Implements ARM SIMD multiplication and subtraction instructions.

runtime/vm/assembler_arm_test.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"
 #if defined(TARGET_ARCH_ARM)
 
 #include "vm/assembler.h"
 #include "vm/os.h"
 #include "vm/unit_test.h"
@@ skipping 1718 matching lines @@
 }
 
 
 ASSEMBLER_TEST_RUN(Vaddqi64, test) {
   EXPECT(test != NULL);
   typedef int (*Tst)();
   EXPECT_EQ(4, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
 }
 
 
+ASSEMBLER_TEST_GENERATE(Vsubqi8, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S1, R0);
+    __ vmovsr(S2, R0);
+    __ vmovsr(S3, R0);

[regis, 2013/07/16 00:23:30]

Being picky, but a test using a different value in each register would have a
broader coverage.

[zra, 2013/07/16 16:11:01]

Done.

+    __ mov(R0, ShifterOperand(3));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S5, R0);
+    __ vmovsr(S6, R0);
+    __ vmovsr(S7, R0);
+
+    __ vsubqi(0, Q2, Q1, Q0);

[regis, 2013/07/16 00:23:30]

Comment about the meaning of 0 until you introduce an enum?

[zra, 2013/07/16 16:11:01]

Done.

+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R1, S9);
+    __ vmovrs(R2, S10);
+    __ vmovrs(R3, S11);
+
+    __ add(R0, R0, ShifterOperand(R1));
+    __ add(R0, R0, ShifterOperand(R2));
+    __ add(R0, R0, ShifterOperand(R3));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 8);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vsubqi8, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vsubqi16, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S1, R0);
+    __ vmovsr(S2, R0);
+    __ vmovsr(S3, R0);
+    __ mov(R0, ShifterOperand(3));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S5, R0);
+    __ vmovsr(S6, R0);
+    __ vmovsr(S7, R0);
+
+    __ vsubqi(1, Q2, Q1, Q0);

[regis, 2013/07/16 00:23:30]

ditto for 1?

[zra, 2013/07/16 16:11:01]

Done.

+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R1, S9);
+    __ vmovrs(R2, S10);
+    __ vmovrs(R3, S11);
+
+    __ add(R0, R0, ShifterOperand(R1));
+    __ add(R0, R0, ShifterOperand(R2));
+    __ add(R0, R0, ShifterOperand(R3));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 8);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vsubqi16, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vsubqi32, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S1, R0);
+    __ vmovsr(S2, R0);
+    __ vmovsr(S3, R0);
+    __ mov(R0, ShifterOperand(3));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S5, R0);
+    __ vmovsr(S6, R0);
+    __ vmovsr(S7, R0);
+
+    __ vsubqi(2, Q2, Q1, Q0);
+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R1, S9);
+    __ vmovrs(R2, S10);
+    __ vmovrs(R3, S11);
+
+    __ add(R0, R0, ShifterOperand(R1));
+    __ add(R0, R0, ShifterOperand(R2));
+    __ add(R0, R0, ShifterOperand(R3));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 8);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vsubqi32, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vsubqi64, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S2, R0);
+    __ mov(R0, ShifterOperand(3));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S6, R0);
+
+    __ vsubqi(3, Q2, Q1, Q0);
+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R2, S10);
+
+    __ add(R0, R0, ShifterOperand(R2));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 4);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vsubqi64, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(4, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vmulqi8, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S1, R0);
+    __ vmovsr(S2, R0);
+    __ vmovsr(S3, R0);
+    __ mov(R0, ShifterOperand(2));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S5, R0);
+    __ vmovsr(S6, R0);
+    __ vmovsr(S7, R0);
+
+    __ vmulqi(0, Q2, Q1, Q0);
+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R1, S9);
+    __ vmovrs(R2, S10);
+    __ vmovrs(R3, S11);
+
+    __ add(R0, R0, ShifterOperand(R1));
+    __ add(R0, R0, ShifterOperand(R2));
+    __ add(R0, R0, ShifterOperand(R3));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 8);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vmulqi8, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vmulqi16, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S1, R0);
+    __ vmovsr(S2, R0);
+    __ vmovsr(S3, R0);
+    __ mov(R0, ShifterOperand(2));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S5, R0);
+    __ vmovsr(S6, R0);
+    __ vmovsr(S7, R0);
+
+    __ vmulqi(1, Q2, Q1, Q0);
+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R1, S9);
+    __ vmovrs(R2, S10);
+    __ vmovrs(R3, S11);
+
+    __ add(R0, R0, ShifterOperand(R1));
+    __ add(R0, R0, ShifterOperand(R2));
+    __ add(R0, R0, ShifterOperand(R3));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 8);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vmulqi16, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vmulqi32, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ mov(R0, ShifterOperand(1));
+    __ vmovsr(S0, R0);
+    __ vmovsr(S1, R0);
+    __ vmovsr(S2, R0);
+    __ vmovsr(S3, R0);
+    __ mov(R0, ShifterOperand(2));
+    __ vmovsr(S4, R0);
+    __ vmovsr(S5, R0);
+    __ vmovsr(S6, R0);
+    __ vmovsr(S7, R0);
+
+    __ vmulqi(2, Q2, Q1, Q0);
+
+    __ vmovrs(R0, S8);
+    __ vmovrs(R1, S9);
+    __ vmovrs(R2, S10);
+    __ vmovrs(R3, S11);
+
+    __ add(R0, R0, ShifterOperand(R1));
+    __ add(R0, R0, ShifterOperand(R2));
+    __ add(R0, R0, ShifterOperand(R3));
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 8);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vmulqi32, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
 ASSEMBLER_TEST_GENERATE(Vaddqs, assembler) {
   if (CPUFeatures::neon_supported()) {
     __ LoadSImmediate(S0, 1.0);
     __ vmovs(S1, S0);
     __ vmovs(S2, S0);
     __ vmovs(S3, S0);
     __ vmovs(S4, S0);
     __ vmovs(S5, S0);
     __ vmovs(S6, S0);
     __ vmovs(S7, S0);
@@ skipping 15 matching lines @@
 }
 
