Skip to contentSkip to conversation

Pointer arithmetic error

DISCUSSION David Given 6 participants 20 messages Jun 26, 2008 – Jun 29, 2008 1 message not in the archive Sparse checker

lore
Share & tools
lore
#0 lore raw

I've just found what I think is a bug. In evaluate_ptr_add(), in evaluate.c, are the following lines:

	/* Get the size of whatever the pointer points to */
	multiply = base->bit_size >> 3;

This divides the bit size by 8 to get the size in bytes. However, this doesn't take into account that bits_in_char might not be 8. I think this should actually be:

	/* Get the size of whatever the pointer points to */
	multiply = base->bit_size / bits_in_char;

(This now produces the right result on my non-8-bit-char 'platform'.)

Signature · 4 lines
┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup
#1 lore raw
Quoted 9 lines · David Given
       /* Get the size of whatever the pointer points to */
       multiply = base->bit_size >> 3;

This divides the bit size by 8 to get the size in bytes. However, this doesn't take into account that bits_in_char might not be 8. I think this should actually be:

       /* Get the size of whatever the pointer points to */
       multiply = base->bit_size / bits_in_char;

I don't think so. The offset part of the ptr_add op is mean to be the absolute offset, not how many element of the pointer index. In other words, it is how many address number it need to add.

Chris

#2 lore raw

Chris Li wrote:
[...]

Quoted 3 lines

I don't think so. The offset part of the ptr_add op is mean to be the absolute offset, not how many element of the pointer index. In other words, it is how many address number it need to add.

Hmm. True. But even so, it's still not right on my system, which doesn't use 8-bit bytes. (It's word addressable where each word can contain any value, so sizeof(int) == sizeof(double) == sizeof(char) == 1.)

Should there, then, be another symbol to define the number of bits in a byte, distinct from the number of bits in a char?

Signature · 4 lines
┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup
maintainer
#3 lore raw
Quoted 3 lines · David Given

Hmm. True. But even so, it's still not right on my system, which doesn't use 8-bit bytes. (It's word addressable where each word can contain any value, so sizeof(int) == sizeof(double) == sizeof(char) == 1.)

Ah, I see. I was thinking some thing else.
What platform exactly is that?

Quoted 2 lines

Should there, then, be another symbol to define the number of bits in a byte, distinct from the number of bits in a char?

Byte need to big enough to hold the char. Using bits_in_byte is better. There might be other place in sparse assume byte is 8 bits.

Chris

#4 lore raw
In reply to David Given
Quoted 9 lines · David Given

Chris Li wrote:
[...]

Quoted 3 lines

I don't think so. The offset part of the ptr_add op is mean to be the absolute offset, not how many element of the pointer index. In other words, it is how many address number it need to add.

Hmm. True. But even so, it's still not right on my system, which doesn't use 8-bit bytes. (It's word addressable where each word can contain any value, so sizeof(int) == sizeof(double) == sizeof(char) == 1.)

/usr/include/limits.h should have a correct #define CHAR_BIT.

Quoted 2 lines

Should there, then, be another symbol to define the number of bits in a byte, distinct from the number of bits in a char?

In C, there is no type "byte" (unless you typedef oder #define it). "byte" is usually (but not necessarily) meant as "unsigned char".

Bernd

Signature · 3 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services
#5 lore raw

Bernd Petrovitsch wrote:
[...]

Quoted 2 lines

In C, there is no type "byte" (unless you typedef oder #define it). "byte" is usually (but not necessarily) meant as "unsigned char".

The issue here is that sparse is helpful and converts pointer offsets into byte offsets when generating code for pointer arithmetic. So:

const int* p = (const char*) 1234;
p += 10;

-->

set.32 %p0 <- 1234
add.32 %p1 <- %p0, 40
(In fact, in this example it'd collapse these together and use 1274 instead.)

This appears to be done using hard-coded knowledge that a byte (one unit of addressingness) is 8 bits wide.

Quoted 2 lines · Chris Li

Byte need to big enough to hold the char. Using bits_in_byte is better. There might be other place in sparse assume byte is 8 bits.

IIRC C specifies that sizeof() returns values measured in chars, but I don't believe it specifies any mapping between the size of chars and the underlying addressing units --- it should be possible to use 16-bit chars, for example, on an 8-bit byte system. Using 32-bit ints, sizeof(int) would then return 2; but you wouldn't be able to access individual bytes from C.

That's assuming I've remembered the spec correctly, of course --- this stuff is annoying subtle.

If you like I'll go and have a look to see if this can be easily fixed --- I need to do it anyway. In fact, I have vague recollections that load and store were actually using offsets like 0, 1, 2, etc (the 'right' values), which means they must have been using bits_per_char to do their calculations; I need to check up on this.

Signature · 2 lines
David Given
[email protected]
#6 lore raw
Quoted 13 lines · David Given

Bernd Petrovitsch wrote:
[...]

Quoted 2 lines

In C, there is no type "byte" (unless you typedef oder #define it). "byte" is usually (but not necessarily) meant as "unsigned char".

The issue here is that sparse is helpful and converts pointer offsets into byte offsets when generating code for pointer arithmetic. So:

const int* p = (const char*) 1234;
p += 10;

-->

Quoted 7 lines

set.32 %p0 <- 1234
add.32 %p1 <- %p0, 40
(In fact, in this example it'd collapse these together and use 1274 instead.)

This appears to be done using hard-coded knowledge that a byte (one unit of addressingness) is 8 bits wide.

It is done IMHO with the false knowledge that "sizeof(int) == 4 * sizeof(char)".
IIUC your hardware (a DSP or what is it exactly) has "sizeof(int) == sizeof(char)" so the size of a char (or int) as such is irrelevant.

Quoted 5 lines
Quoted 2 lines · Chris Li

Byte need to big enough to hold the char. Using bits_in_byte is better. There might be other place in sparse assume byte is 8 bits.

IIRC C specifies that sizeof() returns values measured in chars, but I

ACK. Therefore "sizeof(char) == 1" must always hold.

Quoted 2 lines

don't believe it specifies any mapping between the size of chars and the underlying addressing units --- it should be possible to use 16-bit

Yes, that's what CHAR_BIT is for.

Quoted 3 lines

chars, for example, on an 8-bit byte system. Using 32-bit ints, sizeof(int) would then return 2; but you wouldn't be able to access individual bytes from C.

ACK (apart from "shift and mask it" ans similar).

Bernd

Signature · 3 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services
#7 lore raw
Quoted 2 lines · Bernd Petrovitsch

It is done IMHO with the false knowledge that "sizeof(int) == 4 * sizeof(char)".

No, it's correctly been told that ints are 32 bits wide, but then it's converting this to *bytes* (not chars) by dividing by the hard-coded constant 8.

(By 'byte' I'm referring to the quantum of addressingness of the underlying machine architecture. In other words, the numbers that are used as parameters as offsets into load and store. This isn't something that's exposed to C except on architectures where sizeof(byte) == sizeof(char), i.e., all sensible ones.)

[...]

Quoted 1 line

ACK. Therefore "sizeof(char) == 1" must always hold.

Yes; but that is only true from C's perspective.

We're dealing with things from the machine code perspective, where sizeof(byte) == 1, and sizeof(char) is not necessarily the same as sizeof(byte).

I find it really helps with this stuff if you are capable of holding two completely contradictory beliefs at the same time. A certain level of insanity can help too...

Signature · 2 lines
David Given
[email protected]
#8 lore raw
Quoted 6 lines · David Given
Quoted 2 lines · Bernd Petrovitsch

It is done IMHO with the false knowledge that "sizeof(int) == 4 * sizeof(char)".

No, it's correctly been told that ints are 32 bits wide, but then it's converting this to *bytes* (not chars) by dividing by the hard-coded

That's the bug. there is no difference between "byte" and "char". Tell it that a char has 32 bits too *if* it's the case.

Quoted 1 line

constant 8.

Why do you (or sparse) want to divide a the size of a type where "sizeof(type) == 1"?
Use (a correct) CHAR_BIT for that.

Quoted 5 lines

(By 'byte' I'm referring to the quantum of addressingness of the underlying machine architecture. In other words, the numbers that are used as parameters as offsets into load and store. This isn't something that's exposed to C except on architectures where sizeof(byte) == sizeof(char), i.e., all sensible ones.)

Such hardware-specific issues must be implemented by the compiler. So you probably have to extend "sparse" for that feature.

Quoted 4 lines

[...]

Quoted 1 line

ACK. Therefore "sizeof(char) == 1" must always hold.

Yes; but that is only true from C's perspective.

Is there another remotely relevant perspective (let alone more important) in a C parser?

Quoted 3 lines

We're dealing with things from the machine code perspective, where sizeof(byte) == 1, and sizeof(char) is not necessarily the same as sizeof(byte).

And that's AFAIU the problem: "sizeof(char) == 1" in C (K&R, c8x, c99) per definitionem. There is no way around it. For all architectures in the world, past, present and future.

If your hardware has 32bit char, short and int types, there is no way in C to define a "byte" in a somewhat sane way. Yes, there are bitfields but that tend to get somewhat clumsy.

Perhaps the are some compiler-specific extensions for such strange hardware. But then you leave the defined ground and need to implement them into "sparse" too.

Quoted 2 lines

I find it really helps with this stuff if you are capable of holding two completely contradictory beliefs at the same time. A certain level of

I'm just pointing out the view of "C" (or what I understand it is). Since "sparse" is a C code parser, this should matter most (if not all and only) IMNSHO.
I don't think the (above newly beyond "C" introduced) concept of "byte" for your specific, unknown and secret CPU/architecture is helpful in any way.

Quoted 1 line

insanity can help too...

Bernd

Signature · 3 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services
#9 lore raw

Bernd Petrovitsch wrote:
[...]

Quoted 2 lines

That's the bug. there is no difference between "byte" and "char". Tell it that a char has 32 bits too *if* it's the case.

Having checked the standard it turns out that we've been talking at cross purposes as I've been using the wrong terminology --- it actually defines (unhelpfully) that byte and char are the same size. Sorry for the confusion.

