diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index e6d3375fb..e06a1fb88 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -11,7 +11,13 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, macos-latest]
+        include:
+          - os: ubuntu-latest
+          - os: macos-latest
+          # cross-compile from Linux to Windows using mingw
+          - os: ubuntu-latest
+            env:
+              TARGET_TRIPLE: x86_64-pc-windows-gnu
 
     steps:
     - uses: actions/checkout@v2
@@ -36,6 +42,12 @@ jobs:
         path: target
         key: ${{ runner.os }}-cargo-build-target-${{ hashFiles('rust-toolchain', '**/Cargo.lock') }}
 
+    - name: Install MinGW toolchain and wine
+      if: matrix.os == 'ubuntu-latest' && matrix.env.TARGET_TRIPLE == 'x86_64-pc-windows-gnu'
+      run: |
+        sudo apt-get install -y gcc-mingw-w64-x86-64 wine-stable
+        rustup target add x86_64-pc-windows-gnu
+
     - name: Prepare dependencies
       run: |
         git config --global user.email "user@example.com"
@@ -43,6 +55,8 @@ jobs:
         ./prepare.sh
 
     - name: Test
+      env:
+        TARGET_TRIPLE: ${{ matrix.env.TARGET_TRIPLE }}
       run: |
         # Enable backtraces for easier debugging
         export RUST_BACKTRACE=1
@@ -57,6 +71,7 @@ jobs:
       run: tar cvfJ cg_clif.tar.xz build
 
     - name: Upload prebuilt cg_clif
+      if: matrix.env.TARGET_TRIPLE != 'x86_64-pc-windows-gnu'
       uses: actions/upload-artifact@v2
       with:
         name: cg_clif-${{ runner.os }}
diff --git a/Cargo.lock b/Cargo.lock
index 76d9f0d27..1ed186032 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -39,16 +39,16 @@ checksum = "baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd"
 
 [[package]]
 name = "cranelift-bforest"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "cranelift-entity",
 ]
 
 [[package]]
 name = "cranelift-codegen"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "byteorder",
  "cranelift-bforest",
@@ -65,8 +65,8 @@ dependencies = [
 
 [[package]]
 name = "cranelift-codegen-meta"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "cranelift-codegen-shared",
  "cranelift-entity",
@@ -74,18 +74,18 @@ dependencies = [
 
 [[package]]
 name = "cranelift-codegen-shared"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 
 [[package]]
 name = "cranelift-entity"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 
 [[package]]
 name = "cranelift-frontend"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "cranelift-codegen",
  "log",
@@ -95,8 +95,8 @@ dependencies = [
 
 [[package]]
 name = "cranelift-jit"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "anyhow",
  "cranelift-codegen",
@@ -113,8 +113,8 @@ dependencies = [
 
 [[package]]
 name = "cranelift-module"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "anyhow",
  "cranelift-codegen",
@@ -125,8 +125,8 @@ dependencies = [
 
 [[package]]
 name = "cranelift-native"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "cranelift-codegen",
  "target-lexicon",
@@ -134,8 +134,8 @@ dependencies = [
 
 [[package]]
 name = "cranelift-object"
-version = "0.70.0"
-source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#cdb60ec5a9df087262ae8960a31067e88cd80058"
+version = "0.71.0"
+source = "git+https://github.com/bytecodealliance/wasmtime/?branch=main#e41d88214455987751997b9d6173c5dee93204da"
 dependencies = [
  "anyhow",
  "cranelift-codegen",
diff --git a/build.sh b/build.sh
index 090349e54..76bc18843 100755
--- a/build.sh
+++ b/build.sh
@@ -55,6 +55,7 @@ ln target/$CHANNEL/*rustc_codegen_cranelift* "$target_dir"/lib
 ln rust-toolchain scripts/config.sh scripts/cargo.sh "$target_dir"
 
