[WASM SIMD] Implement horizontal add for float and integer types.

src/arm/assembler-arm.h

Index: src/arm/assembler-arm.h
diff --git a/src/arm/assembler-arm.h b/src/arm/assembler-arm.h
index f16aee6e715d33b4f4d00760c76e116372fbf210..48557e3124a57a5946ff683906f3573451a58fb8 100644
--- a/src/arm/assembler-arm.h
+++ b/src/arm/assembler-arm.h
@@ -77,7 +77,7 @@ namespace internal {
   V(d0)  V(d1)  V(d2)  V(d3)  V(d4)  V(d5)  V(d6)  V(d7)  \
   V(d8)  V(d9)  V(d10) V(d11) V(d12)                      \
   V(d16) V(d17) V(d18) V(d19) V(d20) V(d21) V(d22) V(d23) \
-  V(d24) V(d25) V(d26) V(d27) V(d28) V(d29) V(d30) V(d31)
+  V(d24) V(d25) V(d26) V(d27) V(d28) V(d29)

[georgia.kouveli, 2017/04/20 14:53:06]

Why make d30 and d31 non-allocatable?

[bbudge, 2017/04/21 20:18:58]

That's my mistake, from a local merge. I'll restore those.

 
 #define ALLOCATABLE_NO_VFP32_DOUBLE_REGISTERS(V)          \
   V(d0)  V(d1)  V(d2)  V(d3)  V(d4)  V(d5)  V(d6)  V(d7)  \
@@ -1371,6 +1371,9 @@ class Assembler : public AssemblerBase {
   void vmax(QwNeonRegister dst, QwNeonRegister src1, QwNeonRegister src2);
   void vmax(NeonDataType dt, QwNeonRegister dst,
             QwNeonRegister src1, QwNeonRegister src2);
+  void vpadd(DwVfpRegister dst, DwVfpRegister src1, DwVfpRegister src2);
+  void vpadd(NeonSize size, DwVfpRegister dst, DwVfpRegister src1,
+             DwVfpRegister src2);
   void vpmin(NeonDataType dt, DwVfpRegister dst, DwVfpRegister src1,
              DwVfpRegister src2);
   void vpmax(NeonDataType dt, DwVfpRegister dst, DwVfpRegister src1,