What I was referring to when I previously said (erroneously) 'byte' was 'an address delta of 1', as understood by the assembler. Let's just call this a 'unit' for clarity. This is not necessarily the same size as a char.

[...]

Quoted 5 lines
Quoted 2 lines
Quoted 1 line

ACK. Therefore "sizeof(char) == 1" must always hold.

Yes; but that is only true from C's perspective.

Is there another remotely relevant perspective (let alone more important) in a C parser?

The issue here is that sparse is not just a C parser. It is also a compiler, and needs to know how to manipulate addresses that are represented as units. (This is why I'm using it for my project.) It has to be able to express an offset into a structure in units to pass on to the 'assembler' (with stock sparse, this is the human who's reading the output of test-linearise).

Currently I believe that some parts of sparse are assuming that a unit is 8 bits, and other parts are assuming that a unit is CHAR_BITS bits, neither of which is necessarily correct. However, I need to go and check up on this; I'm away from the machine with my project on it right now.

I'm proposing adding a bits_in_unit (or something) setting and then going through and tracking down these places and changing them to use it. That way it should still work fine on exotic architectures like mine.

(Wikipedia has a rather interesting table of machines with interesting word/byte/char sizes:

http://en.wikipedia.org/wiki/Word_(computing)#Table_of_word_sizes

The last machine with a freaky word size was the Cray C-90 in 1991, where a unit was 64 bits but a char was 8 bits...)

Signature · 2 lines
David Given
[email protected]
maintainer
#10 lore raw
Quoted 2 lines

In C, there is no type "byte" (unless you typedef oder #define it). "byte" is usually (but not necessarily) meant as "unsigned char".

In C spec, there is a concept of "byte". The union return by sizeof() is byte. Char must fit in a byte. But char does not necessary have the same bits as byte. Char can have more.

C99: 3.6, 3.7.1

Because char can always fit in byte, sizeof(char) == 1.

Quoted 2 lines

IIRC C specifies that sizeof() returns values measured in chars, but I don't believe it specifies any mapping between the size of chars and the underlying addressing units --- it should be possible to use 16-bit chars, for example, on an 8-bit byte system. Using 32-bit ints, sizeof(int) would then return 2; but you wouldn't be able to access individual bytes from C.

sizeof() return value measure in _byte_.
C99: 6.5.3.4

Quoted 5 lines · David Given

Bernd Petrovitsch wrote:
[...]

Quoted 3 lines

That's the bug. there is no difference between "byte" and "char". Tell it that a char has 32 bits too *if* it's the case.

No, there is a different between "byte" and "char". See above.

Quoted 8 lines

Having checked the standard it turns out that we've been talking at cross purposes as I've been using the wrong terminology --- it actually defines (unhelpfully) that byte and char are the same size. Sorry for the confusion.

What I was referring to when I previously said (erroneously) 'byte' was 'an address delta of 1', as understood by the assembler. Let's just call this a 'unit' for clarity. This is not necessarily the same size as a char.

In C's term, that is call a "byte" :-)

Quoted 3 lines

I'm proposing adding a bits_in_unit (or something) setting and then going through and tracking down these places and changing them to use it. That way it should still work fine on exotic architectures like mine.

You are right that point out a bug (assumption) of sparse which byte is 8 bits. Using bits_in_byte is instead of 8 is better there.
Using bits_in_char assumes char has same bits as byte. That is my read of the C spec.

Thanks

Chris

#11 lore raw

Christopher Li wrote:
[...]

Quoted 4 lines

You are right that point out a bug (assumption) of sparse which byte is 8 bits. Using bits_in_byte is instead of 8 is better there.
Using bits_in_char assumes char has same bits as byte. That is my read of the C spec.

Yes, indeed, I'd managed to get my terminology muddled to the confusion of everyone.

Okay, I've gone and looked at the implementation of this stuff; it would appear that modifying the code to supply the back end with units scaled in something other than C chars is actually quite complicated, so I haven't tried to do that. However, I'm enclosing a patch that should, hopefully, fix the cases where it's assuming 8 bit bytes. Hopefully I've
managed to catch all the cases, without breaking the octal parse code...

(I haven't used git before; is this the right format?)

Signature · 4 lines
┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup

commit 0752e02d2c7c3f67b7c1835c86bf77f11c8384dd

Author: dg <[email protected]>
Date:   Sat Jun 28 00:24:36 2008 +0100

Changed use of hardcoded 8s when converting bits to bytes to use bits_in_char instead.

compile-i386.c +2 -2
index 8526408..3bbc9c7 100644
@@ -1584,7 +1584,7 @@ static struct storage *emit_select_expr(struct expression *expr)
 	/*
 	 * Do the actual select: check the conditional for zero,
 	 * move false over true if zero
-	 */ 
+	 */
 	insn("test", reg_cond, reg_cond, NULL);
 	insn("cmovz", reg_false, reg_true, NULL);
 
@@ -2081,7 +2081,7 @@ static struct storage *x86_call_expression(struct expression *expr)
 		insn("pushl", new, NULL,
 		     !framesize ? "begin function call" : NULL);
 
-		framesize += size >> 3;
+		framesize += size / bits_in_char;
 	} END_FOR_EACH_PTR_REVERSE(arg);
 
 	fn = expr->fn;
evaluate.c +12 -12
index 2a126dd..6018e0c 100644
@@ -87,13 +87,13 @@ static struct symbol *evaluate_string(struct expression *expr)
 	array->ctype.alignment = 1;
 	array->ctype.modifiers = MOD_STATIC;
 	array->ctype.base_type = &char_ctype;
-	
+
 	addr->symbol = sym;
 	addr->ctype = &lazy_ptr_ctype;
 
 	expr->type = EXPR_PREOP;
 	expr->op = '*';
-	expr->unop = addr;  
+	expr->unop = addr;
 	expr->ctype = sym;
 	return sym;
 }
@@ -579,7 +579,7 @@ static struct symbol *evaluate_ptr_add(struct expression *expr, struct symbol *i
 	}
 
 	/* Get the size of whatever the pointer points to */
-	multiply = base->bit_size >> 3;
+	multiply = base->bit_size / bits_in_char;
 
 	if (ctype == &null_ctype)
 		ctype = &ptr_ctype;
@@ -831,7 +831,7 @@ static struct symbol *evaluate_ptr_sub(struct expression *expr)
 		struct expression *sub = alloc_expression(expr->pos, EXPR_BINOP);
 		struct expression *div = expr;
 		struct expression *val = alloc_expression(expr->pos, EXPR_VALUE);
-		unsigned long value = lbase->bit_size >> 3;
+		unsigned long value = lbase->bit_size / bits_in_char;
 
 		val->ctype = size_t_ctype;
 		val->value = value;
@@ -850,7 +850,7 @@ static struct symbol *evaluate_ptr_sub(struct expression *expr)
 		div->left = sub;
 		div->right = val;
 	}
-		
+
 	return ssize_t_ctype;
 }
 
@@ -1591,7 +1591,7 @@ static struct symbol *degenerate(struct expression *expr)
 				e3->op = '+';
 				e3->left = e0;
 				e3->right = alloc_const_expression(expr->pos,
-							expr->r_bitpos >> 3);
+							expr->r_bitpos / bits_in_char);
 				e3->ctype = &lazy_ptr_ctype;
 			} else {
 				e3 = e0;
@@ -1727,7 +1727,7 @@ static struct symbol *evaluate_postop(struct expression *expr)
 	} else if (class == TYPE_PTR) {
 		struct symbol *target = examine_pointer_target(ctype);
 		if (!is_function(target))
-			multiply = target->bit_size >> 3;
+			multiply = target->bit_size / bits_in_char;
 	}
 
 	if (multiply) {
@@ -1850,7 +1850,7 @@ static struct symbol *find_identifier(struct ident *ident, struct symbol_list *_
 					continue;
 				*offset += sym->offset;
 				return sub;
-			}	
+			}
 		}
 	} while ((list = list->next) != head);
 	return NULL;
@@ -1949,7 +1949,7 @@ static struct symbol *evaluate_member_dereference(struct expression *expr)
 			expr->base = deref->base;
 			expr->r_bitpos = deref->r_bitpos;
 		}
-		expr->r_bitpos += offset << 3;
+		expr->r_bitpos += offset * bits_in_char;
 		expr->type = EXPR_SLICE;
 		expr->r_nrbits = member->bit_size;
 		expr->r_bitpos += member->bit_offset;
@@ -2040,7 +2040,7 @@ static struct symbol *evaluate_sizeof(struct expression *expr)
 	if ((size < 0) || (size & 7))
 		expression_error(expr, "cannot size expression");
 	expr->type = EXPR_VALUE;
-	expr->value = size >> 3;
+	expr->value = size / bits_in_char;
 	expr->taint = 0;
 	expr->ctype = size_t_ctype;
 	return size_t_ctype;
@@ -2074,7 +2074,7 @@ static struct symbol *evaluate_ptrsizeof(struct expression *expr)
 	if (size & 7)
 		size = 0;
 	expr->type = EXPR_VALUE;
-	expr->value = size >> 3;
+	expr->value = size / bits_in_char;
 	expr->taint = 0;
 	expr->ctype = size_t_ctype;
 	return size_t_ctype;
@@ -2865,7 +2865,7 @@ static struct symbol *evaluate_offsetof(struct expression *expr)
 			unrestrict(idx, i_class, &i_type);
 			idx = cast_to(idx, size_t_ctype);
 			m = alloc_const_expression(expr->pos,
-						   ctype->bit_size >> 3);
+						   ctype->bit_size / bits_in_char);
 			m->ctype = size_t_ctype;
 			m->flags = Int_const_expr;
 			expr->type = EXPR_BINOP;
example.c +12 -12
index ae897dc..b844d4f 100644
@@ -27,7 +27,7 @@ static const char *opcodes[] = {
 	[OP_INVOKE] = "invoke",
 	[OP_COMPUTEDGOTO] = "jmp *",
 	[OP_UNWIND] = "unwind",
-	
+
 	/* Binary */
 	[OP_ADD] = "add",
 	[OP_SUB] = "sub",
@@ -40,7 +40,7 @@ static const char *opcodes[] = {
 	[OP_SHL] = "shl",
 	[OP_LSR] = "lsr",
 	[OP_ASR] = "asr",
-	
+
 	/* Logical */
 	[OP_AND] = "and",
 	[OP_OR] = "or",
@@ -66,7 +66,7 @@ static const char *opcodes[] = {
 
