[simd.js] Add SIMD store functions for Phase 1.

Float32x4, Int32x4, Uint32x4:
  store, store1, store2, store3

Int16x8, Int8x16, Uint16x8, Uint8x16:
  store

BUG=v8:4124
LOG=N

R=bbudge@chromium.org, littledan@chromium.org, jarin@chromium.org

Review URL: https://codereview.chromium.org/1304183004

Cr-Commit-Position: refs/heads/master@{#30429}

diff --git a/src/harmony-simd.js b/src/harmony-simd.js
index ee58724..9f8d916 100644 (file)
--- a/src/harmony-simd.js
+++ b/src/harmony-simd.js
@@ -247,6 +247,10 @@ function NAMEGreaterThanOrEqualJS(a, b) {
 function NAMELoadJS(tarray, index) {
   return %NAMELoad(tarray, index);
 }
+
+function NAMEStoreJS(tarray, index, a) {
+  return %NAMEStore(tarray, index, a);
+}
 endmacro
 
 SIMD_NUMERIC_TYPES(DECLARE_NUMERIC_FUNCTIONS)
@@ -368,6 +372,10 @@ macro DECLARE_LOADN_STOREN_FUNCTIONS(NAME, COUNT)
 function NAMELoadCOUNTJS(tarray, index) {
   return %NAMELoadCOUNT(tarray, index);
 }
+
+function NAMEStoreCOUNTJS(tarray, index, a) {
+  return %NAMEStoreCOUNT(tarray, index, a);
+}
 endmacro
 
 SIMD_LOADN_STOREN_TYPES(DECLARE_LOADN_STOREN_FUNCTIONS)
@@ -628,6 +636,10 @@ utils.InstallFunctions(GlobalFloat32x4, DONT_ENUM, [
   'load1', Float32x4Load1JS,
   'load2', Float32x4Load2JS,
   'load3', Float32x4Load3JS,
+  'store', Float32x4StoreJS,
+  'store1', Float32x4Store1JS,
+  'store2', Float32x4Store2JS,
+  'store3', Float32x4Store3JS,
 ]);
 
 utils.InstallFunctions(GlobalInt32x4, DONT_ENUM, [
@@ -668,6 +680,10 @@ utils.InstallFunctions(GlobalInt32x4, DONT_ENUM, [
   'load1', Int32x4Load1JS,
   'load2', Int32x4Load2JS,
   'load3', Int32x4Load3JS,
+  'store', Int32x4StoreJS,
+  'store1', Int32x4Store1JS,
+  'store2', Int32x4Store2JS,
+  'store3', Int32x4Store3JS,
 ]);
 
 utils.InstallFunctions(GlobalUint32x4, DONT_ENUM, [
@@ -708,6 +724,10 @@ utils.InstallFunctions(GlobalUint32x4, DONT_ENUM, [
   'load1', Uint32x4Load1JS,
   'load2', Uint32x4Load2JS,
   'load3', Uint32x4Load3JS,
+  'store', Uint32x4StoreJS,
+  'store1', Uint32x4Store1JS,
+  'store2', Uint32x4Store2JS,
+  'store3', Uint32x4Store3JS,
 ]);
 
 utils.InstallFunctions(GlobalBool32x4, DONT_ENUM, [
@@ -761,6 +781,7 @@ utils.InstallFunctions(GlobalInt16x8, DONT_ENUM, [
   'fromInt8x16Bits', Int16x8FromInt8x16BitsJS,
   'fromUint8x16Bits', Int16x8FromUint8x16BitsJS,
   'load', Int16x8LoadJS,
+  'store', Int16x8StoreJS,
 ]);
 
 utils.InstallFunctions(GlobalUint16x8, DONT_ENUM, [
@@ -801,6 +822,7 @@ utils.InstallFunctions(GlobalUint16x8, DONT_ENUM, [
   'fromInt8x16Bits', Uint16x8FromInt8x16BitsJS,
   'fromUint8x16Bits', Uint16x8FromUint8x16BitsJS,
   'load', Uint16x8LoadJS,
+  'store', Uint16x8StoreJS,
 ]);
 
 utils.InstallFunctions(GlobalBool16x8, DONT_ENUM, [
@@ -854,6 +876,7 @@ utils.InstallFunctions(GlobalInt8x16, DONT_ENUM, [
   'fromUint16x8Bits', Int8x16FromUint16x8BitsJS,
   'fromUint8x16Bits', Int8x16FromUint8x16BitsJS,
   'load', Int8x16LoadJS,
+  'store', Int8x16StoreJS,
 ]);
 
