modules/dbe: simplify Lz4 encoder, improve naming

- Simplify Lz4 encoder by removing corner case handling required by Lz4
  block format. This allows to remove 1 FIFO, 2 FSM states, and a bunch
  of convoluted logic that can be easily moved to an external module.
- Hash function is factored out into a separate function
- Variable naming is improved

Signed-off-by: Roman Dobrodii <rdobrodii@antmicro.com>

xls/modules/dbe/BUILD

@@ -112,8 +112,6 @@ xls_dslx_library(
     deps = [
         ":dbe_common_dslx",
         "//xls/examples:ram_dslx",
-        # decoder is referenced by test code
-        ":dbe_lz4_decoder_dslx",
     ],
 )
 
@@ -136,7 +134,7 @@ xls_dslx_ir(
 xls_ir_opt_ir(
     name = "dbe_lz4_encoder_8k_opt_ir",
     src = "dbe_lz4_encoder_8k_ir.ir",
-    top = "__lz4_encoder__encoder_8k__encoder_base_0__16_3_13_12_13_65536_8192_4_16_8_next",
+    top = "__lz4_encoder__encoder_8k__encoder_base_0__1_4_13_16_16_8_next",
     ram_rewrites = [
         ":lz4_encoder_8k_ram1r1w_rewrites.textproto",
     ],

xls/modules/dbe/common.x

@@ -33,21 +33,24 @@ pub fn is_error(mark: Mark) -> bool {
 }
 
 pub enum TokenKind : u2 {
-    LITERAL = 0,
-    COPY_POINTER = 1,
-    MARK = 2,
+    UNMATCHED_SYMBOL = 0,
+    MATCHED_SYMBOL = 1,
+    MATCH = 2,
+    MARKER = 3,
 }
 
-pub struct Token<SYM_WIDTH: u32, PTR_WIDTH: u32, CNT_WIDTH: u32> {
+pub struct Token<
+    SYMBOL_WIDTH: u32, MATCH_OFFSET_WIDTH: u32, MATCH_COUNT_WIDTH: u32
+>{
     kind: TokenKind,
-    literal: uN[SYM_WIDTH],
-    copy_pointer_offset: uN[PTR_WIDTH],
-    copy_pointer_count: uN[CNT_WIDTH],
+    symbol: uN[SYMBOL_WIDTH],
+    match_offset: uN[MATCH_OFFSET_WIDTH],
+    match_count: uN[MATCH_COUNT_WIDTH],
     mark: Mark
 }
 
 pub struct PlainData<DATA_WIDTH: u32> {
-    is_mark: bool,
-    data: uN[DATA_WIDTH],    // symbol
-    mark: Mark,             // marker code
+    is_marker: bool,
+    data: uN[DATA_WIDTH],
+    mark: Mark,
 }