 	/* Special three-input */
 	[OP_SEL] = "select",
-	
+
 	/* Memory */
 	[OP_MALLOC] = "malloc",
 	[OP_FREE] = "free",
@@ -818,7 +818,7 @@ static const char *generic(struct bb_state *state, pseudo_t pseudo)
 		put_operand(state, op);
 		reg = target_reg(state, pseudo, NULL);
 		output_insn(state, "lea %s,%s", show_op(state, op), reg->name);
-		return reg->name;		
+		return reg->name;
 
 	default:
 		str = show_op(state, op);
@@ -870,7 +870,7 @@ static void kill_dead_reg(struct hardreg *reg)
 {
 	if (reg->dead) {
 		pseudo_t p;
-		
+
 		FOR_EACH_PTR(reg->contains, p) {
 			if (CURRENT_TAG(p) & TAG_DEAD) {
 				DELETE_CURRENT_PTR(p);
@@ -1011,7 +1011,7 @@ static void kill_pseudo(struct bb_state *state, pseudo_t pseudo)
 				continue;
 			if (CURRENT_TAG(p) & TAG_DEAD)
 				reg->dead--;
-			output_comment(state, "removing pseudo %s from reg %s", 
+			output_comment(state, "removing pseudo %s from reg %s",
 				show_pseudo(pseudo), reg->name);
 			DELETE_CURRENT_PTR(p);
 		} END_FOR_EACH_PTR(p);
@@ -1069,7 +1069,7 @@ static const char *conditional[] = {
 	[OP_SET_BE] = "be",
 	[OP_SET_AE] = "ae"
 };
-	
+
 
 static void generate_branch(struct bb_state *state, struct instruction *br)
 {
@@ -1187,7 +1187,7 @@ static void replace_asm_percent(const char **src_p, char **dst_p, struct asm_arg
 		if (index < nr)
 			replace_asm_arg(dst_p, args+index);
 		break;
-	}	
+	}
 	*src_p = src;
 	return;
 }
@@ -1583,7 +1583,7 @@ static int final_pseudo_flush(struct bb_state *state, pseudo_t pseudo, struct ha
 	struct hardreg *dst;
 
 	/*
-	 * Since this pseudo is live at exit, we'd better have output 
+	 * Since this pseudo is live at exit, we'd better have output
 	 * storage for it..
 	 */
 	hash = find_storage_hash(pseudo, state->outputs);
@@ -1829,8 +1829,8 @@ static void set_up_arch_entry(struct entrypoint *ep, struct instruction *entry)
 
 			in->type = REG_FRAME;
 			in->offset = offset;
-			
-			offset += bits >> 3;
+
+			offset += bits / bits_in_char;
 		}
 		i++;
 		NEXT_PTR_LIST(argtype);
@@ -1936,7 +1936,7 @@ static int compile(struct symbol_list *list)
 		if (ep)
 			output(ep);
 	} END_FOR_EACH_PTR(sym);
-	
+
 	return 0;
 }
 
expand.c +7 -7
index 032f0c5..d1aa019 100644
@@ -100,7 +100,7 @@ Int:
 	// Stop here unless checking for truncation
 	if (!Wcast_truncate || conservative)
 		return;
-	
+
 	// Check if we dropped any bits..
 	oldsignmask = 1ULL << (old_size-1);
 	oldmask = oldsignmask | (oldsignmask-1);
@@ -179,7 +179,7 @@ static int simplify_int_binop(struct expression *expr, struct symbol *ctype)
 		sl |= ~(mask-1);
 	if (is_signed && (sr & mask))
 		sr |= ~(mask-1);
-	
+
 	switch (CONVERT(expr->op,is_signed)) {
 	case SIGNED('+'):
 	case UNSIGNED('+'):
@@ -224,7 +224,7 @@ static int simplify_int_binop(struct expression *expr, struct symbol *ctype)
 
 	case UNSIGNED('/'):
 		if (!r) goto Div;
-		v = l / r; 
+		v = l / r;
 		break;
 
 	case SIGNED('%'):
@@ -241,7 +241,7 @@ static int simplify_int_binop(struct expression *expr, struct symbol *ctype)
 	case SIGNED(SPECIAL_LEFTSHIFT):
 	case UNSIGNED(SPECIAL_LEFTSHIFT):
 		v = l << r;
-		break; 
+		break;
 
 	case SIGNED(SPECIAL_RIGHTSHIFT):
 		v = sl >> r;
@@ -536,7 +536,7 @@ static int expand_conditional(struct expression *expr)
 
 	return cost + cond_cost + BRANCH_COST;
 }
-		
+
 static int expand_assignment(struct expression *expr)
 {
 	expand_expression(expr->left);
@@ -822,7 +822,7 @@ static int expand_expression_list(struct expression_list *list)
 	return cost;
 }
 
-/* 
+/*
  * We can simplify nested position expressions if
  * this is a simple (single) positional expression.
  */
@@ -880,7 +880,7 @@ static unsigned long bit_offset(const struct expression *expr)
 {
 	unsigned long offset = 0;
 	while (expr->type == EXPR_POS) {
-		offset += expr->init_offset << 3;
+		offset += expr->init_offset * bits_in_char;
 		expr = expr->init_expr;
 	}
 	if (expr && expr->ctype)
flow.c +11 -9
index 82fb23a..bdb2c25 100644
@@ -16,6 +16,7 @@
 #include "expression.h"
 #include "linearize.h"
 #include "flow.h"
+#include "target.h"
 
 unsigned long bb_generation;
 
@@ -265,8 +266,8 @@ void convert_load_instruction(struct instruction *insn, pseudo_t src)
 
 static int overlapping_memop(struct instruction *a, struct instruction *b)
 {
-	unsigned int a_start = a->offset << 3;
-	unsigned int b_start = b->offset << 3;
+	unsigned int a_start = a->offset * bits_in_char;
+	unsigned int b_start = b->offset * bits_in_char;
 	unsigned int a_size = a->size;
 	unsigned int b_size = b->size;
 
@@ -364,11 +365,11 @@ found_dominator:
 		use_pseudo(insn, phi, add_pseudo(dominators, phi));
 	} END_FOR_EACH_PTR(parent);
 	return 1;
-}		
+}
 
 /*
  * We should probably sort the phi list just to make it easier to compare
- * later for equality. 
+ * later for equality.
  */
 void rewrite_load_instruction(struct instruction *insn, struct pseudo_list *dominators)
 {
@@ -529,7 +530,7 @@ static void kill_dead_stores(pseudo_t pseudo, unsigned long generation, struct b
  * This should see if the "insn" trivially dominates some previous store, and kill the
  * store if unnecessary.
  */
-static void kill_dominated_stores(pseudo_t pseudo, struct instruction *insn, 
+static void kill_dominated_stores(pseudo_t pseudo, struct instruction *insn,
 	unsigned long generation, struct basic_block *bb, int local, int found)
 {
 	struct instruction *one;
@@ -581,13 +582,14 @@ void check_access(struct instruction *insn)
 	pseudo_t pseudo = insn->src;
 
 	if (insn->bb && pseudo->type == PSEUDO_SYM) {
-		int offset = insn->offset, bit = (offset << 3) + insn->size;
+		int offset = insn->offset, bit = (offset * bits_in_char) + insn->size;
 		struct symbol *sym = pseudo->sym;
 
 		if (sym->bit_size > 0 && (offset < 0 || bit > sym->bit_size))
 			warning(insn->pos, "invalid access %s '%s' (%d %d)",
 				offset < 0 ? "below" : "past the end of",
-				show_ident(sym->ident), offset, sym->bit_size >> 3);
+				show_ident(sym->ident), offset,
+				sym->bit_size / bits_in_char);
 	}
 }
 
@@ -705,7 +707,7 @@ external_visibility:
 			} END_FOR_EACH_PTR(bb);
 		}
 	}
-			
+
 	return;
 }
 
@@ -887,7 +889,7 @@ static void vrfy_children(struct basic_block *bb)
 	default:
 		break;
 	}
-		
+
 	FOR_EACH_PTR(bb->children, tmp) {
 		vrfy_bb_in_list(bb, tmp->parents);
 	} END_FOR_EACH_PTR(tmp);
show-parse.c +14 -14
index 064af32..63d9144 100644
@@ -246,8 +246,8 @@ deeper:
 
 		s = modifier_string(mod);
 		len = strlen(s);
-		name->start -= len;    
-		memcpy(name->start, s, len);  
+		name->start -= len;
+		memcpy(name->start, s, len);
 		mod = 0;
 		as = 0;
 	}
@@ -412,7 +412,7 @@ void show_symbol(struct symbol *sym)
 		struct statement *stmt = type->stmt;
 		if (stmt) {
 			int val;
-			printf("\n");		
+			printf("\n");
 			val = show_statement(stmt);
 			if (val)
 				printf("\tmov.%d\t\tretval,%d\n", stmt->ret->bit_size, val);
@@ -589,7 +589,7 @@ int show_statement(struct statement *stmt)
 		if (pre_condition) {
 			if (pre_condition->type == EXPR_VALUE) {
 				if (!pre_condition->value) {
-					loop_bottom = new_label();   
+					loop_bottom = new_label();
 					printf("\tjmp\t\t.L%d\n", loop_bottom);
 				}
 			} else {
@@ -623,7 +623,7 @@ int show_statement(struct statement *stmt)
 	}
 	case STMT_NONE:
 		break;
-	
+
 	case STMT_LABEL:
 		printf(".L%p:\n", stmt->label_identifier);
 		show_statement(stmt->label_statement);
@@ -649,9 +649,9 @@ int show_statement(struct statement *stmt)
 		int val = show_expression(stmt->range_expression);
 		int low = show_expression(stmt->range_low);
 		int high = show_expression(stmt->range_high);
-		printf("\trange( %d %d-%d)\n", val, low, high); 
+		printf("\trange( %d %d-%d)\n", val, low, high);
 		break;
-	}	
+	}
 	}
 	return 0;
 }
@@ -673,7 +673,7 @@ static int show_call_expression(struct expression *expr)
 		int new = show_expression(arg);
 		int size = arg->ctype->bit_size;
 		printf("\tpush.%d\t\tv%d\n", size, new);
-		framesize += size >> 3;
+		framesize += size / bits_in_char;
 	} END_FOR_EACH_PTR_REVERSE(arg);
 
