[SIMD] Add extend + mul simd operations to interpreter (#4021)

2 files changed, 125 insertions, 13 deletions

diff --git a/src/wasm/literal.cpp b/src/wasm/literal.cpp
index 5a8ec3df3..7e36097c2 100644
--- a/src/wasm/literal.cpp
+++ b/src/wasm/literal.cpp
@@ -2400,41 +2400,54 @@ Literal Literal::extendHighUToVecI64x2() const {
   return extend<2, uint32_t, uint64_t, LaneOrder::High>(*this);
 }
 
+template<size_t Lanes, typename LaneFrom, typename LaneTo, LaneOrder Side>
+Literal extMul(const Literal& a, const Literal& b) {
+  LaneArray<Lanes* 2> lhs = getLanes<LaneFrom, Lanes * 2>(a);
+  LaneArray<Lanes* 2> rhs = getLanes<LaneFrom, Lanes * 2>(b);
+  LaneArray<Lanes> result;
+  for (size_t i = 0; i < Lanes; ++i) {
+    size_t idx = (Side == LaneOrder::Low) ? i : i + Lanes;
+    result[i] = Literal((LaneTo)(LaneFrom)lhs[idx].geti32() *
+                        (LaneTo)(LaneFrom)rhs[idx].geti32());
+  }
+  return Literal(result);
+}
+
 Literal Literal::extMulLowSI16x8(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<8, int8_t, int16_t, LaneOrder::Low>(*this, other);
 }
 Literal Literal::extMulHighSI16x8(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<8, int8_t, int16_t, LaneOrder::High>(*this, other);
 }
 Literal Literal::extMulLowUI16x8(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<8, uint8_t, uint16_t, LaneOrder::Low>(*this, other);
 }
 Literal Literal::extMulHighUI16x8(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<8, uint8_t, uint16_t, LaneOrder::High>(*this, other);
 }
 Literal Literal::extMulLowSI32x4(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<4, int16_t, int32_t, LaneOrder::Low>(*this, other);
 }
 Literal Literal::extMulHighSI32x4(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<4, int16_t, int32_t, LaneOrder::High>(*this, other);
 }
 Literal Literal::extMulLowUI32x4(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<4, uint16_t, uint32_t, LaneOrder::Low>(*this, other);
 }
 Literal Literal::extMulHighUI32x4(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<4, uint16_t, uint32_t, LaneOrder::High>(*this, other);
 }
 Literal Literal::extMulLowSI64x2(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<2, int32_t, int64_t, LaneOrder::Low>(*this, other);
 }
 Literal Literal::extMulHighSI64x2(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<2, int32_t, int64_t, LaneOrder::High>(*this, other);
 }
 Literal Literal::extMulLowUI64x2(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<2, uint32_t, uint64_t, LaneOrder::Low>(*this, other);
 }
 Literal Literal::extMulHighUI64x2(const Literal& other) const {
-  WASM_UNREACHABLE("TODO: implement SIMD extending multiplications");
+  return extMul<2, uint32_t, uint64_t, LaneOrder::High>(*this, other);
 }
 
 Literal Literal::swizzleVec8x16(const Literal& other) const {
diff --git a/test/spec/simd.wast b/test/spec/simd.wast
index 31c44b453..aa9ed48ba 100644
--- a/test/spec/simd.wast
+++ b/test/spec/simd.wast
@@ -152,7 +152,18 @@
  (func (export "i16x8.max_u") (param $0 v128) (param $1 v128) (result v128) (i16x8.max_u (local.get $0) (local.get $1)))
  (func (export "i16x8.avgr_u") (param $0 v128) (param $1 v128) (result v128) (i16x8.avgr_u (local.get $0) (local.get $1)))
  (func (export "i16x8.q15mulr_sat_s") (param $0 v128) (param $1 v128) (result v128) (i16x8.q15mulr_sat_s (local.get $0) (local.get $1)))
- ;; TODO: extending multiplications
+ (func (export "i16x8.extmul_low_i8x16_s") (param $0 v128) (param $1 v128) (result v128) (i16x8.extmul_low_i8x16_s (local.get $0) (local.get $1)))
+ (func (export "i16x8.extmul_high_i8x16_s") (param $0 v128) (param $1 v128) (result v128) (i16x8.extmul_high_i8x16_s (local.get $0) (local.get $1)))
+ (func (export "i16x8.extmul_low_i8x16_u") (param $0 v128) (param $1 v128) (result v128) (i16x8.extmul_low_i8x16_u (local.get $0) (local.get $1)))
+ (func (export "i16x8.extmul_high_i8x16_u") (param $0 v128) (param $1 v128) (result v128) (i16x8.extmul_high_i8x16_u (local.get $0) (local.get $1)))
+ (func (export "i32x4.extmul_low_i16x8_s") (param $0 v128) (param $1 v128) (result v128) (i32x4.extmul_low_i16x8_s (local.get $0) (local.get $1)))
+ (func (export "i32x4.extmul_high_i16x8_s") (param $0 v128) (param $1 v128) (result v128) (i32x4.extmul_high_i16x8_s (local.get $0) (local.get $1)))
+ (func (export "i32x4.extmul_low_i16x8_u") (param $0 v128) (param $1 v128) (result v128) (i32x4.extmul_low_i16x8_u (local.get $0) (local.get $1)))
+ (func (export "i32x4.extmul_high_i16x8_u") (param $0 v128) (param $1 v128) (result v128) (i32x4.extmul_high_i16x8_u (local.get $0) (local.get $1)))
+ (func (export "i64x2.extmul_low_i32x4_s") (param $0 v128) (param $1 v128) (result v128) (i64x2.extmul_low_i32x4_s (local.get $0) (local.get $1)))
+ (func (export "i64x2.extmul_high_i32x4_s") (param $0 v128) (param $1 v128) (result v128) (i64x2.extmul_high_i32x4_s (local.get $0) (local.get $1)))
+ (func (export "i64x2.extmul_low_i32x4_u") (param $0 v128) (param $1 v128) (result v128) (i64x2.extmul_low_i32x4_u (local.get $0) (local.get $1)))
+ (func (export "i64x2.extmul_high_i32x4_u") (param $0 v128) (param $1 v128) (result v128) (i64x2.extmul_high_i32x4_u (local.get $0) (local.get $1)))
  (func (export "i32x4.abs") (param $0 v128) (result v128) (i32x4.abs (local.get $0)))
  (func (export "i32x4.neg") (param $0 v128) (result v128) (i32x4.neg (local.get $0)))
  (func (export "i32x4.all_true") (param $0 v128) (result i32) (i32x4.all_true (local.get $0)))
@@ -791,6 +802,94 @@
   (v128.const i16x8 0 8192 1 1 32767 0 8193 1)
 )
 
+
+(assert_return
+  (invoke "i16x8.extmul_low_i8x16_s"
+    (v128.const i8x16 63 -63 128 -127 -128 -128 255 255 0 0 0 0 0 0 0 0)
+    (v128.const i8x16 64 -64   1   -1   -1 -127   1 255 0 0 0 0 0 0 0 0)
+  )
+  (v128.const i16x8 4032 4032 -128 127 128 16256 -1 1)
+)
+(assert_return
+  (invoke "i16x8.extmul_high_i8x16_s"
+    (v128.const i8x16 0 0 0 0 0 0 0 0 63 -63 128 -127 -128 -128 255 255)
+    (v128.const i8x16 0 0 0 0 0 0 0 0 64 -64   1   -1   -1 -127   1 255)
+  )
+  (v128.const i16x8 4032 4032 -128 127 128 16256 -1 1)
+)
+(assert_return
+  (invoke "i16x8.extmul_low_i8x16_u"
+    (v128.const i8x16  1 -1 63 -65 -126 -128  255 255 0 0 0 0 0 0 0 0)
+    (v128.const i8x16 -1 -1 64 -64 -1   -128 -128 255 0 0 0 0 0 0 0 0)
+  )
+  (v128.const i16x8 255 -511 4032 -28864 -32386 16384 32640 -511)
+)
+(assert_return
+  (invoke "i16x8.extmul_high_i8x16_u"
+    (v128.const i8x16 0 0 0 0 0 0 0 0  1 -1 63 -65 -126 -128  255 255)
+    (v128.const i8x16 0 0 0 0 0 0 0 0 -1 -1 64 -64 -1   -128 -128 255)
+  )
+  (v128.const i16x8 255 -511 4032 -28864 -32386 16384 32640 -511)
+)
+
+(assert_return
+  (invoke "i32x4.extmul_low_i16x8_s"
+    (v128.const i16x8  1 16384 -32766 65535 0 0 0 0)
+    (v128.const i16x8 -1 16384     -1    -1 0 0 0 0)
+  )
+  (v128.const i32x4 -1 268435456 32766 1)
+)
+(assert_return
+  (invoke "i32x4.extmul_high_i16x8_s"
+    (v128.const i16x8 0 0 0 0 16383 32768 -32766  65535)
+    (v128.const i16x8 0 0 0 0 16384     1     -1 -32768)
+  )
+  (v128.const i32x4 268419072 -32768 32766 32768)
+)
+(assert_return
+  (invoke "i32x4.extmul_low_i16x8_u"
+    (v128.const i16x8 16383 -16383 -32768 65535 0 0 0 0)
+    (v128.const i16x8 16384 -16384     -1 65535 0 0 0 0)
+  )
+  (v128.const i32x4 268419072 -1878999040 2147450880 -131071)
+)
+(assert_return
+  (invoke "i32x4.extmul_high_i16x8_u"
+    (v128.const i16x8 0 0 0 0 16383 -16383 -32768 65535)
+    (v128.const i16x8 0 0 0 0 16384 -16384     -1 65535)
+  )
+  (v128.const i32x4 268419072 -1878999040 2147450880 -131071)
+)
+
+(assert_return
+  (invoke "i64x2.extmul_low_i32x4_s"
+    (v128.const i32x4 1073741823 -2147483648 0 0)
+    (v128.const i32x4 1073741824          -1 0 0)
+  )
+  (v128.const i64x2 1152921503533105152 2147483648)
+)
+(assert_return
+  (invoke "i64x2.extmul_high_i32x4_s"
+    (v128.const i32x4 0 0 -1073741825 4294967295)
+    (v128.const i32x4 0 0 -1073741824 4294967295)
+  )
+  (v128.const i64x2 1152921505680588800 1)
+)
+(assert_return
+  (invoke "i64x2.extmul_low_i32x4_u"
+    (v128.const i32x4 -1 -1073741825 0 0)
+    (v128.const i32x4 -1 -1073741824 0 0)
+  )
+  (v128.const i64x2 -8589934591 -8070450535469154304)
+)
+(assert_return
+  (invoke "i64x2.extmul_high_i32x4_u"
+    (v128.const i32x4 0 0 -2147483648 4294967295)
+    (v128.const i32x4 0 0          -1 4294967295)
+  )
+  (v128.const i64x2 9223372034707292160 -8589934591)
+)
+
 ;; i32x4 arithmetic
 (assert_return (invoke "i32x4.abs" (v128.const i32x4 0 1 0x80000000 0x80000001)) (v128.const i32x4 0 1 0x80000000 0x7fffffff))
 (assert_return (invoke "i32x4.neg" (v128.const i32x4 0 1 0x80000000 0x80000001)) (v128.const i32x4 0 -1 0x80000000 0x7fffffff))