fix mixed marshalling of mixed float/double argument lists on armhf

When we skip a single-precision register to ensure a double-precision
load is aligned, we need to remember that in case we see another
single-precision argument later on, which we must backfill into that
register we skipped according to the ABI.

src/arm.h

@@ -147,6 +147,7 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
   const unsigned VfpCount = 16;
   uintptr_t vfpTable[VfpCount];
   unsigned vfpIndex = 0;
+  unsigned vfpBackfillIndex = 0;
 
   uintptr_t stack[(argumentCount * 8) / BytesPerWord]; // is > argumentSize to account for padding
   unsigned stackIndex = 0;
@@ -159,6 +160,7 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
       {
         if (vfpIndex + Alignment <= VfpCount) {
           if (vfpIndex % Alignment) {
+            vfpBackfillIndex = vfpIndex;
             ++ vfpIndex;
           }
           
@@ -177,7 +179,10 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
       } break;
 
     case FLOAT_TYPE:
-      if (vfpIndex < VfpCount) {
+      if (vfpBackfillIndex) {
+        vfpTable[vfpBackfillIndex] = arguments[ai];
+        vfpBackfillIndex = 0;
+      } else if (vfpIndex < VfpCount) {
         vfpTable[vfpIndex++] = arguments[ai];
       } else {
         stack[stackIndex++] = arguments[ai];

test/JNI.java

@@ -19,6 +19,12 @@ public class JNI {
      float a13, float a14, float a15, float a16, float a17, float a18,
      float a19, float a20);
 
+  private static native double addMix
+    (float a1, double a2, float a3, double a4, float a5, float a6,
+     float a7, float a8, float a9, float a10, float a11, float a12,
+     float a13, float a14, float a15, double a16, float a17, float a18,
+     float a19, float a20);
+
   public static void main(String[] args) {
     expect(addDoubles
            (1.0d, 2.0d, 3.0d, 4.0d, 5.0d, 6.0d, 7.0d, 8.0d, 9.0d, 10.0d, 11.0d,
@@ -29,5 +35,10 @@ public class JNI {
            (1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f,
             12.0f, 13.0f, 14.0f, 15.0f, 16.0f, 17.0f, 18.0f, 19.0f, 20.0f)
            == 210.0f);
+
+    expect(addMix
+           (1.0f, 2.0d, 3.0f, 4.0d, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f,
+            12.0f, 13.0f, 14.0f, 15.0f, 16.0d, 17.0f, 18.0f, 19.0f, 20.0f)
+           == 210.0d);
   }
 }

test/jni.cpp

@@ -24,3 +24,15 @@ Java_JNI_addFloats
   return a1 + a2 + a3 + a4 + a5 + a6 + a7 + a8 + a9 + a10 + a11 + a12 + a13
     + a14 + a15 + a16 + a17 + a18 + a19 + a20;
 }
+
+extern "C" JNIEXPORT jdouble JNICALL
+Java_JNI_addMix
+(JNIEnv*, jclass,
+ jfloat a1, jdouble a2, jfloat a3, jdouble a4, jfloat a5, jfloat a6,
+ jfloat a7, jfloat a8, jfloat a9, jfloat a10, jfloat a11, jfloat a12,
+ jfloat a13, jfloat a14, jfloat a15, jdouble a16, jfloat a17, jfloat a18,
+ jfloat a19, jfloat a20)
+{
+  return a1 + a2 + a3 + a4 + a5 + a6 + a7 + a8 + a9 + a10 + a11 + a12 + a13
+    + a14 + a15 + a16 + a17 + a18 + a19 + a20;
+}