 	fn = expr->fn;
@@ -845,7 +845,7 @@ static int show_inc_dec(struct expression *expr, int postop)
 	printf("\t%s.%d\t\tv%d,v%d,$1\n", opname, bits, new, retval);
 	show_store_gen(bits, new, expr->unop, addr);
 	return retval;
-}	
+}
 
 static int show_preop(struct expression *expr)
 {
@@ -864,7 +864,7 @@ static int show_preop(struct expression *expr)
 static int show_postop(struct expression *expr)
 {
 	return show_inc_dec(expr, 1);
-}	
+}
 
 static int show_symbol_expr(struct symbol *sym)
 {
@@ -918,7 +918,7 @@ static int show_cast_expr(struct expression *expr)
 
 	old_type = expr->cast_expression->ctype;
 	new_type = expr->cast_type;
-	
+
 	oldbits = old_type->bit_size;
 	newbits = new_type->bit_size;
 	if (oldbits >= newbits)
@@ -1017,7 +1017,7 @@ again:
 			entry = entry->ident_expression;
 			goto again;
 		}
-			
+
 		if (entry->type == EXPR_INDEX) {
 			printf(" AT '%d..%d:\n", entry->idx_from, entry->idx_to);
 			entry = entry->idx_expression;
@@ -1057,11 +1057,11 @@ int show_expression(struct expression *expr)
 			pos->line, pos->pos);
 		return 0;
 	}
-		
+
 	switch (expr->type) {
 	case EXPR_CALL:
 		return show_call_expression(expr);
-		
+
 	case EXPR_ASSIGNMENT:
 		return show_assignment(expr);
 
symbol.c +9 -9
index 3292907..a22731f 100644
@@ -22,7 +22,7 @@
 
 /*
  * Secondary symbol list for stuff that needs to be output because it
- * was used. 
+ * was used.
  */
 struct symbol_list *translation_unit_used_list = NULL;
 
@@ -117,7 +117,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
 	}
 
 	bit_size = info->bit_size;
-	base_size = sym->bit_size; 
+	base_size = sym->bit_size;
 
 	/*
 	 * Unsized arrays cause us to not align the resulting
@@ -128,7 +128,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
 		base_size = 0;
 	}
 
-	align_bit_mask = (sym->ctype.alignment << 3) - 1;
+	align_bit_mask = (sym->ctype.alignment * bits_in_char) - 1;
 
 	/*
 	 * Bitfields have some very special rules..
@@ -143,7 +143,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
 			bit_size = (bit_size + align_bit_mask) & ~align_bit_mask;
 			bit_offset = 0;
 		}
-		sym->offset = (bit_size - bit_offset) >> 3;
+		sym->offset = (bit_size - bit_offset) / bits_in_char;
 		sym->bit_offset = bit_offset;
 		sym->ctype.base_type->bit_offset = bit_offset;
 		info->bit_size = bit_size + width;
@@ -156,7 +156,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
 	 * Otherwise, just align it right and add it up..
 	 */
 	bit_size = (bit_size + align_bit_mask) & ~align_bit_mask;
-	sym->offset = bit_size >> 3;
+	sym->offset = bit_size / bits_in_char;
 
 	info->bit_size = bit_size + base_size;
 	// warning (sym->pos, "regular: offset=%d", sym->offset);
@@ -182,7 +182,7 @@ static struct symbol * examine_struct_union_type(struct symbol *sym, int advance
 		sym->ctype.alignment = info.max_align;
 	bit_size = info.bit_size;
 	if (info.align_size) {
-		bit_align = (sym->ctype.alignment << 3)-1;
+		bit_align = (sym->ctype.alignment * bits_in_char)-1;
 		bit_size = (bit_size + bit_align) & ~bit_align;
 	}
 	sym->bit_size = bit_size;
@@ -340,7 +340,7 @@ static struct symbol * examine_node_type(struct symbol *sym)
 		if (node_type && node_type->bit_size >= 0)
 			bit_size = node_type->bit_size * count;
 	}
-	
+
 	sym->bit_size = bit_size;
 	return sym;
 }
@@ -654,7 +654,7 @@ static int expand_warning(struct expression *expr, int cost)
 	FOR_EACH_PTR (arglist, arg) {
 		/*
 		 * Constant strings get printed out as a warning. By the
-		 * time we get here, the EXPR_STRING has been fully 
+		 * time we get here, the EXPR_STRING has been fully
 		 * evaluated, so by now it's an anonymous symbol with a
 		 * string initializer.
 		 *
@@ -877,7 +877,7 @@ void init_ctype(void)
 		struct symbol *sym = ctype->ptr;
 		unsigned long bit_size = ctype->bit_size ? *ctype->bit_size : -1;
 		unsigned long maxalign = ctype->maxalign ? *ctype->maxalign : 0;
-		unsigned long alignment = (bit_size + 7) >> 3;
+		unsigned long alignment = (bit_size + bits_in_char - 1) / bits_in_char;
 
 		if (alignment > maxalign)
 			alignment = maxalign;
target.c +2 -0
index bf1bb8f..a0ee010 100644
@@ -11,6 +11,8 @@ struct symbol *ssize_t_ctype = &int_ctype;
  */
 int max_alignment = 16;
 
+int bits_in_unit = 8;
+
 /*
  * Integer data types
  */
tokenize.c +5 -5
index e72c56e..67daa97 100644
@@ -96,7 +96,7 @@ static char *charstr(char *ptr, unsigned char c, unsigned char escape, unsigned
 	}
 	if (!isdigit(next))
 		return ptr + sprintf(ptr, "%o", c);
-		
+
 	return ptr + sprintf(ptr, "%03o", c);
 }
 
@@ -161,7 +161,7 @@ const char *show_token(const struct token *token)
 	case TOKEN_STREAMEND:
 		sprintf(buffer, "<end of '%s'>", stream_name(token->pos.stream));
 		return buffer;
-	
+
 	default:
 		return "WTF???";
 	}
@@ -483,7 +483,7 @@ static int escapechar(int first, int type, stream_t *stream, int *valp)
 				int nr = 2;
 				value -= '0';
 				while (next >= '0' && next <= '9') {
-					value = (value << 3) + (next-'0');
+					value = (value*8) + (next-'0');
 					next = nextchar(stream);
 					if (!--nr)
 						break;
@@ -572,7 +572,7 @@ static int get_string_token(int next, stream_t *stream)
 	token_type(token) = TOKEN_STRING;
 	token->string = string;
 	add_token(stream);
-	
+
 	return next;
 }
 
@@ -875,7 +875,7 @@ static int get_one_identifier(int c, stream_t *stream)
 	token->ident = ident;
 	add_token(stream);
 	return next;
-}		
+}
 
 static int get_one_token(int c, stream_t *stream)
 {
maintainer
#12 lore raw

Your patch is white space damaged.

While you are there, you might want to consider macro

bits_to_byte(x)
byte_to_bits(x)

To isolate out the conversion. It should generate the same code. Marginally more readable.

Chris

Quoted 678 lines · David Given

Christopher Li wrote:
[...]

Quoted 4 lines

You are right that point out a bug (assumption) of sparse which byte is 8 bits. Using bits_in_byte is instead of 8 is better there.
Using bits_in_char assumes char has same bits as byte. That is my read of the C spec.

Yes, indeed, I'd managed to get my terminology muddled to the confusion of everyone.

Okay, I've gone and looked at the implementation of this stuff; it would appear that modifying the code to supply the back end with units scaled in something other than C chars is actually quite complicated, so I haven't tried to do that. However, I'm enclosing a patch that should, hopefully, fix the cases where it's assuming 8 bit bytes. Hopefully I've
managed to catch all the cases, without breaking the octal parse code...

(I haven't used git before; is this the right format?)

--

┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup

commit 0752e02d2c7c3f67b7c1835c86bf77f11c8384dd

Author: dg <[email protected]>
Date:   Sat Jun 28 00:24:36 2008 +0100

Changed use of hardcoded 8s when converting bits to bytes to use bits_in_char instead.

compile-i386.c +2 -2
index 8526408..3bbc9c7 100644
@@ -1584,7 +1584,7 @@ static struct storage *emit_select_expr(struct expression *expr)
       /*
        * Do the actual select: check the conditional for zero,
        * move false over true if zero
-        */
+        */
       insn("test", reg_cond, reg_cond, NULL);
       insn("cmovz", reg_false, reg_true, NULL);
 
@@ -2081,7 +2081,7 @@ static struct storage *x86_call_expression(struct expression *expr)
               insn("pushl", new, NULL,
                    !framesize ? "begin function call" : NULL);
 
-               framesize += size >> 3;
+               framesize += size / bits_in_char;
       } END_FOR_EACH_PTR_REVERSE(arg);
 
       fn = expr->fn;
evaluate.c +12 -12
index 2a126dd..6018e0c 100644
@@ -87,13 +87,13 @@ static struct symbol *evaluate_string(struct expression *expr)
       array->ctype.alignment = 1;
       array->ctype.modifiers = MOD_STATIC;
       array->ctype.base_type = &char_ctype;
-
+
       addr->symbol = sym;
       addr->ctype = &lazy_ptr_ctype;
 
       expr->type = EXPR_PREOP;
       expr->op = '*';
-       expr->unop = addr;
+       expr->unop = addr;
       expr->ctype = sym;
       return sym;
 }
@@ -579,7 +579,7 @@ static struct symbol *evaluate_ptr_add(struct expression *expr, struct symbol *i
       }
 
