liblzma: Optimize the loop conditions in BCJ filters

Compilers cannot optimize the addition "i + 4" away since theoretically
it could overflow.

src/liblzma/simple/arm.c

@@ -18,8 +18,10 @@ arm_code(void *simple lzma_attribute((__unused__)),
 		uint32_t now_pos, bool is_encoder,
 		uint8_t *buffer, size_t size)
 {
+	size &= ~(size_t)3;
+
 	size_t i;
-	for (i = 0; i + 4 <= size; i += 4) {
+	for (i = 0; i < size; i += 4) {
 		if (buffer[i + 3] == 0xEB) {
 			uint32_t src = ((uint32_t)(buffer[i + 2]) << 16)
 					| ((uint32_t)(buffer[i + 1]) << 8)

src/liblzma/simple/arm64.c

@@ -28,6 +28,8 @@ arm64_code(void *simple lzma_attribute((__unused__)),
 		uint32_t now_pos, bool is_encoder,
 		uint8_t *buffer, size_t size)
 {
+	size &= ~(size_t)3;
+
 	size_t i;
 
 	// Clang 14.0.6 on x86-64 makes this four times bigger and 40 % slower
@@ -37,7 +39,7 @@ arm64_code(void *simple lzma_attribute((__unused__)),
 #ifdef __clang__
 #	pragma clang loop vectorize(disable)
 #endif
-	for (i = 0; i + 4 <= size; i += 4) {
+	for (i = 0; i < size; i += 4) {
 		uint32_t pc = (uint32_t)(now_pos + i);
 		uint32_t instr = read32le(buffer + i);
 

src/liblzma/simple/armthumb.c

@@ -18,8 +18,13 @@ armthumb_code(void *simple lzma_attribute((__unused__)),
 		uint32_t now_pos, bool is_encoder,
 		uint8_t *buffer, size_t size)
 {
+	if (size < 4)
+		return 0;
+
+	size -= 4;
+
 	size_t i;
-	for (i = 0; i + 4 <= size; i += 2) {
+	for (i = 0; i <= size; i += 2) {
 		if ((buffer[i + 1] & 0xF8) == 0xF0
 				&& (buffer[i + 3] & 0xF8) == 0xF8) {
 			uint32_t src = (((uint32_t)(buffer[i + 1]) & 7) << 19)

src/liblzma/simple/ia64.c

@@ -25,8 +25,10 @@ ia64_code(void *simple lzma_attribute((__unused__)),
 		4, 4, 0, 0, 4, 4, 0, 0
 	};
 
+	size &= ~(size_t)15;
+
 	size_t i;
-	for (i = 0; i + 16 <= size; i += 16) {
+	for (i = 0; i < size; i += 16) {
 		const uint32_t instr_template = buffer[i] & 0x1F;
 		const uint32_t mask = BRANCH_TABLE[instr_template];
 		uint32_t bit_pos = 5;

src/liblzma/simple/powerpc.c

@@ -18,8 +18,10 @@ powerpc_code(void *simple lzma_attribute((__unused__)),
 		uint32_t now_pos, bool is_encoder,
 		uint8_t *buffer, size_t size)
 {
+	size &= ~(size_t)3;
+
 	size_t i;
-	for (i = 0; i + 4 <= size; i += 4) {
+	for (i = 0; i < size; i += 4) {
 		// PowerPC branch 6(48) 24(Offset) 1(Abs) 1(Link)
 		if ((buffer[i] >> 2) == 0x12
 				&& ((buffer[i + 3] & 3) == 1)) {

src/liblzma/simple/sparc.c

@@ -18,9 +18,10 @@ sparc_code(void *simple lzma_attribute((__unused__)),
 		uint32_t now_pos, bool is_encoder,
 		uint8_t *buffer, size_t size)
 {
-	size_t i;
-	for (i = 0; i + 4 <= size; i += 4) {
+	size &= ~(size_t)3;
 
+	size_t i;
+	for (i = 0; i < size; i += 4) {
 		if ((buffer[i] == 0x40 && (buffer[i + 1] & 0xC0) == 0x00)
 				|| (buffer[i] == 0x7F
 				&& (buffer[i + 1] & 0xC0) == 0xC0)) {