 utils.InstallFunctions(GlobalUint8x16, DONT_ENUM, [
@@ -894,6 +917,7 @@ utils.InstallFunctions(GlobalUint8x16, DONT_ENUM, [
   'fromUint16x8Bits', Uint8x16FromUint16x8BitsJS,
   'fromInt8x16Bits', Uint8x16FromInt8x16BitsJS,
   'load', Uint8x16LoadJS,
+  'store', Uint8x16StoreJS,
 ]);
 
 utils.InstallFunctions(GlobalBool8x16, DONT_ENUM, [

diff --git a/src/runtime/runtime-simd.cc b/src/runtime/runtime-simd.cc
index c9ec4e1..7ec9737 100644 (file)
--- a/src/runtime/runtime-simd.cc
+++ b/src/runtime/runtime-simd.cc
@@ -960,14 +960,16 @@ SIMD_FROM_BITS_TYPES(SIMD_FROM_BITS_FUNCTION)
 
 //-------------------------------------------------------------------
 
-// Load functions.
+// Load and Store functions.
+
 #define SIMD_LOADN_STOREN_TYPES(FUNCTION) \
   FUNCTION(Float32x4, float, 4)           \
   FUNCTION(Int32x4, int32_t, 4)           \
   FUNCTION(Uint32x4, uint32_t, 4)
 
 
-// Common Load Functions
+// Common Load and Store Functions
+
 #define SIMD_LOAD(type, lane_type, lane_count, count, result)          \
   static const int kLaneCount = lane_count;                            \
   DCHECK(args.length() == 2);                                          \
@@ -986,6 +988,27 @@ SIMD_FROM_BITS_TYPES(SIMD_FROM_BITS_FUNCTION)
   Handle<type> result = isolate->factory()->New##type(lanes);
 
 
+#define SIMD_STORE(type, lane_type, lane_count, count, a)              \
+  static const int kLaneCount = lane_count;                            \
+  DCHECK(args.length() == 3);                                          \
+  CONVERT_ARG_HANDLE_CHECKED(JSTypedArray, tarray, 0);                 \
+  CONVERT_INT32_ARG_CHECKED(index, 1)                                  \
+  CONVERT_ARG_HANDLE_CHECKED(type, a, 2);                              \
+  size_t bpe = tarray->element_size();                                 \
+  uint32_t bytes = count * sizeof(lane_type);                          \
+  size_t byte_length = NumberToSize(isolate, tarray->byte_length());   \
+  RUNTIME_ASSERT(index >= 0 && index * bpe + bytes <= byte_length);    \
+  size_t tarray_offset = NumberToSize(isolate, tarray->byte_offset()); \
+  uint8_t* tarray_base =                                               \
+      static_cast<uint8_t*>(tarray->GetBuffer()->backing_store()) +    \
+      tarray_offset;                                                   \
+  lane_type lanes[kLaneCount];                                         \
+  for (int i = 0; i < kLaneCount; i++) {                               \
+    lanes[i] = a->get_lane(i);                                         \
+  }                                                                    \
+  memcpy(tarray_base + index * bpe, lanes, bytes);
+
+
 #define SIMD_LOAD_FUNCTION(type, lane_type, lane_count)         \
   RUNTIME_FUNCTION(Runtime_##type##Load) {                      \
     HandleScope scope(isolate);                                 \
@@ -1018,10 +1041,46 @@ SIMD_FROM_BITS_TYPES(SIMD_FROM_BITS_FUNCTION)
   }
 
 
+#define SIMD_STORE_FUNCTION(type, lane_type, lane_count)    \
+  RUNTIME_FUNCTION(Runtime_##type##Store) {                 \
+    HandleScope scope(isolate);                             \
+    SIMD_STORE(type, lane_type, lane_count, lane_count, a); \
+    return *a;                                              \
+  }
+
+
+#define SIMD_STORE1_FUNCTION(type, lane_type, lane_count) \
+  RUNTIME_FUNCTION(Runtime_##type##Store1) {              \
+    HandleScope scope(isolate);                           \
+    SIMD_STORE(type, lane_type, lane_count, 1, a);        \
+    return *a;                                            \
+  }
+
+
+#define SIMD_STORE2_FUNCTION(type, lane_type, lane_count) \
+  RUNTIME_FUNCTION(Runtime_##type##Store2) {              \
+    HandleScope scope(isolate);                           \
+    SIMD_STORE(type, lane_type, lane_count, 2, a);        \
+    return *a;                                            \
+  }
+
+
+#define SIMD_STORE3_FUNCTION(type, lane_type, lane_count) \
+  RUNTIME_FUNCTION(Runtime_##type##Store3) {              \
+    HandleScope scope(isolate);                           \
+    SIMD_STORE(type, lane_type, lane_count, 3, a);        \
+    return *a;                                            \
+  }
+
+
 SIMD_NUMERIC_TYPES(SIMD_LOAD_FUNCTION)
 SIMD_LOADN_STOREN_TYPES(SIMD_LOAD1_FUNCTION)
 SIMD_LOADN_STOREN_TYPES(SIMD_LOAD2_FUNCTION)
 SIMD_LOADN_STOREN_TYPES(SIMD_LOAD3_FUNCTION)
+SIMD_NUMERIC_TYPES(SIMD_STORE_FUNCTION)
+SIMD_LOADN_STOREN_TYPES(SIMD_STORE1_FUNCTION)
+SIMD_LOADN_STOREN_TYPES(SIMD_STORE2_FUNCTION)
+SIMD_LOADN_STOREN_TYPES(SIMD_STORE3_FUNCTION)
 