 
 ASSEMBLER_TEST_RUN(Vaddqs, test) {
   EXPECT(test != NULL);
   typedef int (*Tst)();
   EXPECT_EQ(8, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
 }
 
 
+ASSEMBLER_TEST_GENERATE(Vsubqs, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ LoadSImmediate(S0, 1.0);
+    __ vmovs(S1, S0);
+    __ vmovs(S2, S0);
+    __ vmovs(S3, S0);
+    __ LoadSImmediate(S4, 4.0);
+    __ vmovs(S5, S4);
+    __ vmovs(S6, S4);
+    __ vmovs(S7, S4);
+
+    __ vsubqs(Q2, Q1, Q0);
+
+    __ vadds(S8, S8, S9);
+    __ vadds(S8, S8, S10);
+    __ vadds(S8, S8, S11);
+
+    __ vcvtis(S0, S8);
+    __ vmovrs(R0, S0);
+
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 12);
+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vsubqs, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(12, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
+ASSEMBLER_TEST_GENERATE(Vmulqs, assembler) {
+  if (CPUFeatures::neon_supported()) {
+    __ LoadSImmediate(S0, 2.0);
+    __ vmovs(S1, S0);
+    __ vmovs(S2, S0);
+    __ vmovs(S3, S0);
+    __ LoadSImmediate(S4, 3.0);
+    __ vmovs(S5, S4);
+    __ vmovs(S6, S4);
+    __ vmovs(S7, S4);
+
+    __ vmulqs(Q2, Q1, Q0);
+
+    __ vadds(S8, S8, S9);
+    __ vadds(S8, S8, S10);
+    __ vadds(S8, S8, S11);
+
+    __ vcvtis(S0, S8);
+    __ vmovrs(R0, S0);
+
+    __ bx(LR);
+  } else {
+    __ LoadImmediate(R0, 42);

[regis, 2013/07/16 00:23:30]

24?

[zra, 2013/07/16 16:11:01]

Done.

+    __ bx(LR);
+  }
+}
+
+
+ASSEMBLER_TEST_RUN(Vmulqs, test) {
+  EXPECT(test != NULL);
+  typedef int (*Tst)();
+  EXPECT_EQ(24, EXECUTE_TEST_CODE_INT32(Tst, test->entry()));
+}
+
+
 // Called from assembler_test.cc.
 // LR: return address.
 // R0: context.
 // R1: value.
 // R2: growable array.
 ASSEMBLER_TEST_GENERATE(StoreIntoObject, assembler) {
   __ PushList((1 << CTX) | (1 << LR));
   __ mov(CTX, ShifterOperand(R0));
   __ StoreIntoObject(R2,
                      FieldAddress(R2, GrowableObjectArray::data_offset()),
                      R1);
   __ PopList((1 << CTX) | (1 << LR));
   __ Ret();
 }
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM