1 files changed, 24 insertions, 0 deletions
diff --git a/js/src/jit-test/tests/wasm/simd/ion-bug1688713.js b/js/src/jit-test/tests/wasm/simd/ion-bug1688713.js
new file mode 100644
index 0000000000..86a2ff0b3c
--- /dev/null
+++ b/js/src/jit-test/tests/wasm/simd/ion-bug1688713.js
@@ -0,0 +1,24 @@
+// |jit-test| skip-if: !wasmSimdEnabled() || !hasDisassembler() || wasmCompileMode() != "ion" || !getBuildConfiguration().x64 || getBuildConfiguration().simulator || isAvxPresent(); include:codegen-x64-test.js
+
+// This checks that we emit a REX prefix that includes the SIB index when
+// appropriate.
+//
+// This test case is a little tricky.  On Win64, the arg registers are rcx, rdx,
+// r8, r9; so we want to use local 2 or 3 as the index.  But on other x64
+// platforms, the arg registers are rdi, rsi, rdx, rcx, r8, r9; so we want to
+// use local 4 or 5 as the index.  This test uses both, and then looks for a hit
+// on the REX byte which must be 0x43.  Before the bugfix, since the index
+// register was ignored, the byte would always be 0x41, as it will continue to
+// be for the access that does not use an extended register.
+//
+// The test is brittle: the register allocator can easily make a mess of it.
+// But for now it works.
+
+codegenTestX64_adhoc(
+`(module
+   (memory 1)
+   (func $f (export "f") (param i32) (param i32) (param i32) (param i32) (param i32) (result v128)
+     (i32x4.add (v128.load8x8_s (local.get 4)) (v128.load8x8_s (local.get 2)))))`,
+    'f',
+    `66 43 0f 38 20 .. ..      pmovsxbwq \\(%r15,%r(8|9|10|11|12|13),1\\), %xmm[0-9]+`,
+    {no_prefix: true, no_suffix: true, log:true});