From fccf24a51c7ee756d1b6cd3e44a431218139aae3 Mon Sep 17 00:00:00 2001
From: Richard Burgess <>
Date: Tue, 17 Jan 1995 09:02:22 +0000
Subject: [PATCH] autocommit for file dated 1995-01-17 09:02:22

---
 msamples/cm32m/optimize.h | 236 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 236 insertions(+)
 create mode 100644 msamples/cm32m/optimize.h

diff --git a/msamples/cm32m/optimize.h b/msamples/cm32m/optimize.h
new file mode 100644
index 0000000..e1b4259
--- /dev/null
+++ b/msamples/cm32m/optimize.h
@@ -0,0 +1,236 @@
+/*
+ * Peephole Optimizer for the C Minus32 386 compiler.
+ *
+ * The strategy behind the peephole optimizer is to keep the peephole
+ * buffer completly filled.  The list of optimizations is in priority order
+ * in the table below. Each buffer full is checked against each entry
+ * in the table.  Even after a substitution is made, the new substituted
+ * code is left in the buffer to check for further optimization.
+ * See the rules for table entries
+ * just prior to the table below.  All entries are searched even after
+ * new code is inserted to ensure all optimization sequences used.
+ *
+ * Copyright 1990 Dave Dunfield
+ * Copyright 1992 R.A. Burgess
+ */
+#include <ctype.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+#define	OBUF_SIZE		10		/* number of entries in peephole buffer */
+#define	OLINE_SIZE		100		/* maximum size of input line */
+#define OSYMBOLS		8		/* maximum # symbols per peep */
+#define OSYMBOL_SIZE	25		/* maximum size of symbol */
+
+/*
+ * Peephole optimization table:
+ *
+ * Each two entries represent the instruction sequences:
+ * The first is the string of instructions to SEARCH for,
+ * the second is the optimized REPLACEMENT string.
+ *
+ * 1) The REPLACEMENT instructions are in reverse order!
+ * 2) The highest priority entries should be entered first in the table.
+ * 3) Tabs (\t) should begin each entry.
+ * 4) Linefeeds (\n) should end each except for the last entry of a line.
+ *
+ */
+
+char *peep_table[] = {
+
+/* 80386/80486 Optimizations */
+
+/* data movement & register usage optimizations */
+
+"\tMOV \200,\201\n\tMOV \201,\200",
+"\tMOV \200,\201",
+
+"\tMOV \200,\201\n\tMOV \200,\202",
+"\tMOV \200,\202",
+
+"\tMOV \200,\201\n\tMOV \202,\200\n\tMOV \201,\200",
+"\tMOV \202,\200\n\tMOV \200,\201",
+
+"\tMOV EAX,\200\n\tMOV ECX,EAX",
+"\tMOV ECX,\200",
+
+"\tMOV ESI,\200\n\tMOV EAX,ESI",
+"\tMOV EAX,\200",
+
+"\tLEA ESI,\200\n\tMOV EAX,ESI",
+"\tLEA EAX,\200",
+
+"\tPUSH EAX\n\tLEA ESI,\200\n\tMOV EAX,ESI\n\tPOP EBX\n\tADD EAX,EBX",
+"\tADD EAX,ESI\n\tLEA ESI,\200",
+
+"\tPOP EBX\n\tMOV EAX,EBX",
+"\tPOP EAX",
+
+/* indexing operations */
+
+"\tMOV ECX,32\n\tMUL ECX",
+"\tSHL EAX,5",
+
+"\tMOV ECX,32\n\tIMUL ECX",
+"\tSHL EAX,5",
+
+"\tMOV ECX,16\n\tMUL ECX",
+"\tSHL EAX,4",
+
+"\tMOV ECX,16\n\tIMUL ECX",
+"\tSHL EAX,4",
+
+"\tMOV ECX,8\n\tMUL ECX",
+"\tSHL EAX,3",
+
+"\tMOV ECX,8\n\tIMUL ECX",
+"\tSHL EAX,3",
+
+"\tMOV ECX,4\n\tMUL ECX",
+"\tSHL EAX,2",
+
+"\tMOV ECX,4\n\tIMUL ECX",
+"\tSHL EAX,2",
+
+"\tMOV ECX,2\n\tMUL ECX",
+"\tSHL EAX,1",
+
+"\tMOV ECX,2\n\tIMUL ECX",
+"\tSHL EAX,1",
+
+/* jump optimizations */
+
+"\tJMP \200\n\200:",
+"\200:",
+
+"\tJMP \200\n\201:\n\200:",
+"\200:\n\201:",
+
+"\tJMP \200\n\tJMP \201",
+"\tJMP \200",
+
+"\tJNZ \200\n\tJMP \201\n\200:\n\tJMP \202\n\201:",
+"\201:\n\tJMP \202\n\tJZ \201",
+
+"\tJZ \200\n\tJMP \201\n\200:\n\tJMP \202\n\201:",
+"\201:\n\tJMP \202\n\tJNZ \201",
+
+"\tJNZ \200\n\tJMP \201\n\200:\n\tJMP SHORT \202",
+"\tJMP \201\n\tJNZ \202",
+
+"\tJZ \200\n\tJMP \201\n\200:\n\tJMP SHORT \202",
+"\tJMP \201\n\tJZ \202",
+
+/* conversion optimizations */
+
+"\tMOV AL,\200\n\tMOVSX EAX,AL",
+"\tMOVSX EAX,BYTE PTR \200",
+
+"\tMOV AL,\200\n\tMOVZX EAX,AL",
+"\tMOVZX EAX,BYTE PTR \200",
+
+"\tMOV AX,\200\n\tMOVSX EAX,AX",
+"\tMOVSX EAX,WORD PTR \200",
+
+"\tMOV AX,\200\n\tMOVZX EAX,AX",
+"\tMOVZX EAX,WORD PTR \200",
+
+/* comparisons to ECX, or MOV into EAX for compare (when not needed) */
+
+"\tMOV ECX,\200\n\tCMP EAX,ECX",
+"\tCMP EAX,\200",
+
+"\tMOVSX ECX,BYTE PTR\200\n\tCMP EAX,ECX",
+"\tCMP AL,BYTE PTR\200",
+
+"\tMOVZX ECX,BYTE PTR\200\n\tCMP EAX,ECX",
+"\tCMP AL,BYTE PTR\200",
+
+/* stack/parameters */
+
+"\tMOV EAX,[\200\n\tPUSH EAX",
+"\tPUSH DWORD PTR [\200",
+
+"\tMOV EAX,\200\n\tPUSH EAX",
+"\tPUSH \200",
+
+/* more simple optimizations */
+
+"\tMOV EAX,0",
+"\tXOR EAX,EAX",
+
+"\tMOV AX,0",
+"\tXOR AX,AX",
+
+"\tMOV AL,0",
+"\tXOR AL,AL",
+
+"\tMOV AL,0",
+"\tXOR AL,AL",
+
+0 };
+
+/* circular peep hole buffer & read/write pointers */
+
+	char peep_buffer[OBUF_SIZE][OLINE_SIZE];
+
+	unsigned peep_top = 0,		/* first entry in circular buffer */
+			 peep_next = 0;		/* Next place to put new entry    */
+
+	/* If ((next+1) mod n entries) == top then buffer is full */
+	/* If next == top then buffer is empty */
+
+/* Symbol table */
+
+	char symbols[OSYMBOLS][OSYMBOL_SIZE];
+
+/* End of Optimize.h */
+
+/* candidates
+OLD:
+	SHL EAX,2

+	MOV ESI,OFFSET xxxx

+	ADD ESI,EAX

+	MOV EAX,[ESI]

+NEW:
+	MOV ESI,OFFSET xxxx

+	MOV EAX, [ESI+EAX*2]
+
+OLD:
+	ADD ESI,xx
+	MOV rr,nnn
+	MOV [ESI],rr
+NEW
+	MOV rr,nnn
+	MOV [ESI+xx],rr
+
+OLD
+	ADD ESI,xx
+	MOV rr, [ESI]
+NEW
+	MOV rr, [ESI+xx]
+
+
+OLD (candidates):
+	ADD ESI,EAX
+	ADD ESI,4
+	MOVZX EAX,WORD PTR _CrntStrucDef
+	MOV [ESI],AX
+
+	ADD ESI,EAX
+	MOV EAX,[EBP-16]
+	MOV [ESI],EAX
+
+	ADD ESI,EAX
+	ADD ESI,10
+	MOVZX EAX,WORD PTR [ESI]
+
+	ADD ESI,EAX
+	ADD ESI,8
+	MOV EAX,DWORD PTR [ESI]
+
+	ADD ESI,EAX
+	MOV EAX,DWORD PTR [ESI]
+
+*/
-- 
2.40.0