diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
new file mode 100644
index 0000000..431400a
--- /dev/null
+++ b/.github/workflows/ci.yml
@@ -0,0 +1,97 @@
+name: CI
+
+on: [push, pull_request]
+
+jobs:
+  build:
+    name: ${{ matrix.target }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - target: amd64
+            flags: -O3 -fomit-frame-pointer
+          - target: x86
+            flags: -m32 -O3 -fomit-frame-pointer
+    steps:
+      - uses: actions/checkout@v3
+      - name: Prepare
+        run: |
+          sudo apt-get update -q -y
+          sudo apt-get install -y gcc-multilib g++-multilib valgrind
+      - name: Build
+        working-directory: tests
+        run: make
+      - name: Test
+        working-directory: tests
+        run: make test-all
+
+  build-qemu:
+    name: ${{ matrix.target }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - target: aarch64
+            arch: aarch64
+    steps:
+      - uses: actions/checkout@v3
+      - uses: uraimo/run-on-arch-action@v2
+        with:
+          arch: ${{ matrix.arch }}
+          distro: ubuntu_latest
+          install: |
+            apt-get update -q -y
+            apt-get install -q -y --no-install-recommends build-essential valgrind
+          env: |
+            target: ${{ matrix.target }}
+          run: |
+            cd tests
+            make
+            make test-all
+
+  build-windows:
+    name: Windows (${{ matrix.arch }})
+    runs-on: windows-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        arch: [amd64, x86]
+    steps:
+      - uses: ilammy/msvc-dev-cmd@v1
+        with:
+          arch: ${{ matrix.arch }}
+      - uses: actions/checkout@v3
+      - name: Build
+        working-directory: tests
+        run: |
+            cl /I .. testsuite.c
+            cl /I .. example.c
+            cl /I .. mt-example.c
+            cl /I .. simple.c
+            cl /I .. coro_timing.c
+      - name: Test
+        working-directory: tests
+        run: |
+            .\testsuite.exe
+            .\example.exe
+            .\mt-example.exe
+            .\simple.exe
+            .\coro_timing.exe
+
+  sanitizers:
+    name: Linux full - thread, undefined, leak, address
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+    steps:
+      - uses: actions/checkout@v3
+      - name: Prepare
+        run: |
+          sudo apt-get update -q -y
+          sudo apt-get install -y gcc-multilib g++-multilib valgrind
+      - name: Build and run sanitizer tests
+        working-directory: tests
+        run: make test-linux-full
diff --git a/tests/Makefile b/tests/Makefile
index 2aad113..1167905 100644
--- a/tests/Makefile
+++ b/tests/Makefile
@@ -26,7 +26,7 @@ WASMTIME=wasmtime
 # OUTEXT=.exe
 # RUNNER=
 
-all: testsuite example mt-example simple benchmark
+all: testsuite example mt-example simple benchmark coro_timing
 
 test: testsuite
 	$(RUNNER) ./testsuite$(OUTEXT)
@@ -34,11 +34,12 @@ test: testsuite
 test-example: example
 	$(RUNNER) ./example$(OUTEXT)
 
-test-all: testsuite example mt-example simple
+test-all: testsuite example mt-example simple coro_timing
 	$(RUNNER) ./testsuite$(OUTEXT)
 	$(RUNNER) ./example$(OUTEXT)
 	$(RUNNER) ./mt-example$(OUTEXT)
 	$(RUNNER) ./simple$(OUTEXT)
+	$(RUNNER) ./coro_timing$(OUTEXT)
 
 bench: benchmark
 	./benchmark$(OUTEXT)
@@ -46,6 +47,9 @@ bench: benchmark
 testsuite: testsuite.c ../minicoro.h Makefile
 	$(CC) testsuite.c -o testsuite $(EXTRA_CFLAGS) $(CFLAGS)
 
+coro_timing: coro_timing.c ../minicoro.h Makefile
+	$(CC) coro_timing.c -o coro_timing $(EXTRA_CFLAGS) $(CFLAGS)
+
 example: example.c ../minicoro.h Makefile
 	$(CC) example.c -o example $(EXTRA_CFLAGS) $(CFLAGS)
 
@@ -59,7 +63,7 @@ benchmark: benchmark.c ../minicoro.h Makefile
 	$(CC) benchmark.c -o benchmark $(EXTRA_CFLAGS) $(CFLAGS) $(RELEASE_CFLAGS) -std=c99
 
 clean:
-	rm -f testsuite$(OUTEXT) example$(OUTEXT) mt-example$(OUTEXT) simple$(OUTEXT) benchmark$(OUTEXT)
+	rm -f testsuite$(OUTEXT) example$(OUTEXT) mt-example$(OUTEXT) simple$(OUTEXT) benchmark$(OUTEXT) coro_timing$(OUTEXT)
 
 test-riscv64:
 	$(MAKE) --no-print-directory CC=riscv64-elf-gcc CFLAGS="-march=rv64gc -mabi=lp64d" RUNNER=qemu-riscv64 clean test test-example
diff --git a/tests/coro_timing.c b/tests/coro_timing.c
new file mode 100644
index 0000000..f6b7f1e
--- /dev/null
+++ b/tests/coro_timing.c
@@ -0,0 +1,65 @@
+#define MINICORO_IMPL
+#define MCO_NO_DEBUG 1
+#include "minicoro.h"
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+
+enum
+{
+    Iterations = 500000000
+};
+
+static int counter;
+
+void co_timingtest(mco_coro *co)
+{
+    for (;;)
+    {
+        counter++;
+        mco_yield(co);
+    }
+}
+
+void sub_timingtest()
+{
+    counter++;
+}
+
+int main()
+{
+    printf("context-switching timing test\n\n");
+    time_t start, end;
+    mco_coro *thread_y;
+    int i, t1, t2;
+
+    start = clock();
+    for (counter = 0, i = 0; i < Iterations; i++)
+    {
+        sub_timingtest();
+    }
+    end = clock();
+
+    t1 = (int)difftime(end, start);
+    printf("%2.3f seconds per  50 million subroutine calls (%d iterations)\n", (float)t1 / CLOCKS_PER_SEC, counter);
+
+    mco_desc desc = mco_desc_init(co_timingtest, 0);
+    desc.user_data = NULL;
+    mco_create(&thread_y, &desc);
+    mco_resume(thread_y);
+    start = clock();
+    for (counter = 0, i = 0; i < Iterations; i++)
+    {
+        mco_resume(thread_y);
+    }
+    end = clock();
+
+    mco_destroy(thread_y);
+
+    t2 = (int)difftime(end, start);
+    printf("%2.3f seconds per 100 million mco_resume calls (%d iterations)\n", (float)t2 / CLOCKS_PER_SEC, counter);
+
+    printf("mco_resume skew = %fx\n\n", (double)t2 / (double)t1);
+    return 0;
+}