(
 
     let consider_ad = cfg!(llvm_enzyme) && config.autodiff.contains(&config::AutoDiff::Enable);
     let run_enzyme = autodiff_stage == AutodiffStage::DuringAD;
+    let print_before_enzyme = config.autodiff.contains(&config::AutoDiff::PrintModBefore);
+    let print_after_enzyme = config.autodiff.contains(&config::AutoDiff::PrintModAfter);
+    let print_passes = config.autodiff.contains(&config::AutoDiff::PrintPasses);
+    let merge_functions;
     let unroll_loops;
     let vectorize_slp;
     let vectorize_loop;
 
     // When we build rustc with enzyme/autodiff support, we want to postpone size-increasing
     // optimizations until after differentiation. Our pipeline is thus: (opt + enzyme), (full opt).
     // We therefore have two calls to llvm_optimize, if autodiff is used.
+    //
+    // We also must disable merge_functions, since autodiff placeholder/dummy bodies tend to be
+    // identical. We run opts before AD, so there is a chance that LLVM will merge our dummies.
+    // In that case, we lack some dummy bodies and can't replace them with the real AD code anymore.
+    // We then would need to abort compilation. This was especially common in test cases.
     if consider_ad && autodiff_stage != AutodiffStage::PostAD {
+        merge_functions = false;
         unroll_loops = false;
         vectorize_slp = false;
         vectorize_loop = false;
     } else {
         unroll_loops =
             opt_level != config::OptLevel::Size && opt_level != config::OptLevel::SizeMin;
+        merge_functions = config.merge_functions;
         vectorize_slp = config.vectorize_slp;
         vectorize_loop = config.vectorize_loop;
     }
@@ -663,13 +674,16 @@ pub(crate) unsafe fn llvm_optimize(
             thin_lto_buffer,
             config.emit_thin_lto,
             config.emit_thin_lto_summary,
-            config.merge_functions,
+            merge_functions,
             unroll_loops,
             vectorize_slp,
             vectorize_loop,
             config.no_builtins,
             config.emit_lifetime_markers,
             run_enzyme,
+            print_before_enzyme,
+            print_after_enzyme,
+            print_passes,
             sanitizer_options.as_ref(),
             pgo_gen_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()),
             pgo_use_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()),

compiler/rustc_codegen_llvm/src/builder/autodiff.rs

@@ -473,7 +473,7 @@ pub(crate) fn differentiate<'ll>(
         return Err(diag_handler.handle().emit_almost_fatal(AutoDiffWithoutEnable));
     }
 
-    // Before dumping the module, we want all the TypeTrees to become part of the module.
+    // Here we replace the placeholder code with the actual autodiff code, which calls Enzyme.
     for item in diff_items.iter() {
         let name = item.source.clone();
         let fn_def: Option<&llvm::Value> = cx.get_function(&name);

compiler/rustc_codegen_llvm/src/llvm/ffi.rs

@@ -2454,6 +2454,9 @@ unsafe extern "C" {
         DisableSimplifyLibCalls: bool,
         EmitLifetimeMarkers: bool,
         RunEnzyme: bool,
+        PrintBeforeEnzyme: bool,
+        PrintAfterEnzyme: bool,
+        PrintPasses: bool,
         SanitizerOptions: Option<&SanitizerOptions>,
         PGOGenPath: *const c_char,
         PGOUsePath: *const c_char,

compiler/rustc_const_eval/src/interpret/intrinsics.rs

@@ -158,6 +158,31 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
                 self.copy_op(&val, dest)?;
             }
 
+            sym::fadd_algebraic
+            | sym::fsub_algebraic
+            | sym::fmul_algebraic
+            | sym::fdiv_algebraic
+            | sym::frem_algebraic => {
+                let a = self.read_immediate(&args[0])?;
+                let b = self.read_immediate(&args[1])?;
+
+                let op = match intrinsic_name {
+                    sym::fadd_algebraic => BinOp::Add,
+                    sym::fsub_algebraic => BinOp::Sub,
+                    sym::fmul_algebraic => BinOp::Mul,
+                    sym::fdiv_algebraic => BinOp::Div,
+                    sym::frem_algebraic => BinOp::Rem,
+
+                    _ => bug!(),
+                };
+
+                let res = self.binary_op(op, &a, &b)?;
+                // `binary_op` already called `generate_nan` if needed.
+
+                // FIXME: Miri should add some non-determinism to the result here to catch any dependences on exact computations. This has previously been done, but the behaviour was removed as part of constification.
+                self.write_immediate(*res, dest)?;
+            }
+
             sym::ctpop
             | sym::cttz
             | sym::cttz_nonzero

compiler/rustc_lint/src/types.rs

@@ -1,7 +1,9 @@
 use std::iter;
 use std::ops::ControlFlow;
 
-use rustc_abi::{BackendRepr, TagEncoding, VariantIdx, Variants, WrappingRange};
+use rustc_abi::{
+    BackendRepr, Integer, IntegerType, TagEncoding, VariantIdx, Variants, WrappingRange,
+};
 use rustc_data_structures::fx::FxHashSet;
 use rustc_errors::DiagMessage;
 use rustc_hir::intravisit::VisitorExt;
@@ -1243,6 +1245,14 @@ impl<'a, 'tcx> ImproperCTypesVisitor<'a, 'tcx> {
                             };
                         }
 
