From 5c596d3797fe5a322485f2b865f02314524b45e1 Mon Sep 17 00:00:00 2001
From: Marisa Heit <rheit@users.noreply.github.com>
Date: Fri, 21 Oct 2016 22:08:12 -0500
Subject: [PATCH] Change VM to increment PC after each instruction rather than
 before

- VC++ generated horribly stupid code for x64 when incrementing pc at the
  beginning of each instruction by storing hundreds of copies of it for
  every opcode executed. Incrementing pc at the end avoids this madness.
- It is possible I messed something up with this change. Hopefully not.
---
 src/zscript/vmexec.cpp | 22 +++++++++++-----------
 src/zscript/vmexec.h   | 24 +++++++++++++++---------
 2 files changed, 26 insertions(+), 20 deletions(-)

diff --git a/src/zscript/vmexec.cpp b/src/zscript/vmexec.cpp
index 454f02d03..7ee89329a 100644
--- a/src/zscript/vmexec.cpp
+++ b/src/zscript/vmexec.cpp
@@ -11,21 +11,21 @@
 
 #if COMPGOTO
 #define OP(x)	x
-#define NEXTOP	do { unsigned op = pc->op; a = pc->a; pc++; goto *ops[op]; } while(0)
+#define NEXTOP	do { pc++; unsigned op = pc->op; a = pc->a; goto *ops[op]; } while(0)
 #else
 #define OP(x)	case OP_##x
-#define NEXTOP	break
+#define NEXTOP	pc++; break
 #endif
 
 #define luai_nummod(a,b)        ((a) - floor((a)/(b))*(b))
 
-#define A				(pc[-1].a)
-#define B				(pc[-1].b)
-#define C				(pc[-1].c)
-#define Cs				(pc[-1].cs)
-#define BC				(pc[-1].i16u)
-#define BCs				(pc[-1].i16)
-#define ABCs			(pc[-1].i24)
+#define A				(pc[0].a)
+#define B				(pc[0].b)
+#define C				(pc[0].c)
+#define Cs				(pc[0].cs)
+#define BC				(pc[0].i16u)
+#define BCs				(pc[0].i16)
+#define ABCs			(pc[0].i24)
 #define JMPOFS(x)		((x)->i24)
 
 #define KC				(konstd[C])
@@ -48,8 +48,8 @@
 
 #define CMPJMP(test) \
 	if ((test) == (a & CMP_CHECK)) { \
-		assert(pc->op == OP_JMP); \
-		pc += 1 + JMPOFS(pc); \
+		assert(pc[1].op == OP_JMP); \
+		pc += 1 + JMPOFS(pc+1); \
 	} else { \
 		pc += 1; \
 	}
diff --git a/src/zscript/vmexec.h b/src/zscript/vmexec.h
index adf2986c5..3637f2c1c 100644
--- a/src/zscript/vmexec.h
+++ b/src/zscript/vmexec.h
@@ -52,11 +52,17 @@ begin:
 	{
 #if !COMPGOTO
 	VM_UBYTE op;
-	for(;;) switch(op = pc->op, a = pc->a, pc++, op)
+	for(;;) switch(op = pc->op, a = pc->a, op)
 #else
+	pc--;
 	NEXTOP;
 #endif
 	{
+#if !COMPGOTO
+	default:
+		assert(0 && "Undefined opcode hit");
+		NEXTOP;
+#endif
 	OP(LI):
 		ASSERTD(a);
 		reg.d[a] = BCs;
@@ -367,13 +373,13 @@ begin:
 		}
 		NEXTOP;
 	OP(JMP):
-		pc += JMPOFS(pc - 1);
+		pc += JMPOFS(pc);
 		NEXTOP;
 	OP(IJMP):
 		ASSERTD(a);
 		pc += (BCs + reg.d[a]);
-		assert(pc->op == OP_JMP);
-		pc += 1 + JMPOFS(pc);
+		assert(pc[1].op == OP_JMP);
+		pc += 1 + JMPOFS(pc+1);
 		NEXTOP;
 	OP(PARAMI):
 		assert(f->NumParam < sfunc->MaxParam);
@@ -490,7 +496,7 @@ begin:
 			VMReturn returns[MAX_RETURNS];
 			int numret;
 
-			FillReturns(reg, f, returns, pc, C);
+			FillReturns(reg, f, returns, pc+1, C);
 			if (call->Native)
 			{
 				numret = static_cast<VMNativeFunction *>(call)->NativeCall(stack, reg.param + f->NumParam - B, B, returns, C);
@@ -603,8 +609,8 @@ begin:
 		{
 			THROW(X_TOO_MANY_TRIES);
 		}
-		assert((pc + JMPOFS(pc - 1))->op == OP_CATCH);
-		exception_frames[try_depth++] = pc + JMPOFS(pc - 1);
+		assert((pc + JMPOFS(pc) + 1)->op == OP_CATCH);
+		exception_frames[try_depth++] = pc + JMPOFS(pc) + 1;
 		NEXTOP;
 	OP(UNTRY):
 		assert(a <= try_depth);
@@ -704,8 +710,8 @@ begin:
 			}
 			if (cmp == (a & CMP_CHECK))
 			{
-				assert(pc->op == OP_JMP);
-				pc += 1 + JMPOFS(pc);
+				assert(pc[1].op == OP_JMP);
+				pc += 1 + JMPOFS(pc+1);
 			}
 			else
 			{