huff0: Use bmi1 on GOAMD64=v3 (#519)

Go v1.18 feature. Set `GOAMD64=v3` to enable. Nothing worth having separate codepaths for. Allows breaking dependency chain a bit. ``` benchmark old ns/op new ns/op delta BenchmarkDecompress4XNoTable/gettysburg/10000-32 11464 11195 -2.35% BenchmarkDecompress4XNoTable/gettysburg/262143-32 322679 319985 -0.83% BenchmarkDecompress4XNoTable/twain/10000-32 11505 11238 -2.32% BenchmarkDecompress4XNoTable/twain/262143-32 373751 370410 -0.89% BenchmarkDecompress4XNoTable/pngdata.001/10000-32 11957 11461 -4.15% BenchmarkDecompress4XNoTable/pngdata.001/262143-32 306403 300566 -1.91% ```
klauspost · Mar 9, 2022 · 0ff8ec1 · 0ff8ec1
1 parent 275e1fc
commit 0ff8ec1
Show file tree

Hide file tree

Showing 3 changed files with 244 additions and 57 deletions.
diff --git a/huff0/decompress_8b_amd64.s b/huff0/decompress_8b_amd64.s
@@ -137,8 +137,8 @@ skip_fill0:
 	ADDQ    CX, br_bits_read // bits_read += n
 
 	// these two writes get coalesced
-	// buf[stream][off] = uint8(v0.entry >> 8)
-	// buf[stream][off+1] = uint8(v1.entry >> 8)
+	// buf[stream][off+2] = uint8(v2.entry >> 8)
+	// buf[stream][off+3] = uint8(v3.entry >> 8)
 	MOVW BX, 0+2(buffer)(off*1)
 
 	// update the bitrader reader structure
@@ -241,8 +241,8 @@ skip_fill1:
 	ADDQ    CX, br_bits_read // bits_read += n
 
 	// these two writes get coalesced
-	// buf[stream][off] = uint8(v0.entry >> 8)
-	// buf[stream][off+1] = uint8(v1.entry >> 8)
+	// buf[stream][off+2] = uint8(v2.entry >> 8)
+	// buf[stream][off+3] = uint8(v3.entry >> 8)
 	MOVW BX, 256+2(buffer)(off*1)
 
 	// update the bitrader reader structure
@@ -345,8 +345,8 @@ skip_fill2:
 	ADDQ    CX, br_bits_read // bits_read += n
 
 	// these two writes get coalesced
-	// buf[stream][off] = uint8(v0.entry >> 8)
-	// buf[stream][off+1] = uint8(v1.entry >> 8)
+	// buf[stream][off+2] = uint8(v2.entry >> 8)
+	// buf[stream][off+3] = uint8(v3.entry >> 8)
 	MOVW BX, 512+2(buffer)(off*1)
 
 	// update the bitrader reader structure
@@ -449,8 +449,8 @@ skip_fill3:
 	ADDQ    CX, br_bits_read // bits_read += n
 
 	// these two writes get coalesced
-	// buf[stream][off] = uint8(v0.entry >> 8)
-	// buf[stream][off+1] = uint8(v1.entry >> 8)
+	// buf[stream][off+2] = uint8(v2.entry >> 8)
+	// buf[stream][off+3] = uint8(v3.entry >> 8)
 	MOVW BX, 768+2(buffer)(off*1)
 
 	// update the bitrader reader structure