 mkdir -p "$target_dir/lib/rustlib/$TARGET_TRIPLE/lib/"
+mkdir -p "$target_dir/lib/rustlib/$HOST_TRIPLE/lib/"
 if [[ "$TARGET_TRIPLE" == "x86_64-pc-windows-gnu" ]]; then
     cp $(rustc --print sysroot)/lib/rustlib/$TARGET_TRIPLE/lib/*.o "$target_dir/lib/rustlib/$TARGET_TRIPLE/lib/"
 fi
@@ -64,12 +65,18 @@ case "$build_sysroot" in
         ;;
     "llvm")
         cp -r $(rustc --print sysroot)/lib/rustlib/$TARGET_TRIPLE/lib "$target_dir/lib/rustlib/$TARGET_TRIPLE/"
+        if [[ "$HOST_TRIPLE" != "$TARGET_TRIPLE" ]]; then
+            cp -r $(rustc --print sysroot)/lib/rustlib/$HOST_TRIPLE/lib "$target_dir/lib/rustlib/$HOST_TRIPLE/"
+        fi
         ;;
     "clif")
         echo "[BUILD] sysroot"
         dir=$(pwd)
         cd "$target_dir"
         time "$dir/build_sysroot/build_sysroot.sh"
+        if [[ "$HOST_TRIPLE" != "$TARGET_TRIPLE" ]]; then
+            time TARGET_TRIPLE="$HOST_TRIPLE" "$dir/build_sysroot/build_sysroot.sh"
+        fi
         cp lib/rustlib/*/lib/libstd-* lib/
         ;;
     *)
diff --git a/example/mini_core.rs b/example/mini_core.rs
index 7c6d7fc10..c4834c804 100644
--- a/example/mini_core.rs
+++ b/example/mini_core.rs
@@ -621,6 +621,7 @@ struct PanicLocation {
 }
 
 #[no_mangle]
+#[cfg(not(windows))]
 pub fn get_tls() -> u8 {
     #[thread_local]
     static A: u8 = 42;
diff --git a/example/mini_core_hello_world.rs b/example/mini_core_hello_world.rs
index 237f4d11d..08ceaeb65 100644
--- a/example/mini_core_hello_world.rs
+++ b/example/mini_core_hello_world.rs
@@ -239,7 +239,7 @@ fn main() {
 
     assert_eq!(((|()| 42u8) as fn(()) -> u8)(()), 42);
 
-    #[cfg(not(jit))]
+    #[cfg(not(any(jit, windows)))]
     {
         extern {
             #[linkage = "extern_weak"]
@@ -292,7 +292,7 @@ fn main() {
 
     from_decimal_string();
 
-    #[cfg(not(jit))]
+    #[cfg(not(any(jit, windows)))]
     test_tls();
 
     #[cfg(all(not(jit), target_os = "linux"))]
diff --git a/example/std_example.rs b/example/std_example.rs
index 015bbdfed..221b512e3 100644
--- a/example/std_example.rs
+++ b/example/std_example.rs
@@ -16,7 +16,8 @@ fn main() {
     let mut stderr = stderr.lock();
 
     // FIXME support lazy jit when multi threading
-    #[cfg(not(lazy_jit))]
+    // FIXME support TLS on windows
+    #[cfg(not(any(lazy_jit, windows)))]
     std::thread::spawn(move || {
         println!("Hello from another thread!");
     });
diff --git a/scripts/tests.sh b/scripts/tests.sh
index f9a9fb091..3afcea8f0 100755
--- a/scripts/tests.sh
+++ b/scripts/tests.sh
@@ -71,14 +71,20 @@ function base_sysroot_tests() {
     echo "[AOT] mod_bench"
     $MY_RUSTC example/mod_bench.rs --crate-type bin --target "$TARGET_TRIPLE"
     $RUN_WRAPPER ./target/out/mod_bench
+}
 
+function extended_sysroot_tests() {
     pushd rand
-    rm -r ./target || true
-    ../build/cargo.sh test --workspace
+    cargo clean
+    if [[ "$HOST_TRIPLE" = "$TARGET_TRIPLE" ]]; then
+        echo "[TEST] rust-random/rand"
+        ../build/cargo.sh test --workspace
+    else
+        echo "[AOT] rust-random/rand"
+        ../build/cargo.sh build --workspace --target $TARGET_TRIPLE --tests
+    fi
     popd
-}
 
-function extended_sysroot_tests() {
     pushd simple-raytracer
     if [[ "$HOST_TRIPLE" = "$TARGET_TRIPLE" ]]; then
         echo "[BENCH COMPILE] ebobby/simple-raytracer"
@@ -92,27 +98,40 @@ function extended_sysroot_tests() {
     else
         echo "[BENCH COMPILE] ebobby/simple-raytracer (skipped)"
         echo "[COMPILE] ebobby/simple-raytracer"
-        ../cargo.sh build
+        ../build/cargo.sh build --target $TARGET_TRIPLE
         echo "[BENCH RUN] ebobby/simple-raytracer (skipped)"
     fi
     popd
 
     pushd build_sysroot/sysroot_src/library/core/tests
     echo "[TEST] libcore"
-    rm -r ./target || true
-    ../../../../../build/cargo.sh test
+    cargo clean
+    if [[ "$HOST_TRIPLE" = "$TARGET_TRIPLE" ]]; then
+        ../../../../../build/cargo.sh test
+    else
+        ../../../../../build/cargo.sh build --target $TARGET_TRIPLE --tests
+    fi
     popd
 