       /* Get the size of whatever the pointer points to */
-       multiply = base->bit_size >> 3;
+       multiply = base->bit_size / bits_in_char;
 
       if (ctype == &null_ctype)
               ctype = &ptr_ctype;
@@ -831,7 +831,7 @@ static struct symbol *evaluate_ptr_sub(struct expression *expr)
               struct expression *sub = alloc_expression(expr->pos, EXPR_BINOP);
               struct expression *div = expr;
               struct expression *val = alloc_expression(expr->pos, EXPR_VALUE);
-               unsigned long value = lbase->bit_size >> 3;
+               unsigned long value = lbase->bit_size / bits_in_char;
 
               val->ctype = size_t_ctype;
               val->value = value;
@@ -850,7 +850,7 @@ static struct symbol *evaluate_ptr_sub(struct expression *expr)
               div->left = sub;
               div->right = val;
       }
-
+
       return ssize_t_ctype;
 }
 
@@ -1591,7 +1591,7 @@ static struct symbol *degenerate(struct expression *expr)
                               e3->op = '+';
                               e3->left = e0;
                               e3->right = alloc_const_expression(expr->pos,
-                                                       expr->r_bitpos >> 3);
+                                                       expr->r_bitpos / bits_in_char);
                               e3->ctype = &lazy_ptr_ctype;
                       } else {
                               e3 = e0;
@@ -1727,7 +1727,7 @@ static struct symbol *evaluate_postop(struct expression *expr)
       } else if (class == TYPE_PTR) {
               struct symbol *target = examine_pointer_target(ctype);
               if (!is_function(target))
-                       multiply = target->bit_size >> 3;
+                       multiply = target->bit_size / bits_in_char;
       }
 
       if (multiply) {
@@ -1850,7 +1850,7 @@ static struct symbol *find_identifier(struct ident *ident, struct symbol_list *_
                                       continue;
                               *offset += sym->offset;
                               return sub;
-                       }
+                       }
               }
       } while ((list = list->next) != head);
       return NULL;
@@ -1949,7 +1949,7 @@ static struct symbol *evaluate_member_dereference(struct expression *expr)
                       expr->base = deref->base;
                       expr->r_bitpos = deref->r_bitpos;
               }
-               expr->r_bitpos += offset << 3;
+               expr->r_bitpos += offset * bits_in_char;
               expr->type = EXPR_SLICE;
               expr->r_nrbits = member->bit_size;
               expr->r_bitpos += member->bit_offset;
@@ -2040,7 +2040,7 @@ static struct symbol *evaluate_sizeof(struct expression *expr)
       if ((size < 0) || (size & 7))
               expression_error(expr, "cannot size expression");
       expr->type = EXPR_VALUE;
-       expr->value = size >> 3;
+       expr->value = size / bits_in_char;
       expr->taint = 0;
       expr->ctype = size_t_ctype;
       return size_t_ctype;
@@ -2074,7 +2074,7 @@ static struct symbol *evaluate_ptrsizeof(struct expression *expr)
       if (size & 7)
               size = 0;
       expr->type = EXPR_VALUE;
-       expr->value = size >> 3;
+       expr->value = size / bits_in_char;
       expr->taint = 0;
       expr->ctype = size_t_ctype;
       return size_t_ctype;
@@ -2865,7 +2865,7 @@ static struct symbol *evaluate_offsetof(struct expression *expr)
                       unrestrict(idx, i_class, &i_type);
                       idx = cast_to(idx, size_t_ctype);
                       m = alloc_const_expression(expr->pos,
-                                                  ctype->bit_size >> 3);
+                                                  ctype->bit_size / bits_in_char);
                       m->ctype = size_t_ctype;
                       m->flags = Int_const_expr;
                       expr->type = EXPR_BINOP;
example.c +12 -12
index ae897dc..b844d4f 100644
@@ -27,7 +27,7 @@ static const char *opcodes[] = {
       [OP_INVOKE] = "invoke",
       [OP_COMPUTEDGOTO] = "jmp *",
       [OP_UNWIND] = "unwind",
-
+
       /* Binary */
       [OP_ADD] = "add",
       [OP_SUB] = "sub",
@@ -40,7 +40,7 @@ static const char *opcodes[] = {
       [OP_SHL] = "shl",
       [OP_LSR] = "lsr",
       [OP_ASR] = "asr",
-
+
       /* Logical */
       [OP_AND] = "and",
       [OP_OR] = "or",
@@ -66,7 +66,7 @@ static const char *opcodes[] = {
 
       /* Special three-input */
       [OP_SEL] = "select",
-
+
       /* Memory */
       [OP_MALLOC] = "malloc",
       [OP_FREE] = "free",
@@ -818,7 +818,7 @@ static const char *generic(struct bb_state *state, pseudo_t pseudo)
               put_operand(state, op);
               reg = target_reg(state, pseudo, NULL);
               output_insn(state, "lea %s,%s", show_op(state, op), reg->name);
-               return reg->name;
+               return reg->name;
 
       default:
               str = show_op(state, op);
@@ -870,7 +870,7 @@ static void kill_dead_reg(struct hardreg *reg)
 {
       if (reg->dead) {
               pseudo_t p;
-
+
               FOR_EACH_PTR(reg->contains, p) {
                       if (CURRENT_TAG(p) & TAG_DEAD) {
                               DELETE_CURRENT_PTR(p);
@@ -1011,7 +1011,7 @@ static void kill_pseudo(struct bb_state *state, pseudo_t pseudo)
                               continue;
                       if (CURRENT_TAG(p) & TAG_DEAD)
                               reg->dead--;
-                       output_comment(state, "removing pseudo %s from reg %s",
+                       output_comment(state, "removing pseudo %s from reg %s",
                               show_pseudo(pseudo), reg->name);
                       DELETE_CURRENT_PTR(p);
               } END_FOR_EACH_PTR(p);
@@ -1069,7 +1069,7 @@ static const char *conditional[] = {
       [OP_SET_BE] = "be",
       [OP_SET_AE] = "ae"
 };
-
+
 
 static void generate_branch(struct bb_state *state, struct instruction *br)
 {
@@ -1187,7 +1187,7 @@ static void replace_asm_percent(const char **src_p, char **dst_p, struct asm_arg
               if (index < nr)
                       replace_asm_arg(dst_p, args+index);
               break;
-       }
+       }
       *src_p = src;
       return;
 }
@@ -1583,7 +1583,7 @@ static int final_pseudo_flush(struct bb_state *state, pseudo_t pseudo, struct ha
       struct hardreg *dst;
 
       /*
-        * Since this pseudo is live at exit, we'd better have output
+        * Since this pseudo is live at exit, we'd better have output
        * storage for it..
        */
       hash = find_storage_hash(pseudo, state->outputs);
@@ -1829,8 +1829,8 @@ static void set_up_arch_entry(struct entrypoint *ep, struct instruction *entry)
 
                       in->type = REG_FRAME;
                       in->offset = offset;
-
-                       offset += bits >> 3;
+
+                       offset += bits / bits_in_char;
               }
               i++;
               NEXT_PTR_LIST(argtype);
@@ -1936,7 +1936,7 @@ static int compile(struct symbol_list *list)
               if (ep)
                       output(ep);
       } END_FOR_EACH_PTR(sym);
-
+
       return 0;
 }
 
expand.c +7 -7
index 032f0c5..d1aa019 100644
@@ -100,7 +100,7 @@ Int:
       // Stop here unless checking for truncation
       if (!Wcast_truncate || conservative)
               return;
-
+
       // Check if we dropped any bits..
       oldsignmask = 1ULL << (old_size-1);
       oldmask = oldsignmask | (oldsignmask-1);
@@ -179,7 +179,7 @@ static int simplify_int_binop(struct expression *expr, struct symbol *ctype)
               sl |= ~(mask-1);
       if (is_signed && (sr & mask))
               sr |= ~(mask-1);
-
+
       switch (CONVERT(expr->op,is_signed)) {
       case SIGNED('+'):
       case UNSIGNED('+'):
@@ -224,7 +224,7 @@ static int simplify_int_binop(struct expression *expr, struct symbol *ctype)
 
       case UNSIGNED('/'):
               if (!r) goto Div;
-               v = l / r;
+               v = l / r;
               break;
 
       case SIGNED('%'):
@@ -241,7 +241,7 @@ static int simplify_int_binop(struct expression *expr, struct symbol *ctype)
       case SIGNED(SPECIAL_LEFTSHIFT):
       case UNSIGNED(SPECIAL_LEFTSHIFT):
               v = l << r;
-               break;
+               break;
 
       case SIGNED(SPECIAL_RIGHTSHIFT):
               v = sl >> r;
@@ -536,7 +536,7 @@ static int expand_conditional(struct expression *expr)
 
       return cost + cond_cost + BRANCH_COST;
 }
-
+
 static int expand_assignment(struct expression *expr)
 {
       expand_expression(expr->left);
@@ -822,7 +822,7 @@ static int expand_expression_list(struct expression_list *list)
       return cost;
 }
 
-/*
+/*
 * We can simplify nested position expressions if
 * this is a simple (single) positional expression.
 */
@@ -880,7 +880,7 @@ static unsigned long bit_offset(const struct expression *expr)
 {
       unsigned long offset = 0;
       while (expr->type == EXPR_POS) {
-               offset += expr->init_offset << 3;
+               offset += expr->init_offset * bits_in_char;
               expr = expr->init_expr;
       }
       if (expr && expr->ctype)
flow.c +11 -9
index 82fb23a..bdb2c25 100644
@@ -16,6 +16,7 @@
 #include "expression.h"
 #include "linearize.h"
 #include "flow.h"
+#include "target.h"
 
 unsigned long bb_generation;
 
@@ -265,8 +266,8 @@ void convert_load_instruction(struct instruction *insn, pseudo_t src)
 
 static int overlapping_memop(struct instruction *a, struct instruction *b)
 {
-       unsigned int a_start = a->offset << 3;
-       unsigned int b_start = b->offset << 3;
+       unsigned int a_start = a->offset * bits_in_char;
+       unsigned int b_start = b->offset * bits_in_char;
       unsigned int a_size = a->size;
       unsigned int b_size = b->size;
 
@@ -364,11 +365,11 @@ found_dominator:
               use_pseudo(insn, phi, add_pseudo(dominators, phi));
       } END_FOR_EACH_PTR(parent);
       return 1;
-}
+}
 
 /*
 * We should probably sort the phi list just to make it easier to compare
- * later for equality.
+ * later for equality.
 */
 void rewrite_load_instruction(struct instruction *insn, struct pseudo_list *dominators)
 {
@@ -529,7 +530,7 @@ static void kill_dead_stores(pseudo_t pseudo, unsigned long generation, struct b
 * This should see if the "insn" trivially dominates some previous store, and kill the
 * store if unnecessary.
 */
-static void kill_dominated_stores(pseudo_t pseudo, struct instruction *insn,
+static void kill_dominated_stores(pseudo_t pseudo, struct instruction *insn,
       unsigned long generation, struct basic_block *bb, int local, int found)
 {
       struct instruction *one;
@@ -581,13 +582,14 @@ void check_access(struct instruction *insn)
       pseudo_t pseudo = insn->src;
 
       if (insn->bb && pseudo->type == PSEUDO_SYM) {
-               int offset = insn->offset, bit = (offset << 3) + insn->size;
+               int offset = insn->offset, bit = (offset * bits_in_char) + insn->size;
               struct symbol *sym = pseudo->sym;
 
               if (sym->bit_size > 0 && (offset < 0 || bit > sym->bit_size))
                       warning(insn->pos, "invalid access %s '%s' (%d %d)",
                               offset < 0 ? "below" : "past the end of",
-                               show_ident(sym->ident), offset, sym->bit_size >> 3);
+                               show_ident(sym->ident), offset,
+                               sym->bit_size / bits_in_char);
       }
 }
 
@@ -705,7 +707,7 @@ external_visibility:
                       } END_FOR_EACH_PTR(bb);
               }
       }
-
+
       return;
 }
 
@@ -887,7 +889,7 @@ static void vrfy_children(struct basic_block *bb)
       default:
               break;
       }
-
+
       FOR_EACH_PTR(bb->children, tmp) {
               vrfy_bb_in_list(bb, tmp->parents);
       } END_FOR_EACH_PTR(tmp);
show-parse.c +14 -14
index 064af32..63d9144 100644
@@ -246,8 +246,8 @@ deeper:
 
               s = modifier_string(mod);
               len = strlen(s);
-               name->start -= len;
-               memcpy(name->start, s, len);
+               name->start -= len;
+               memcpy(name->start, s, len);
               mod = 0;
               as = 0;
       }
@@ -412,7 +412,7 @@ void show_symbol(struct symbol *sym)
               struct statement *stmt = type->stmt;
               if (stmt) {
                       int val;
-                       printf("\n");
+                       printf("\n");
                       val = show_statement(stmt);
                       if (val)
                               printf("\tmov.%d\t\tretval,%d\n", stmt->ret->bit_size, val);
@@ -589,7 +589,7 @@ int show_statement(struct statement *stmt)
               if (pre_condition) {
                       if (pre_condition->type == EXPR_VALUE) {
                               if (!pre_condition->value) {
-                                       loop_bottom = new_label();
+                                       loop_bottom = new_label();
                                       printf("\tjmp\t\t.L%d\n", loop_bottom);
                               }
                       } else {
@@ -623,7 +623,7 @@ int show_statement(struct statement *stmt)
       }
       case STMT_NONE:
               break;
-
+
       case STMT_LABEL:
               printf(".L%p:\n", stmt->label_identifier);
               show_statement(stmt->label_statement);
@@ -649,9 +649,9 @@ int show_statement(struct statement *stmt)
               int val = show_expression(stmt->range_expression);
               int low = show_expression(stmt->range_low);
               int high = show_expression(stmt->range_high);
-               printf("\trange( %d %d-%d)\n", val, low, high);
+               printf("\trange( %d %d-%d)\n", val, low, high);
               break;
-       }
+       }
       }
       return 0;
 }
@@ -673,7 +673,7 @@ static int show_call_expression(struct expression *expr)
               int new = show_expression(arg);
               int size = arg->ctype->bit_size;
               printf("\tpush.%d\t\tv%d\n", size, new);
-               framesize += size >> 3;
+               framesize += size / bits_in_char;
       } END_FOR_EACH_PTR_REVERSE(arg);
 
       fn = expr->fn;
@@ -845,7 +845,7 @@ static int show_inc_dec(struct expression *expr, int postop)
       printf("\t%s.%d\t\tv%d,v%d,$1\n", opname, bits, new, retval);
       show_store_gen(bits, new, expr->unop, addr);
       return retval;
-}
+}
 
 static int show_preop(struct expression *expr)
 {
@@ -864,7 +864,7 @@ static int show_preop(struct expression *expr)
 static int show_postop(struct expression *expr)
 {
       return show_inc_dec(expr, 1);
-}
+}
 
 static int show_symbol_expr(struct symbol *sym)
 {
@@ -918,7 +918,7 @@ static int show_cast_expr(struct expression *expr)
 
       old_type = expr->cast_expression->ctype;
       new_type = expr->cast_type;
-
+
       oldbits = old_type->bit_size;
       newbits = new_type->bit_size;
       if (oldbits >= newbits)
@@ -1017,7 +1017,7 @@ again:
                       entry = entry->ident_expression;
                       goto again;
               }
-
+
               if (entry->type == EXPR_INDEX) {
                       printf(" AT '%d..%d:\n", entry->idx_from, entry->idx_to);
                       entry = entry->idx_expression;
@@ -1057,11 +1057,11 @@ int show_expression(struct expression *expr)
                       pos->line, pos->pos);
               return 0;
       }
-
+
       switch (expr->type) {
       case EXPR_CALL:
               return show_call_expression(expr);
-
+
       case EXPR_ASSIGNMENT:
               return show_assignment(expr);
 
symbol.c +9 -9
index 3292907..a22731f 100644
@@ -22,7 +22,7 @@
 
 /*
 * Secondary symbol list for stuff that needs to be output because it
- * was used.
+ * was used.
 */
 struct symbol_list *translation_unit_used_list = NULL;
 
@@ -117,7 +117,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
       }
 
       bit_size = info->bit_size;
-       base_size = sym->bit_size;
+       base_size = sym->bit_size;
 
       /*
        * Unsized arrays cause us to not align the resulting
@@ -128,7 +128,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
               base_size = 0;
       }
 
-       align_bit_mask = (sym->ctype.alignment << 3) - 1;
+       align_bit_mask = (sym->ctype.alignment * bits_in_char) - 1;
 
       /*
        * Bitfields have some very special rules..
@@ -143,7 +143,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
                       bit_size = (bit_size + align_bit_mask) & ~align_bit_mask;
                       bit_offset = 0;
               }
-               sym->offset = (bit_size - bit_offset) >> 3;
+               sym->offset = (bit_size - bit_offset) / bits_in_char;
               sym->bit_offset = bit_offset;
               sym->ctype.base_type->bit_offset = bit_offset;
               info->bit_size = bit_size + width;
@@ -156,7 +156,7 @@ static void lay_out_struct(struct symbol *sym, struct struct_union_info *info)
        * Otherwise, just align it right and add it up..
        */
       bit_size = (bit_size + align_bit_mask) & ~align_bit_mask;
-       sym->offset = bit_size >> 3;
+       sym->offset = bit_size / bits_in_char;
 
       info->bit_size = bit_size + base_size;
       // warning (sym->pos, "regular: offset=%d", sym->offset);
@@ -182,7 +182,7 @@ static struct symbol * examine_struct_union_type(struct symbol *sym, int advance
               sym->ctype.alignment = info.max_align;
       bit_size = info.bit_size;
       if (info.align_size) {
-               bit_align = (sym->ctype.alignment << 3)-1;
+               bit_align = (sym->ctype.alignment * bits_in_char)-1;
               bit_size = (bit_size + bit_align) & ~bit_align;
       }
       sym->bit_size = bit_size;
@@ -340,7 +340,7 @@ static struct symbol * examine_node_type(struct symbol *sym)
               if (node_type && node_type->bit_size >= 0)
                       bit_size = node_type->bit_size * count;
       }
-
+
       sym->bit_size = bit_size;
       return sym;
 }
@@ -654,7 +654,7 @@ static int expand_warning(struct expression *expr, int cost)
       FOR_EACH_PTR (arglist, arg) {
               /*
                * Constant strings get printed out as a warning. By the
-                * time we get here, the EXPR_STRING has been fully
+                * time we get here, the EXPR_STRING has been fully
                * evaluated, so by now it's an anonymous symbol with a
                * string initializer.
                *
@@ -877,7 +877,7 @@ void init_ctype(void)
               struct symbol *sym = ctype->ptr;
               unsigned long bit_size = ctype->bit_size ? *ctype->bit_size : -1;
               unsigned long maxalign = ctype->maxalign ? *ctype->maxalign : 0;
-               unsigned long alignment = (bit_size + 7) >> 3;
+               unsigned long alignment = (bit_size + bits_in_char - 1) / bits_in_char;
 
               if (alignment > maxalign)
                       alignment = maxalign;
target.c +2 -0
index bf1bb8f..a0ee010 100644
@@ -11,6 +11,8 @@ struct symbol *ssize_t_ctype = &int_ctype;
 */
 int max_alignment = 16;
 
+int bits_in_unit = 8;
+
 /*
 * Integer data types
 */
tokenize.c +5 -5
index e72c56e..67daa97 100644
@@ -96,7 +96,7 @@ static char *charstr(char *ptr, unsigned char c, unsigned char escape, unsigned
       }
       if (!isdigit(next))
               return ptr + sprintf(ptr, "%o", c);
-
+
       return ptr + sprintf(ptr, "%03o", c);
 }
 
@@ -161,7 +161,7 @@ const char *show_token(const struct token *token)
       case TOKEN_STREAMEND:
               sprintf(buffer, "<end of '%s'>", stream_name(token->pos.stream));
               return buffer;
-
+
       default:
               return "WTF???";
       }
@@ -483,7 +483,7 @@ static int escapechar(int first, int type, stream_t *stream, int *valp)
                               int nr = 2;
                               value -= '0';
                               while (next >= '0' && next <= '9') {
-                                       value = (value << 3) + (next-'0');
+                                       value = (value*8) + (next-'0');
                                       next = nextchar(stream);
                                       if (!--nr)
                                               break;
@@ -572,7 +572,7 @@ static int get_string_token(int next, stream_t *stream)
       token_type(token) = TOKEN_STRING;
       token->string = string;
       add_token(stream);
-
+
       return next;
 }
 
@@ -875,7 +875,7 @@ static int get_one_identifier(int c, stream_t *stream)
       token->ident = ident;
       add_token(stream);
       return next;
-}
+}
 
 static int get_one_token(int c, stream_t *stream)
 {
#13 lore raw
Quoted 6 lines · Christopher Li

Your patch is white space damaged.

While you are there, you might want to consider macro

bits_to_byte(x)
byte_to_bits(x)

Yes, that's a good idea. I'll redo it.

(What's the right way to generate the patch? I made that with 'git show

Quoted 1 line

diff', which I thought should do the right thing.)

Signature · 4 lines
┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup
#15 lore raw
In reply to David Given

compile-i386.c static struct storage *emit_select_expr(struct · line 1588

Quoted 11 lines · David Given
compile-i386.c +1 -1
index 8526408..3bbc9c7 100644
@@ -1584,7 +1584,7 @@ static struct storage *emit_select_expr(struct
 expression *expr)
 	/*
 	 * Do the actual select: check the conditional for zero,
 	 * move false over true if zero
-	 */ 
+	 */

Don't include whitespace changes in unrelated patches. You have many whitespace changes in this patch; I haven't commented on all of them.

compile-i386.c static struct symbol *evaluate_sizeof(struct · line 2042

Quoted 4 lines
compile-i386.c +0 -0
@@ -2040,7 +2040,7 @@ static struct symbol *evaluate_sizeof(struct
 expression *expr)
 	if ((size < 0) || (size & 7))
 		expression_error(expr, "cannot size expression");

This "size & 7" represents an assumption about bits_in_char as well; it checks if the size represents an integral number of chars. You'll need to look for other instances of 7 as well, to catch cases like this.

compile-i386.c static struct symbol *evaluate_ptrsizeof(struct · line 2076

Quoted 4 lines
compile-i386.c +0 -0
@@ -2074,7 +2074,7 @@ static struct symbol *evaluate_ptrsizeof(struct
 expression *expr)
 	if (size & 7)
 		size = 0;

Same problem here.

compile-i386.c static void set_up_arch_entry(struct entrypoint · line 1834

Quoted 12 lines
compile-i386.c +2 -2
 	 * storage for it..
 	 */
 	hash = find_storage_hash(pseudo, state->outputs);
@@ -1829,8 +1829,8 @@ static void set_up_arch_entry(struct entrypoint
 *ep, struct instruction *entry)
 
 			in->type = REG_FRAME;
 			in->offset = offset;
-			
-			offset += bits >> 3;
+
+			offset += bits / bits_in_char;

The first line represents a whitespace-only change.

target.c struct symbol *ssize_t_ctype = &int_ctype; · line 15

Quoted 8 lines
target.c +2 -0
@@ -11,6 +11,8 @@ struct symbol *ssize_t_ctype = &int_ctype;
  */
 int max_alignment = 16;
 
+int bits_in_unit = 8;
+

Unused.

target.c static int escapechar(int first, int type, · line 487

Quoted 7 lines
target.c +1 -1
@@ -483,7 +483,7 @@ static int escapechar(int first, int type,
 stream_t *stream, int *valp)
 				int nr = 2;
 				value -= '0';
 				while (next >= '0' && next <= '9') {
-					value = (value << 3) + (next-'0');
+					value = (value*8) + (next-'0');

This does not relate to bits_per_char, and should not change.

- Josh Triplett

--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#14 lore raw
In reply to Christopher Li
Quoted 9 lines · Christopher Li

Your patch is white space damaged.

While you are there, you might want to consider macro

bits_to_byte(x)
byte_to_bits(x)

To isolate out the conversion. It should generate the same code. Marginally more readable.

Yes, good idea. Updated patch enclosed. I've also applied Josh Triplett's fixes.

(This one's generated manually because I've been totally unable to make git behave for me. Can anyone suggest a decent getting-started guide?)

Signature · 4 lines
┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup
sparse/compile-i386.c renamed from compile-i386.c +1 -1
diff -rwu ./compile-i386.c ../sparse/compile-i386.c
@@ -2081,7 +2081,7 @@
 		insn("pushl", new, NULL,
 		     !framesize ? "begin function call" : NULL);
 
-		framesize += size >> 3;
+		framesize += bits_to_bytes(size);
 	} END_FOR_EACH_PTR_REVERSE(arg);
 
 	fn = expr->fn;

Only in .: diff

sparse/evaluate.c renamed from evaluate.c +10 -10
diff -rwu ./evaluate.c ../sparse/evaluate.c
@@ -579,7 +579,7 @@
 	}
 
 	/* Get the size of whatever the pointer points to */
-	multiply = base->bit_size >> 3;
+	multiply = bits_to_bytes(base->bit_size);
 
 	if (ctype == &null_ctype)
 		ctype = &ptr_ctype;
@@ -831,7 +831,7 @@
 		struct expression *sub = alloc_expression(expr->pos, EXPR_BINOP);
 		struct expression *div = expr;
 		struct expression *val = alloc_expression(expr->pos, EXPR_VALUE);
-		unsigned long value = lbase->bit_size >> 3;
+		unsigned long value = bits_to_bytes(lbase->bit_size);
 
 		val->ctype = size_t_ctype;
 		val->value = value;
@@ -1591,7 +1591,7 @@
 				e3->op = '+';
 				e3->left = e0;
 				e3->right = alloc_const_expression(expr->pos,
-							expr->r_bitpos >> 3);
+							bits_to_bytes(expr->r_bitpos));
 				e3->ctype = &lazy_ptr_ctype;
 			} else {
 				e3 = e0;
@@ -1727,7 +1727,7 @@
 	} else if (class == TYPE_PTR) {
 		struct symbol *target = examine_pointer_target(ctype);
 		if (!is_function(target))
-			multiply = target->bit_size >> 3;
+			multiply = bits_to_bytes(target->bit_size);
 	}
 
 	if (multiply) {
@@ -1949,7 +1949,7 @@
 			expr->base = deref->base;
 			expr->r_bitpos = deref->r_bitpos;
 		}
-		expr->r_bitpos += offset << 3;
+		expr->r_bitpos += bytes_to_bits(offset);
 		expr->type = EXPR_SLICE;
 		expr->r_nrbits = member->bit_size;
 		expr->r_bitpos += member->bit_offset;
@@ -2037,10 +2037,10 @@
 		return NULL;
 
 	size = type->bit_size;
-	if ((size < 0) || (size & 7))
+	if ((size < 0) || (size & (bits_in_char - 1)))
 		expression_error(expr, "cannot size expression");
 	expr->type = EXPR_VALUE;
-	expr->value = size >> 3;
+	expr->value = bits_to_bytes(size);
 	expr->taint = 0;
 	expr->ctype = size_t_ctype;
 	return size_t_ctype;
@@ -2071,10 +2071,10 @@
 		return NULL;
 	}
 	size = type->bit_size;
-	if (size & 7)
+	if (size & (bits_in_char-1))
 		size = 0;
 	expr->type = EXPR_VALUE;
-	expr->value = size >> 3;
+	expr->value = bits_to_bytes(size);
 	expr->taint = 0;
 	expr->ctype = size_t_ctype;
 	return size_t_ctype;
@@ -2865,7 +2865,7 @@
 			unrestrict(idx, i_class, &i_type);
 			idx = cast_to(idx, size_t_ctype);
 			m = alloc_const_expression(expr->pos,
-						   ctype->bit_size >> 3);
+						   bits_to_bytes(ctype->bit_size));
 			m->ctype = size_t_ctype;
 			m->flags = Int_const_expr;
 			expr->type = EXPR_BINOP;
sparse/example.c renamed from example.c +1 -1
diff -rwu ./example.c ../sparse/example.c
@@ -1830,7 +1830,7 @@
 			in->type = REG_FRAME;
 			in->offset = offset;
 			
-			offset += bits >> 3;
+			offset += bits_to_bytes(bits);
 		}
 		i++;
 		NEXT_PTR_LIST(argtype);
sparse/expand.c renamed from expand.c +1 -1
diff -rwu ./expand.c ../sparse/expand.c
@@ -880,7 +880,7 @@
 {
 	unsigned long offset = 0;
 	while (expr->type == EXPR_POS) {
-		offset += expr->init_offset << 3;
+		offset += bytes_to_bits(expr->init_offset);
 		expr = expr->init_expr;
 	}
 	if (expr && expr->ctype)
sparse/flow.c renamed from flow.c +6 -4
diff -rwu ./flow.c ../sparse/flow.c
@@ -16,6 +16,7 @@
 #include "expression.h"
 #include "linearize.h"
 #include "flow.h"
+#include "target.h"
 
 unsigned long bb_generation;
 
@@ -265,8 +266,8 @@
 
 static int overlapping_memop(struct instruction *a, struct instruction *b)
 {
-	unsigned int a_start = a->offset << 3;
-	unsigned int b_start = b->offset << 3;
+	unsigned int a_start = bytes_to_bits(a->offset);
+	unsigned int b_start = bytes_to_bits(b->offset);
 	unsigned int a_size = a->size;
 	unsigned int b_size = b->size;
 
@@ -581,13 +582,14 @@
 	pseudo_t pseudo = insn->src;
 
 	if (insn->bb && pseudo->type == PSEUDO_SYM) {
-		int offset = insn->offset, bit = (offset << 3) + insn->size;
+		int offset = insn->offset, bit = bytes_to_bits(offset) + insn->size;
 		struct symbol *sym = pseudo->sym;
 
 		if (sym->bit_size > 0 && (offset < 0 || bit > sym->bit_size))
 			warning(insn->pos, "invalid access %s '%s' (%d %d)",
 				offset < 0 ? "below" : "past the end of",
-				show_ident(sym->ident), offset, sym->bit_size >> 3);
+				show_ident(sym->ident), offset,
+				bits_to_bytes(sym->bit_size));
 	}
 }
 
sparse/show-parse.c renamed from show-parse.c +1 -1
diff -rwu ./show-parse.c ../sparse/show-parse.c
@@ -673,7 +673,7 @@
 		int new = show_expression(arg);
 		int size = arg->ctype->bit_size;
 		printf("\tpush.%d\t\tv%d\n", size, new);
-		framesize += size >> 3;
+		framesize += bits_to_bytes(size);
 	} END_FOR_EACH_PTR_REVERSE(arg);
 
 	fn = expr->fn;
sparse/symbol.c renamed from symbol.c +5 -5
diff -rwu ./symbol.c ../sparse/symbol.c
@@ -128,7 +128,7 @@
 		base_size = 0;
 	}
 
-	align_bit_mask = (sym->ctype.alignment << 3) - 1;
+	align_bit_mask = bytes_to_bits(sym->ctype.alignment) - 1;
 
 	/*
 	 * Bitfields have some very special rules..
@@ -143,7 +143,7 @@
 			bit_size = (bit_size + align_bit_mask) & ~align_bit_mask;
 			bit_offset = 0;
 		}
-		sym->offset = (bit_size - bit_offset) >> 3;
+		sym->offset = bits_to_bytes(bit_size - bit_offset);
 		sym->bit_offset = bit_offset;
 		sym->ctype.base_type->bit_offset = bit_offset;
 		info->bit_size = bit_size + width;
@@ -156,7 +156,7 @@
 	 * Otherwise, just align it right and add it up..
 	 */
 	bit_size = (bit_size + align_bit_mask) & ~align_bit_mask;
-	sym->offset = bit_size >> 3;
+	sym->offset = bits_to_bytes(bit_size);
 
 	info->bit_size = bit_size + base_size;
 	// warning (sym->pos, "regular: offset=%d", sym->offset);
@@ -182,7 +182,7 @@
 		sym->ctype.alignment = info.max_align;
 	bit_size = info.bit_size;
 	if (info.align_size) {
-		bit_align = (sym->ctype.alignment << 3)-1;
+		bit_align = bytes_to_bits(sym->ctype.alignment)-1;
 		bit_size = (bit_size + bit_align) & ~bit_align;
 	}
 	sym->bit_size = bit_size;
@@ -877,7 +877,7 @@
 		struct symbol *sym = ctype->ptr;
 		unsigned long bit_size = ctype->bit_size ? *ctype->bit_size : -1;
 		unsigned long maxalign = ctype->maxalign ? *ctype->maxalign : 0;
-		unsigned long alignment = (bit_size + 7) >> 3;
+		unsigned long alignment = bits_to_bytes(bit_size + bits_in_char - 1);
 
 		if (alignment > maxalign)
 			alignment = maxalign;
sparse/target.h renamed from target.h +10 -0
diff -rwu ./target.h ../sparse/target.h
@@ -42,4 +42,14 @@
 extern int bits_in_enum;
 extern int enum_alignment;
 
+/*
+ * Helper functions for converting bits to bytes and vice versa.
+ */
+
+static inline int bits_to_bytes(int bits)
+{ return bits / bits_in_char; }
+
+static inline int bytes_to_bits(int bytes)
+{ return bytes * bits_in_char; }
+
 #endif
#16 lore raw
In reply to David Given

[Resent for the 3rd time. I seems I can't write to the list]

Quoted 10 lines · David Given

Christopher Li wrote:
[...]

Quoted 5 lines
Quoted 1 line

You are right that point out a bug (assumption) of sparse which byte is 8 bits.

Using bits_in_byte is instead of 8 is better there.
Using bits_in_char assumes char has same bits as byte. That is my read of the C spec.

Quoted 1 line

Yes, indeed, I'd managed to get my terminology muddled to the confusion of everyone.

Quoted 9 lines

Okay, I've gone and looked at the implementation of this stuff; it would appear that modifying the code to supply the back end with units scaled in something other than C chars is actually quite complicated, so I haven't tried to do that. However, I'm enclosing a patch that should, hopefully, fix the cases where it's assuming 8 bit bytes. Hopefully I've
managed to catch all the cases, without breaking the octal parse code...

(I haven't used git before; is this the right format?)

Yes, but it would have been better to have kept all the white space changes for a separate change so we could just see that relevant changes.

This introduces divides all over the place for all users, replacing a very cheap constant shift with an expensive divide. All sparse users would be paying the cost for a feature that is useful for 1 user, I think we should perhaps think carefully about this.

One obvious solution would be to introduce a BITS_IN_CHAR macro in target.h defaulting to 8 and let "exotic" architectures redefine it to bits_in_char. This is very similar to the approach GCC is using.

On a completely unrelated node, I'm excited to see your work on using sparse for compilation. Hopefully your experience will lead to it being easier for the next guy. I would like to see sparse target for a simple abstract machine, but I so far haven't had time to work on it.

Tommy

#18 lore raw
In reply to a message not in the archive

Tommy Thorn wrote:
[...]

Quoted 2 lines

Yes, but it would have been better to have kept all the white space changes for a separate change so we could just see that relevant changes.

Corrected; turned out it had been so long since I've used a diff that didn't use my alias I'd forgotten you have to explicitly tell it to ignore white space.

Quoted 4 lines

This introduces divides all over the place for all users, replacing a very cheap constant shift with an expensive divide. All sparse users would be paying the cost for a feature that is useful for 1 user, I think we should perhaps think carefully about this.

I really don't think this is an issue. Certainly, a division by a variable is more expensive than a shift by a constant... but you'd have to do billions of them before anyone actually started to notice. If it really did turn out to be an issue, it would be simple enough to change later to "value >> bits_in_char_power_of_2".

Quoted 3 lines

One obvious solution would be to introduce a BITS_IN_CHAR macro in target.h defaulting to 8 and let "exotic" architectures redefine it to bits_in_char. This is very similar to the approach GCC is using.

The problem with this approach is that it involves recompiling all of sparse for each target, which is a bit user unfriendly. Currently all configuration can be done at run time, which allows such nice features as having all potential back-ends share the same code via shared libraries.

Quoted 4 lines

On a completely unrelated node, I'm excited to see your work on using sparse for compilation. Hopefully your experience will lead to it being easier for the next guy. I would like to see sparse target for a simple abstract machine, but I so far haven't had time to work on it.

sparse is working rather well, and seems to be one of the better compiler front ends that I've found. The linearising support is particularly helpful. I have, however, been finding that the learning curve is rather steep --- the API shows a lot of signs of having grown organically over time --- and I've had to build big chunks of code that I'm sure are unnecessary. For example, in order to generate the right kind of code, I need to figure out whether a pseudo contains an integer, float or pointer. The only way I've found out to do this is to recursively follow the chain of pseudo->def pointers until I find an instruction with enough type information attached to it to figure it out. I'm sure there must be an easier way of doing this...

(BTW, given a function symbol, how do I find its return type? I can find the list of arguments, but there's nothing in struct symbol that seems to refer to the return value...)

I am intending to release my project once done, of course, and I hope people will find it useful, but I'm not sure how much general use it will be; I've basically ripped out and replaced the register allocator / storage mechanism from my back end as I didn't understand it and didn't need the complexity. That should probably all get rewritten properly at some stage, but right now I'm focusing on getting things working...

Signature · 4 lines
┌─── dg@cowlark.com ───── http://www.cowlark.com ─────
│ "I have always wished for my computer to be as easy to use as my
│ telephone; my wish has come true because I can no longer figure out
│ how to use my telephone." --- Bjarne Stroustrup
#19 lore raw
Quoted 16 lines · David Given

sparse is working rather well, and seems to be one of the better compiler front ends that I've found. The linearising support is particularly helpful. I have, however, been finding that the learning curve is rather steep --- the API shows a lot of signs of having grown organically over time --- and I've had to build big chunks of code that I'm sure are unnecessary. For example, in order to generate the right kind of code, I need to figure out whether a pseudo contains an integer, float or pointer. The only way I've found out to do this is to recursively follow the chain of pseudo->def pointers until I find an instruction with enough type information attached to it to figure it out. I'm sure there must be an easier way of doing this...

(BTW, given a function symbol, how do I find its return type? I can find the list of arguments, but there's nothing in struct symbol that seems to refer to the return value...)

Quoted 7 lines

I am intending to release my project once done, of course, and I hope people will find it useful, but I'm not sure how much general use it will be; I've basically ripped out and replaced the register allocator / storage mechanism from my back end as I didn't understand it and didn't need the complexity. That should probably all get rewritten properly at some stage, but right now I'm focusing on getting things working...

What I had in mind would not even perform register allocation or any attempt at optimization. Such only obscures how to use sparse as a compiler.

Cheers,
Tommy

#20 lore raw
In reply to Christopher Li
Quoted 4 lines · Christopher Li
Quoted 2 lines

In C, there is no type "byte" (unless you typedef oder #define it). "byte" is usually (but not necessarily) meant as "unsigned char".

In C spec, there is a concept of "byte". The union return by sizeof()

I stand corrected. Hmm, I need the time to read C99 thoroughly.

Quoted 6 lines

is byte. Char must fit in a byte. But char does not necessary have the same bits as byte. Char can have more.

C99: 3.6, 3.7.1

Because char can always fit in byte, sizeof(char) == 1.

But how can a char have more bits than a byte?

Quoted 9 lines
Quoted 6 lines

IIRC C specifies that sizeof() returns values measured in chars, but I don't believe it specifies any mapping between the size of chars and the underlying addressing units --- it should be possible to use 16-bit chars, for example, on an 8-bit byte system. > Using 32-bit ints, sizeof(int) would then return 2; but you wouldn't be able to access individual bytes from C.

sizeof() return value measure in _byte_.
C99: 6.5.3.4

Yes. But "sizeof(char)" is always 1 (as stated in the same chapter). So I see no real difference between "byte" and "char" (at least with the size of them).

Bernd

Signature · 3 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services