 //-------------------------------------------------------------------
 

diff --git a/src/runtime/runtime.h b/src/runtime/runtime.h
index 7a43c84..18faeb8 100644 (file)
--- a/src/runtime/runtime.h
+++ b/src/runtime/runtime.h
@@ -615,6 +615,10 @@ namespace internal {
   F(Float32x4Load1, 2, 1)                      \
   F(Float32x4Load2, 2, 1)                      \
   F(Float32x4Load3, 2, 1)                      \
+  F(Float32x4Store, 3, 1)                      \
+  F(Float32x4Store1, 3, 1)                     \
+  F(Float32x4Store2, 3, 1)                     \
+  F(Float32x4Store3, 3, 1)                     \
   F(Int32x4Check, 1, 1)                        \
   F(Int32x4ExtractLane, 2, 1)                  \
   F(Int32x4ReplaceLane, 3, 1)                  \
@@ -651,6 +655,10 @@ namespace internal {
   F(Int32x4Load1, 2, 1)                        \
   F(Int32x4Load2, 2, 1)                        \
   F(Int32x4Load3, 2, 1)                        \
+  F(Int32x4Store, 3, 1)                        \
+  F(Int32x4Store1, 3, 1)                       \
+  F(Int32x4Store2, 3, 1)                       \
+  F(Int32x4Store3, 3, 1)                       \
   F(Uint32x4Check, 1, 1)                       \
   F(Uint32x4ExtractLane, 2, 1)                 \
   F(Uint32x4ReplaceLane, 3, 1)                 \
@@ -687,6 +695,10 @@ namespace internal {
   F(Uint32x4Load1, 2, 1)                       \
   F(Uint32x4Load2, 2, 1)                       \
   F(Uint32x4Load3, 2, 1)                       \
+  F(Uint32x4Store, 3, 1)                       \
+  F(Uint32x4Store1, 3, 1)                      \
+  F(Uint32x4Store2, 3, 1)                      \
+  F(Uint32x4Store3, 3, 1)                      \
   F(Bool32x4Check, 1, 1)                       \
   F(Bool32x4ExtractLane, 2, 1)                 \
   F(Bool32x4ReplaceLane, 3, 1)                 \
@@ -732,6 +744,7 @@ namespace internal {
   F(Int16x8FromInt8x16Bits, 1, 1)              \
   F(Int16x8FromUint8x16Bits, 1, 1)             \
   F(Int16x8Load, 2, 1)                         \
+  F(Int16x8Store, 3, 1)                        \
   F(Uint16x8Check, 1, 1)                       \
   F(Uint16x8ExtractLane, 2, 1)                 \
   F(Uint16x8ReplaceLane, 3, 1)                 \
@@ -768,6 +781,7 @@ namespace internal {
   F(Uint16x8FromInt8x16Bits, 1, 1)             \
   F(Uint16x8FromUint8x16Bits, 1, 1)            \
   F(Uint16x8Load, 2, 1)                        \
+  F(Uint16x8Store, 3, 1)                       \
   F(Bool16x8Check, 1, 1)                       \
   F(Bool16x8ExtractLane, 2, 1)                 \
   F(Bool16x8ReplaceLane, 3, 1)                 \
@@ -813,6 +827,7 @@ namespace internal {
   F(Int8x16FromUint16x8Bits, 1, 1)             \
   F(Int8x16FromUint8x16Bits, 1, 1)             \
   F(Int8x16Load, 2, 1)                         \
+  F(Int8x16Store, 3, 1)                        \
   F(Uint8x16Check, 1, 1)                       \
   F(Uint8x16ExtractLane, 2, 1)                 \
   F(Uint8x16ReplaceLane, 3, 1)                 \
@@ -849,6 +864,7 @@ namespace internal {
   F(Uint8x16FromUint16x8Bits, 1, 1)            \
   F(Uint8x16FromInt8x16Bits, 1, 1)             \
   F(Uint8x16Load, 2, 1)                        \
+  F(Uint8x16Store, 3, 1)                       \
   F(Bool8x16Check, 1, 1)                       \
   F(Bool8x16ExtractLane, 2, 1)                 \
   F(Bool8x16ReplaceLane, 3, 1)                 \