From 063fae50fe25388d10e9ec1915c561dc0f4d519d Mon Sep 17 00:00:00 2001
From: LDj3SNuD <35856442+LDj3SNuD@users.noreply.github.com>
Date: Sun, 15 Jul 2018 05:53:26 +0200
Subject: Fix EmitHighNarrow(), EmitSaturatingNarrowOp() when Rd == Rn || Rd ==
 Rm (& Part != 0). Optimization of EmitVectorTranspose(), EmitVectorUnzip(),
 EmitVectorZip() algorithms (reduction of the number of operations and their
 complexity). Add 12 Tests about Trn1/2, Uzp1/2, Zip1/2 (V) instructions.
 (#268)

* Update CpuTestSimdArithmetic.cs

* Update CpuTestSimd.cs

* Update CpuTestSimdReg.cs

* Update Instructions.cs

* Update AInstEmitSimdArithmetic.cs

* Update AInstEmitSimdHelper.cs

* Update AInstEmitSimdMove.cs

* Delete CpuTestSimdMove.cs
---
 ChocolArm64/Instruction/AInstEmitSimdMove.cs | 47 +++++++++++++++-------------
 1 file changed, 25 insertions(+), 22 deletions(-)

(limited to 'ChocolArm64/Instruction/AInstEmitSimdMove.cs')

diff --git a/ChocolArm64/Instruction/AInstEmitSimdMove.cs b/ChocolArm64/Instruction/AInstEmitSimdMove.cs
index 739f01c6..592cab73 100644
--- a/ChocolArm64/Instruction/AInstEmitSimdMove.cs
+++ b/ChocolArm64/Instruction/AInstEmitSimdMove.cs
@@ -331,17 +331,18 @@ namespace ChocolArm64.Instruction
         {
             AOpCodeSimdReg Op = (AOpCodeSimdReg)Context.CurrOp;
 
-            int Bytes = Op.GetBitsCount() >> 3;
-
-            int Elems = Bytes >> Op.Size;
+            int Words = Op.GetBitsCount() >> 4;
+            int Pairs = Words >> Op.Size;
 
-            for (int Index = 0; Index < Elems; Index++)
+            for (int Index = 0; Index < Pairs; Index++)
             {
-                int Elem = (Index & ~1) + Part;
+                int Idx = Index << 1;
 
-                EmitVectorExtractZx(Context, (Index & 1) == 0 ? Op.Rn : Op.Rm, Elem, Op.Size);
+                EmitVectorExtractZx(Context, Op.Rn, Idx + Part, Op.Size);
+                EmitVectorExtractZx(Context, Op.Rm, Idx + Part, Op.Size);
 
-                EmitVectorInsertTmp(Context, Index, Op.Size);
+                EmitVectorInsertTmp(Context, Idx + 1, Op.Size);
+                EmitVectorInsertTmp(Context, Idx    , Op.Size);
             }
 
             Context.EmitLdvectmp();
@@ -357,18 +358,18 @@ namespace ChocolArm64.Instruction
         {
             AOpCodeSimdReg Op = (AOpCodeSimdReg)Context.CurrOp;
 
-            int Bytes = Op.GetBitsCount() >> 3;
+            int Words = Op.GetBitsCount() >> 4;
+            int Pairs = Words >> Op.Size;
 
-            int Elems = Bytes >> Op.Size;
-            int Half  = Elems >> 1;
-
-            for (int Index = 0; Index < Elems; Index++)
+            for (int Index = 0; Index < Pairs; Index++)
             {
-                int Elem = Part + ((Index & (Half - 1)) << 1);
+                int Idx = Index << 1;
 
-                EmitVectorExtractZx(Context, Index < Half ? Op.Rn : Op.Rm, Elem, Op.Size);
+                EmitVectorExtractZx(Context, Op.Rn, Idx + Part, Op.Size);
+                EmitVectorExtractZx(Context, Op.Rm, Idx + Part, Op.Size);
 
-                EmitVectorInsertTmp(Context, Index, Op.Size);
+                EmitVectorInsertTmp(Context, Pairs + Index, Op.Size);
+                EmitVectorInsertTmp(Context,         Index, Op.Size);
             }
 
             Context.EmitLdvectmp();
@@ -384,18 +385,20 @@ namespace ChocolArm64.Instruction
         {
             AOpCodeSimdReg Op = (AOpCodeSimdReg)Context.CurrOp;
 
-            int Bytes = Op.GetBitsCount() >> 3;
+            int Words = Op.GetBitsCount() >> 4;
+            int Pairs = Words >> Op.Size;
 
-            int Elems = Bytes >> Op.Size;
-            int Half  = Elems >> 1;
+            int Base = Part != 0 ? Pairs : 0;
 
-            for (int Index = 0; Index < Elems; Index++)
+            for (int Index = 0; Index < Pairs; Index++)
             {
-                int Elem = Part * Half + (Index >> 1);
+                int Idx = Index << 1;
 
-                EmitVectorExtractZx(Context, (Index & 1) == 0 ? Op.Rn : Op.Rm, Elem, Op.Size);
+                EmitVectorExtractZx(Context, Op.Rn, Base + Index, Op.Size);
+                EmitVectorExtractZx(Context, Op.Rm, Base + Index, Op.Size);
 
-                EmitVectorInsertTmp(Context, Index, Op.Size);
+                EmitVectorInsertTmp(Context, Idx + 1, Op.Size);
+                EmitVectorInsertTmp(Context, Idx    , Op.Size);
             }
 
             Context.EmitLdvectmp();
-- 
cgit v1.2.3