+                        if let Some(IntegerType::Fixed(Integer::I128, _)) = def.repr().int {
+                            return FfiUnsafe {
+                                ty,
+                                reason: fluent::lint_improper_ctypes_128bit,
+                                help: None,
+                            };
+                        }
+
                         use improper_ctypes::check_non_exhaustive_variant;
 
                         let non_exhaustive = def.variant_list_has_applicable_non_exhaustive();

compiler/rustc_lint_defs/src/builtin.rs

@@ -117,6 +117,7 @@ declare_lint_pass! {
         UNKNOWN_OR_MALFORMED_DIAGNOSTIC_ATTRIBUTES,
         UNNAMEABLE_TEST_ITEMS,
         UNNAMEABLE_TYPES,
+        UNNECESSARY_TRANSMUTES,
         UNREACHABLE_CODE,
         UNREACHABLE_PATTERNS,
         UNSAFE_ATTR_OUTSIDE_UNSAFE,
@@ -4909,6 +4910,30 @@ declare_lint! {
     "detects pointer to integer transmutes in const functions and associated constants",
 }
 
+declare_lint! {
+    /// The `unnecessary_transmutes` lint detects transmutations that have safer alternatives.
+    ///
+    /// ### Example
+    ///
+    /// ```rust
+    /// fn bytes_at_home(x: [u8; 4]) -> u32 {
+    ///   unsafe { std::mem::transmute(x) }
+    /// }
+    /// ```
+    ///
+    /// {{produces}}
+    ///
+    /// ### Explanation
+    ///
+    /// Using an explicit method is preferable over calls to
+    /// [`transmute`](https://doc.rust-lang.org/std/mem/fn.transmute.html) as
+    /// they more clearly communicate the intent, are easier to review, and
+    /// are less likely to accidentally result in unsoundness.
+    pub UNNECESSARY_TRANSMUTES,
+    Warn,
+    "detects transmutes that are shadowed by std methods"
+}
+
 declare_lint! {
     /// The `tail_expr_drop_order` lint looks for those values generated at the tail expression location,
     /// that runs a custom `Drop` destructor.

compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp

@@ -14,6 +14,7 @@
 #include "llvm/IR/LegacyPassManager.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/IR/Verifier.h"
+#include "llvm/IRPrinter/IRPrintingPasses.h"
 #include "llvm/LTO/LTO.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/MC/TargetRegistry.h"
@@ -703,7 +704,8 @@ extern "C" LLVMRustResult LLVMRustOptimize(
     bool LintIR, LLVMRustThinLTOBuffer **ThinLTOBufferRef, bool EmitThinLTO,
     bool EmitThinLTOSummary, bool MergeFunctions, bool UnrollLoops,
     bool SLPVectorize, bool LoopVectorize, bool DisableSimplifyLibCalls,
-    bool EmitLifetimeMarkers, bool RunEnzyme,
+    bool EmitLifetimeMarkers, bool RunEnzyme, bool PrintBeforeEnzyme,
+    bool PrintAfterEnzyme, bool PrintPasses,
     LLVMRustSanitizerOptions *SanitizerOptions, const char *PGOGenPath,
     const char *PGOUsePath, bool InstrumentCoverage,
     const char *InstrProfileOutput, const char *PGOSampleUsePath,
@@ -1048,14 +1050,38 @@ extern "C" LLVMRustResult LLVMRustOptimize(
   // now load "-enzyme" pass:
 #ifdef ENZYME
   if (RunEnzyme) {
-    registerEnzymeAndPassPipeline(PB, true);
+
+    if (PrintBeforeEnzyme) {
+      // Handle the Rust flag `-Zautodiff=PrintModBefore`.
+      std::string Banner = "Module before EnzymeNewPM";
+      MPM.addPass(PrintModulePass(outs(), Banner, true, false));
+    }
+
+    registerEnzymeAndPassPipeline(PB, false);
     if (auto Err = PB.parsePassPipeline(MPM, "enzyme")) {
       std::string ErrMsg = toString(std::move(Err));
       LLVMRustSetLastError(ErrMsg.c_str());
       return LLVMRustResult::Failure;
     }
+
+    if (PrintAfterEnzyme) {
+      // Handle the Rust flag `-Zautodiff=PrintModAfter`.
+      std::string Banner = "Module after EnzymeNewPM";
+      MPM.addPass(PrintModulePass(outs(), Banner, true, false));
+    }
   }
 #endif
+  if (PrintPasses) {
+    // Print all passes from the PM:
+    std::string Pipeline;
+    raw_string_ostream SOS(Pipeline);
+    MPM.printPipeline(SOS, [&PIC](StringRef ClassName) {
+      auto PassName = PIC.getPassNameForClassName(ClassName);
+      return PassName.empty() ? ClassName : PassName;
+    });
+    outs() << Pipeline;
+    outs() << "\n";
+  }
 
   // Upgrade all calls to old intrinsics first.
   for (Module::iterator I = TheModule->begin(), E = TheModule->end(); I != E;)