Add MSA (MIPS SIMD Arch) optimized VectorMath functions

third_party/WebKit/Source/platform/cpu/mips/CommonMacrosMSA.h

 // Copyright 2016 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef CommonMacrosMSA_h
 #define CommonMacrosMSA_h
 
 #include <msa.h>
 #include <stdint.h>
 
@@ skipping 500 matching lines @@
 {                                             \
     out0 = in0 / in1;                         \
     out1 = in2 / in3;                         \
 }
 #define DIV4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3)  \
 {                                                                             \
     DIV2(in0, in1, in2, in3, out0, out1);                                     \
     DIV2(in4, in5, in6, in7, out2, out3);                                     \
 }
 
+/* Description : Addition of 2 pairs of vectors
+ * Arguments   : Inputs  - in0, in1, in2, in3
+ *               Outputs - out0, out1
+ * Details     : Each element in 'in0' is added to 'in1' and result is written
+ *               to 'out0'.
+ */

[Raymond Toy, 2016/10/03 16:47:06]

Comment style here seems inconsistent with the nearby commenting style.

[Prashant.Patil, 2016/10/04 11:47:27]

Done.

+#define ADD2(in0, in1, in2, in3, out0, out1)  \
+{                                             \
+    out0 = in0 + in1;                         \
+    out1 = in2 + in3;                         \
+}
+#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3)  \

[Raymond Toy, 2016/10/03 16:47:06]

I know the naming here is consistent with the rest of the file, but it's almost
impossible to know which if the in's get added to produce which out.

[Prashant.Patil, 2016/10/04 11:47:28]

I will add macro description for better understanding.

+{                                                                             \
+    ADD2(in0, in1, in2, in3, out0, out1);                                     \
+    ADD2(in4, in5, in6, in7, out2, out3);                                     \
+}
+
 /* Description : Vector Floating-Point Convert from Unsigned Integer
    Arguments   : Inputs  - in0, in1
                  Outputs - out0, out1
    Details     :
 */
 #define FFINTU_W2(RTYPE, in0, in1, out0, out1)  \
 {                                               \
     out0 = (RTYPE)__msa_ffint_u_w((v4u32)in0);  \
     out1 = (RTYPE)__msa_ffint_u_w((v4u32)in1);  \
 }
@@ skipping 18 matching lines @@
 }
 #define FTRUNCU_W2_UB(...) FTRUNCU_W2(v16u8, __VA_ARGS__)
 
 #define FTRUNCU_W4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3)  \
 {                                                                      \
     FTRUNCU_W2(RTYPE, in0, in1, out0, out1);                           \
     FTRUNCU_W2(RTYPE, in2, in3, out2, out3);                           \
 }
 #define FTRUNCU_W4_UB(...) FTRUNCU_W4(v16u8, __VA_ARGS__)
 
+/* Description : Vector Floating-Point multiply with scale and accumulate
+   Arguments   : Inputs  - in0, in1, scale

[Raymond Toy, 2016/10/03 16:47:06]

Based on the code, out0, and out1 are also inputs.

[Prashant.Patil, 2016/10/04 11:47:27]

Done.

+                 Outputs - out0, out1
+   Details     :
+*/
+#define VSMA2(in0, in1, out0, out1, scale)  \
+{                                           \
+    out0 += in0 * scale;                    \
+    out1 += in1 * scale;                    \
+}
+
+#define VSMA4(in0, in1, in2, in3, out0, out1, out2, out3, scale)  \
+{                                                                 \
+    VSMA2(in0, in1, out0, out1, scale);                           \
+    VSMA2(in2, in3, out2, out3, scale);                           \
+}
+
+/* Description : Vector Floating-Point multiply with scale
+   Arguments   : Inputs  - in0, in1, scale
+                 Outputs - out0, out1
+   Details     :
+*/
+#define VSMUL2(in0, in1, out0, out1, scale)  \
+{                                            \
+    out0 = in0 * scale;                      \
+    out1 = in1 * scale;                      \
+}
+
+#define VSMUL4(in0, in1, in2, in3, out0, out1, out2, out3, scale)  \
+{                                                                  \
+    VSMUL2(in0, in1, out0, out1, scale);                           \
+    VSMUL2(in2, in3, out2, out3, scale);                           \
+}
+
+/* Description : Vector Floating-Point max value with absolute
+   Arguments   : Inputs  - in0, in1
+                 Outputs - out0, out1

[Raymond Toy, 2016/10/03 16:47:06]

There are not outputs out0, out1, just the one output, max.  And there's also
the input mask.

Impossible to tell what mask is for other than looking at the code.

[Prashant.Patil, 2016/10/04 11:47:28]

I will update the macro usage for better code readability.

+   Details     :
+*/
+#define VMAXMGV2(in0, in1, mask, max)                     \
+{                                                         \
+    max = __msa_fmax_w(max, (v4f32)((v16i8)in0 & mask));  \
+    max = __msa_fmax_w(max, (v4f32)((v16i8)in1 & mask));  \
+}
+
+#define VMAXMGV4(in0, in1, in2, in3, mask, max)  \
+{                                                \
+    VMAXMGV2(in0, in1, mask, max);               \
+    VMAXMGV2(in2, in3, mask, max);               \
+}
+
+/* Description : Vector Floating-Point clip to min max
+   Arguments   : Inputs  - in0, in1

[Raymond Toy, 2016/10/03 16:47:06]

You forgot min, max as inputs.

[Prashant.Patil, 2016/10/04 11:47:28]

Done.

+                 Outputs - out0, out1
+   Details     :
+*/
+#define VCLIP2(in0, in1, min, max, out0, out1)         \
+{                                                      \
+    out0 = __msa_fmax_w(__msa_fmin_w(in0, max), min);  \
+    out1 = __msa_fmax_w(__msa_fmin_w(in1, max), min);  \
+}
+
+#define VCLIP4(in0, in1, in2, in3, min, max, out0, out1, out2, out3)  \
+{                                                                     \
+    VCLIP2(in0, in1, min, max, out0, out1);                           \
+    VCLIP2(in2, in3, min, max, out2, out3);                           \
+}
+
 #endif // CommonMacrosMSA_h