     pushd regex
     echo "[TEST] rust-lang/regex example shootout-regex-dna"
-    ../build/cargo.sh clean
+    cargo clean
     # Make sure `[codegen mono items] start` doesn't poison the diff
-    ../build/cargo.sh build --example shootout-regex-dna
-    cat examples/regexdna-input.txt | ../build/cargo.sh run --example shootout-regex-dna | grep -v "Spawned thread" > res.txt
-    diff -u res.txt examples/regexdna-output.txt
+    ../build/cargo.sh build --example shootout-regex-dna --target $TARGET_TRIPLE
+    if [[ "$HOST_TRIPLE" = "$TARGET_TRIPLE" ]]; then
+        cat examples/regexdna-input.txt \
+            | ../build/cargo.sh run --example shootout-regex-dna --target $TARGET_TRIPLE \
+            | grep -v "Spawned thread" > res.txt
+        diff -u res.txt examples/regexdna-output.txt
+    fi
 
-    echo "[TEST] rust-lang/regex tests"
-    ../build/cargo.sh test --tests -- --exclude-should-panic --test-threads 1 -Zunstable-options -q
+    if [[ "$HOST_TRIPLE" = "$TARGET_TRIPLE" ]]; then
+        echo "[TEST] rust-lang/regex tests"
+        ../build/cargo.sh test --tests -- --exclude-should-panic --test-threads 1 -Zunstable-options -q
+    else
+        echo "[AOT] rust-lang/regex tests"
+        ../build/cargo.sh build --tests --target $TARGET_TRIPLE
+    fi
     popd
 }
 
