liblzma: LZMA decoder: Get rid of next_state[].

It's not completely obvious if this is better in the decoder.
It should be good if compiler can avoid creating a branch
(like using CMOV on x86).

This also makes lzma_encoder.c use the new macros.

src/liblzma/lzma/lzma_common.h

@@ -83,6 +83,20 @@ typedef enum {
 				? (state) - 3 \
 				: (state) - 6))
 
+/// Like update_literal(state) but when it is already known that
+/// is_literal_state(state) is true.
+#define update_literal_normal(state) \
+	state = ((state) <= STATE_SHORTREP_LIT_LIT \
+			? STATE_LIT_LIT \
+			: (state) - 3);
+
+/// Like update_literal(state) but when it is already known that
+/// is_literal_state(state) is false.
+#define update_literal_matched(state) \
+	state = ((state) <= STATE_LIT_SHORTREP \
+			? (state) - 3 \
+			: (state) - 6);
+
 /// Indicate that the latest state was a match.
 #define update_match(state) \
 	state = ((state) < LIT_STATES ? STATE_LIT_MATCH : STATE_NONLIT_MATCH)

src/liblzma/lzma/lzma_decoder.c

@@ -307,24 +307,6 @@ lzma_decode(void *coder_ptr, lzma_dict *restrict dictptr,
 		might_finish_without_eopm = true;
 	}
 
-	// Lookup table used to update the literal state.
-	// Compared to other state updates, this would need two branches.
-	// The lookup table is used by both Resumable and Non-resumable modes.
-	static const lzma_lzma_state next_state[] = {
-		STATE_LIT_LIT,
-		STATE_LIT_LIT,
-		STATE_LIT_LIT,
-		STATE_LIT_LIT,
-		STATE_MATCH_LIT_LIT,
-		STATE_REP_LIT_LIT,
-		STATE_SHORTREP_LIT_LIT,
-		STATE_MATCH_LIT,
-		STATE_REP_LIT,
-		STATE_SHORTREP_LIT,
-		STATE_MATCH_LIT,
-		STATE_REP_LIT
-	};
-
 	// The main decoder loop. The "switch" is used to resume the decoder at
 	// correct location. Once resumed, the "switch" is no longer used.
 	// The decoder loops is split into two modes:
@@ -381,16 +363,18 @@ lzma_decode(void *coder_ptr, lzma_dict *restrict dictptr,
 					dict.pos, dict_get(&dict, 0));
 
 			if (is_literal_state(state)) {
+				update_literal_normal(state);
+
 				// Decode literal without match byte.
 				rc_bittree8(probs, 0);
 			} else {
+				update_literal_matched(state);
+
 				// Decode literal with match byte.
 				rc_matched_literal(probs,
 						dict_get(&dict, rep0));
 			}
 
-			state = next_state[state];
-
 			// Write decoded literal to dictionary
 			dict_put(&dict, symbol);
 			continue;
@@ -705,6 +689,8 @@ slow:
 			symbol = 1;
 
 			if (is_literal_state(state)) {
+				update_literal_normal(state);
+
 				// Decode literal without match byte.
 				// The "slow" version does not unroll
 				// the loop.
@@ -714,6 +700,8 @@ slow:
 							SEQ_LITERAL);
 				} while (symbol < (1 << 8));
 			} else {
+				update_literal_matched(state);
+
 				// Decode literal with match byte.
 				len = (uint32_t)(dict_get(&dict, rep0)) << 1;
 
@@ -742,8 +730,6 @@ slow:
 				} while (symbol < (1 << 8));
 			}
 
-			state = next_state[state];
-
 	case SEQ_LITERAL_WRITE:
 			if (dict_put_safe(&dict, symbol)) {
 				coder->sequence = SEQ_LITERAL_WRITE;

src/liblzma/lzma/lzma_encoder.c

@@ -54,18 +54,18 @@ literal(lzma_lzma1_encoder *coder, lzma_mf *mf, uint32_t position)
 	if (is_literal_state(coder->state)) {
 		// Previous LZMA-symbol was a literal. Encode a normal
 		// literal without a match byte.
+		update_literal_normal(coder->state);
 		rc_bittree(&coder->rc, subcoder, 8, cur_byte);
 	} else {
 		// Previous LZMA-symbol was a match. Use the last byte of
 		// the match as a "match byte". That is, compare the bits
 		// of the current literal and the match byte.
+		update_literal_matched(coder->state);
 		const uint8_t match_byte = mf->buffer[
 				mf->read_pos - coder->reps[0] - 1
 				- mf->read_ahead];
 		literal_matched(&coder->rc, subcoder, match_byte, cur_byte);
 	}
-
-	update_literal(coder->state);
 }