implement lzcnt and tzcnt

lib/compiler-singlepass/Cargo.toml

@@ -18,15 +18,16 @@ name = "wasmer_compiler_singlepass"
 wasmer-compiler = { path = "../compiler", package = "wasmer-compiler-near", version = "=2.4.0", features = ["translator"], default-features = false }
 wasmer-vm = { path = "../vm", package = "wasmer-vm-near", version = "=2.4.0" }
 wasmer-types = { path = "../types", package = "wasmer-types-near", version = "=2.4.0", default-features = false, features = ["std"] }
-rayon = { version = "1.5", optional = true }
-hashbrown = { version = "0.11", optional = true }
-more-asserts = "0.2"
+byteorder = "1.3"
 dynasm = "1.0"
 dynasmrt = "1.0"
+enumset = "1.0"
+hashbrown = { version = "0.11", optional = true }
 lazy_static = "1.4"
-byteorder = "1.3"
-smallvec = "1.6"
 memoffset = "0.6"
+more-asserts = "0.2"
+rayon = { version = "1.5", optional = true }
+smallvec = "1.6"
 tracing = "0.1"
 
 [dev-dependencies]

lib/compiler-singlepass/src/codegen_x64.rs

@@ -13,7 +13,7 @@ use wasmer_compiler::{
     CallingConvention, CompiledFunction, CompiledFunctionFrameInfo, CustomSection,
     CustomSectionProtection, FunctionBody, FunctionBodyData, InstructionAddressMap,
     ModuleTranslationState, Relocation, RelocationKind, RelocationTarget, SectionBody,
-    SectionIndex, SourceLoc,
+    SectionIndex, SourceLoc, Target,
 };
 use wasmer_types::{
     entity::{EntityRef, PrimaryMap, SecondaryMap},
@@ -39,6 +39,9 @@ pub(crate) struct FuncGen<'a> {
     /// ModuleInfo compilation config.
     config: &'a Singlepass,
 
+    /// Target to which we compile
+    target: &'a Target,
+
     /// Offsets of vmctx fields.
     vmoffsets: &'a VMOffsets,
 
@@ -1900,6 +1903,7 @@ impl<'a> FuncGen<'a> {
         module: &'a ModuleInfo,
         module_translation_state: &'a ModuleTranslationState,
         config: &'a Singlepass,
+        target: &'a Target,
         vmoffsets: &'a VMOffsets,
         _table_styles: &'a PrimaryMap<TableIndex, TableStyle>,
         local_func_index: LocalFunctionIndex,
@@ -1926,6 +1930,7 @@ impl<'a> FuncGen<'a> {
             module,
             module_translation_state,
             config,
+            target,
             vmoffsets,
             local_types: wasmer_types::partial_sum_map::PartialSumMap::new(),
             assembler,
@@ -2321,7 +2326,7 @@ impl<'a> FuncGen<'a> {
                     }
                 };
 
-                if self.assembler.arch_has_xzcnt() {
+                if self.assembler.arch_has_xzcnt(self.target.cpu_features()) {
                     self.assembler.arch_emit_lzcnt(
                         Size::S32,
                         Location::GPR(src),
@@ -2386,7 +2391,7 @@ impl<'a> FuncGen<'a> {
                     }
                 };
 
-                if self.assembler.arch_has_xzcnt() {
+                if self.assembler.arch_has_xzcnt(self.target.cpu_features()) {
                     self.assembler.arch_emit_tzcnt(
                         Size::S32,
                         Location::GPR(src),
@@ -2552,7 +2557,7 @@ impl<'a> FuncGen<'a> {
                     }
                 };
 
-                if self.assembler.arch_has_xzcnt() {
+                if self.assembler.arch_has_xzcnt(self.target.cpu_features()) {
                     self.assembler.arch_emit_lzcnt(
                         Size::S64,
                         Location::GPR(src),
@@ -2617,7 +2622,7 @@ impl<'a> FuncGen<'a> {
                     }
                 };
 
-                if self.assembler.arch_has_xzcnt() {
+                if self.assembler.arch_has_xzcnt(self.target.cpu_features()) {
                     self.assembler.arch_emit_tzcnt(
                         Size::S64,
                         Location::GPR(src),

lib/compiler-singlepass/src/compiler.rs

@@ -117,6 +117,7 @@ impl Compiler for SinglepassCompiler {
                         module,
                         module_translation,
                         &self.config,
+                        &target,
                         &vmoffsets,
                         &table_styles,
                         i,

lib/compiler-singlepass/src/emitter_x64.rs

@@ -3,6 +3,8 @@ use dynasm::dynasm;
 use dynasmrt::{
     x64::X64Relocation, AssemblyOffset, DynamicLabel, DynasmApi, DynasmLabelApi, VecAssembler,
 };
+use enumset::EnumSet;
+use wasmer_compiler::CpuFeature;
 
 type Assembler = VecAssembler<X64Relocation>;
 
@@ -287,7 +289,7 @@ pub(crate) trait Emitter {
         unimplemented!()
     }
 
-    fn arch_has_xzcnt(&self) -> bool {
+    fn arch_has_xzcnt(&self, _cpu: &EnumSet<CpuFeature>) -> bool {
         false
     }
     fn arch_emit_lzcnt(&mut self, _sz: Size, _src: Location, _dst: Location) {
@@ -1419,6 +1421,26 @@ impl Emitter for Assembler {
         self.emit_jmp_location(Location::GPR(target));
     }
 
+    fn arch_has_xzcnt(&self, cpu: &EnumSet<CpuFeature>) -> bool {
+        cpu.contains(CpuFeature::BMI1) && cpu.contains(CpuFeature::LZCNT)
+    }
+
+    fn arch_emit_lzcnt(&mut self, sz: Size, src: Location, dst: Location) {
+        binop_gpr_gpr!(lzcnt, self, sz, src, dst, {
+            binop_mem_gpr!(lzcnt, self, sz, src, dst, {
+                panic!("singlepass can't emit LZCNT {:?} {:?} {:?}", sz, src, dst)
+            })
+        })
+    }
+
+    fn arch_emit_tzcnt(&mut self, sz: Size, src: Location, dst: Location) {
+        binop_gpr_gpr!(tzcnt, self, sz, src, dst, {
+            binop_mem_gpr!(tzcnt, self, sz, src, dst, {
+                panic!("singlepass can't emit TZCNT {:?} {:?} {:?}", sz, src, dst)
+            })
+        })
+    }
+
     fn arch_mov64_imm_offset(&self) -> usize {
         2
     }