diff --git a/src/codegen_i128.rs b/src/codegen_i128.rs
index ae75e6508..ffe1922ab 100644
--- a/src/codegen_i128.rs
+++ b/src/codegen_i128.rs
@@ -32,18 +32,56 @@ pub(crate) fn maybe_codegen<'tcx>(
         BinOp::Add | BinOp::Sub if !checked => None,
         BinOp::Mul if !checked => {
             let val_ty = if is_signed { fx.tcx.types.i128 } else { fx.tcx.types.u128 };
-            Some(fx.easy_call("__multi3", &[lhs, rhs], val_ty))
+            if fx.tcx.sess.target.is_like_windows {
+                let ret_place = CPlace::new_stack_slot(fx, lhs.layout());
+                let (lhs_ptr, lhs_extra) = lhs.force_stack(fx);
+                let (rhs_ptr, rhs_extra) = rhs.force_stack(fx);
+                assert!(lhs_extra.is_none());
+                assert!(rhs_extra.is_none());
+                let args =
+                    [ret_place.to_ptr().get_addr(fx), lhs_ptr.get_addr(fx), rhs_ptr.get_addr(fx)];
+                fx.lib_call(
+                    "__multi3",
+                    vec![
+                        AbiParam::special(pointer_ty(fx.tcx), ArgumentPurpose::StructReturn),
+                        AbiParam::new(pointer_ty(fx.tcx)),
+                        AbiParam::new(pointer_ty(fx.tcx)),
+                    ],
+                    vec![],
+                    &args,
+                );
+                Some(ret_place.to_cvalue(fx))
+            } else {
+                Some(fx.easy_call("__multi3", &[lhs, rhs], val_ty))
+            }
         }
         BinOp::Add | BinOp::Sub | BinOp::Mul => {
             assert!(checked);
             let out_ty = fx.tcx.mk_tup([lhs.layout().ty, fx.tcx.types.bool].iter());
             let out_place = CPlace::new_stack_slot(fx, fx.layout_of(out_ty));
-            let param_types = vec![
-                AbiParam::special(pointer_ty(fx.tcx), ArgumentPurpose::StructReturn),
-                AbiParam::new(types::I128),
-                AbiParam::new(types::I128),
-            ];
-            let args = [out_place.to_ptr().get_addr(fx), lhs.load_scalar(fx), rhs.load_scalar(fx)];
+            let (param_types, args) = if fx.tcx.sess.target.is_like_windows {
+                let (lhs_ptr, lhs_extra) = lhs.force_stack(fx);
+                let (rhs_ptr, rhs_extra) = rhs.force_stack(fx);
+                assert!(lhs_extra.is_none());
+                assert!(rhs_extra.is_none());
+                (
+                    vec![
+                        AbiParam::special(pointer_ty(fx.tcx), ArgumentPurpose::StructReturn),
+                        AbiParam::new(pointer_ty(fx.tcx)),
+                        AbiParam::new(pointer_ty(fx.tcx)),
+                    ],
+                    [out_place.to_ptr().get_addr(fx), lhs_ptr.get_addr(fx), rhs_ptr.get_addr(fx)],
+                )
+            } else {
+                (
+                    vec![
+                        AbiParam::special(pointer_ty(fx.tcx), ArgumentPurpose::StructReturn),
+                        AbiParam::new(types::I128),
+                        AbiParam::new(types::I128),
+                    ],
+                    [out_place.to_ptr().get_addr(fx), lhs.load_scalar(fx), rhs.load_scalar(fx)],
+                )
+            };
             let name = match (bin_op, is_signed) {
                 (BinOp::Add, false) => "__rust_u128_addo",
                 (BinOp::Add, true) => "__rust_i128_addo",
@@ -57,20 +95,33 @@ pub(crate) fn maybe_codegen<'tcx>(
             Some(out_place.to_cvalue(fx))
         }
         BinOp::Offset => unreachable!("offset should only be used on pointers, not 128bit ints"),
-        BinOp::Div => {
+        BinOp::Div | BinOp::Rem => {
             assert!(!checked);
-            if is_signed {
-                Some(fx.easy_call("__divti3", &[lhs, rhs], fx.tcx.types.i128))
-            } else {
-                Some(fx.easy_call("__udivti3", &[lhs, rhs], fx.tcx.types.u128))
-            }
-        }
-        BinOp::Rem => {
-            assert!(!checked);
-            if is_signed {
-                Some(fx.easy_call("__modti3", &[lhs, rhs], fx.tcx.types.i128))
+            let name = match (bin_op, is_signed) {
+                (BinOp::Div, false) => "__udivti3",
+                (BinOp::Div, true) => "__divti3",
+                (BinOp::Rem, false) => "__umodti3",
+                (BinOp::Rem, true) => "__modti3",
+                _ => unreachable!(),
+            };
+            if fx.tcx.sess.target.is_like_windows {
+                let (lhs_ptr, lhs_extra) = lhs.force_stack(fx);
+                let (rhs_ptr, rhs_extra) = rhs.force_stack(fx);
+                assert!(lhs_extra.is_none());
+                assert!(rhs_extra.is_none());
+                let args = [lhs_ptr.get_addr(fx), rhs_ptr.get_addr(fx)];
+                let ret = fx.lib_call(
+                    name,
+                    vec![AbiParam::new(pointer_ty(fx.tcx)), AbiParam::new(pointer_ty(fx.tcx))],
+                    vec![AbiParam::new(types::I64X2)],
+                    &args,
+                )[0];
+                // FIXME use bitcast instead of store to get from i64x2 to i128
+                let ret_place = CPlace::new_stack_slot(fx, lhs.layout());
+                ret_place.to_ptr().store(fx, ret, MemFlags::trusted());
+                Some(ret_place.to_cvalue(fx))
             } else {
-                Some(fx.easy_call("__umodti3", &[lhs, rhs], fx.tcx.types.u128))
+                Some(fx.easy_call(name, &[lhs, rhs], lhs.layout().ty))
             }
         }
         BinOp::Lt | BinOp::Le | BinOp::Eq | BinOp::Ge | BinOp::Gt | BinOp::Ne => {
diff --git a/src/inline_asm.rs b/src/inline_asm.rs
index 5b3df2bd3..33234f820 100644
--- a/src/inline_asm.rs
+++ b/src/inline_asm.rs
@@ -20,6 +20,10 @@ pub(crate) fn codegen_inline_asm<'tcx>(
     if template.is_empty() {
         // Black box
         return;
+    } else if template[0] == InlineAsmTemplatePiece::String("int $$0x29".to_string()) {
+        let true_ = fx.bcx.ins().iconst(types::I32, 1);
+        fx.bcx.ins().trapnz(true_, TrapCode::User(1));
+        return;
     }
 
     let mut slot_size = Size::from_bytes(0);
diff --git a/src/lib.rs b/src/lib.rs
index e1927ad3a..5db249970 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -313,6 +313,8 @@ fn build_isa(sess: &Session) -> Box<dyn isa::TargetIsa + 'static> {
 
     flags_builder.set("enable_simd", "true").unwrap();
 
+    flags_builder.set("enable_llvm_abi_extensions", "true").unwrap();
+
     use rustc_session::config::OptLevel;
     match sess.opts.optimize {
         OptLevel::No => {