A couple of Aarch64 FIPS delocate fixes.

Clang 12 in opt mode produces a couple of assembly patterns that were
not handled by delocate.

Firstly, two-digit vector indexes were just a simple omission. Fixed.

Secondly, Clang puts symbol deltas in .byte directives, and bit-shifts
them. The .byte directive was not considered to be a symbol-containing
directive because it's too small, but it could store deltas.
Additionally, bit-shifting of symbol expressions was not supported.
Fixed.

Change-Id: I796299821f5ac7d3639fa6243c5d9bd5342bbddf
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47064
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>

util/fipstools/delocate/delocate.peg

@@ -38,14 +38,19 @@ Arg <- QuotedArg / [[0-9a-z%+\-*_@.]]*
 QuotedArg <- '"' QuotedText '"'
 QuotedText <- (EscapedChar / [^"])*
 LabelContainingDirective <- LabelContainingDirectiveName WS SymbolArgs
-LabelContainingDirectiveName <- ".xword" / ".word" / ".long" / ".set" / ".8byte" / ".4byte" / ".quad" / ".tc" / ".localentry" / ".size" / ".type" / ".uleb128" / ".sleb128"
+LabelContainingDirectiveName <- ".xword" / ".word" / ".long" / ".set" / ".byte" / ".8byte" / ".4byte" / ".quad" / ".tc" / ".localentry" / ".size" / ".type" / ".uleb128" / ".sleb128"
 SymbolArgs <- SymbolArg ((WS? ',' WS?) SymbolArg)*
-SymbolArg <- Offset /
-             SymbolType /
-             (Offset / LocalSymbol / SymbolName / Dot) WS? Operator WS? (Offset / LocalSymbol / SymbolName) /
-             LocalSymbol TCMarker? /
-             SymbolName Offset /
-             SymbolName TCMarker?
+SymbolShift <- ('<<' / '>>') WS? [0-9]+
+SymbolArg <- (OpenParen WS?)? (
+               Offset /
+               SymbolType /
+               (Offset / LocalSymbol / SymbolName / Dot) WS? Operator WS? (Offset / LocalSymbol / SymbolName) /
+               LocalSymbol TCMarker? /
+               SymbolName Offset /
+               SymbolName TCMarker?)
+             (WS? CloseParen)? (WS? SymbolShift)?
+OpenParen <- '('
+CloseParen <- ')'
 SymbolType <- [@%] ('function' / 'object')
 Dot <- '.'
 TCMarker <- '[TC]'
@@ -73,7 +78,7 @@ RegisterOrConstant <- (('%'[[A-Z]][[A-Z0-9]]*) /
                        ARMRegister)
                       ![fb:(+\-]
 ARMConstantTweak <- ("lsl" / "sxtw" / "uxtw" / "uxtb" / "lsr" / "ror" / "asr") (WS '#' Offset)?
-ARMRegister <- "sp" / ([xwdqs] [0-9] [0-9]?) / "xzr" / "wzr" / ARMVectorRegister / ('{' WS? ARMVectorRegister (',' WS? ARMVectorRegister)* WS? '}' ('[' [0-9] ']')? )
+ARMRegister <- "sp" / ([xwdqs] [0-9] [0-9]?) / "xzr" / "wzr" / ARMVectorRegister / ('{' WS? ARMVectorRegister (',' WS? ARMVectorRegister)* WS? '}' ('[' [0-9] [0-9]? ']')? )
 ARMVectorRegister <- "v" [0-9] [0-9]? ('.' [0-9]* [bsdhq] ('[' [0-9] [0-9]? ']')? )?
 # Compilers only output a very limited number of expression forms. Rather than
 # implement a full expression parser, this enumerate those forms plus a few

util/fipstools/delocate/testdata/aarch64-Basic/in.s

@@ -43,6 +43,10 @@ foo:
 
 	bl bss_symbol_bss_get
 
+	# Regression test for a two-digit index.
+	ld1 { v1.b }[10], [x9]
+
+
 local_function:
 
 // BSS data

util/fipstools/delocate/testdata/aarch64-Basic/out.s

@@ -88,6 +88,10 @@ foo:
 
 	bl bss_symbol_bss_get
 
+	# Regression test for a two-digit index.
+	ld1 { v1.b }[10], [x9]
+
+
 .Llocal_function_local_target:
 local_function:
 

util/fipstools/delocate/testdata/x86_64-LabelRewrite/in2.s

@@ -12,3 +12,8 @@
 	.uleb128 .L2-.L1
 	.sleb128 .L2-.L1
 
+# .byte was not parsed as a symbol-containing directive on the
+# assumption that it's too small to hold a pointer. But Clang
+# will store offsets in it.
+.byte   (.LBB231_40-.LBB231_19)>>2, 4, .Lfoo, (.Lfoo), .Lfoo<<400, (   .Lfoo ) <<  66
+.byte   421

util/fipstools/delocate/testdata/x86_64-LabelRewrite/out.s

@@ -89,6 +89,12 @@ bar:
 # WAS .sleb128 .L2-.L1
 	.sleb128	.L2_BCM_1-.L1_BCM_1
 
+# .byte was not parsed as a symbol-containing directive on the
+# assumption that it's too small to hold a pointer. But Clang
+# will store offsets in it.
+# WAS .byte   (.LBB231_40-.LBB231_19)>>2, 4, .Lfoo, (.Lfoo), .Lfoo<<400, (   .Lfoo ) <<  66
+	.byte	(.LBB231_40_BCM_1-.LBB231_19_BCM_1)>>2, 4, .Lfoo_BCM_1, (.Lfoo_BCM_1), .Lfoo_BCM_1<<400, (   .Lfoo_BCM_1 ) <<  66
+.byte   421
 .text
 .loc 1 2 0
 BORINGSSL_bcm_text_end: