#16298: Awful(?) code in AArch64 stg_BLACKHOLE entry code -------------------------------------+------------------------------------- Reporter: bgamari | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.7 Keywords: | Operating System: Unknown/Multiple Architecture: aarch64 | Type of failure: Runtime | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- It's been a while since I've look at ARM assembler but this doesn't look right: {{{ Dump of assembler code for function stg_BLACKHOLE_info$def: 0x0000ffffb66d6118 <+0>: mov x23, x22 0x0000ffffb66d611c <+4>: ldr x22, [x22, #8] 0x0000ffffb66d6120 <+8>: tst x22, #0x7 0x0000ffffb66d6124 <+12>: b.ne 0xffffb66d6224 <stg_BLACKHOLE_info$def+268> // b.any 0x0000ffffb66d6128 <+16>: adrp x25, 0xffffb66fb000 0x0000ffffb66d612c <+20>: adrp x26, 0xffffb66fb000 0x0000ffffb66d6130 <+24>: adrp x27, 0xffffb66fb000 0x0000ffffb66d6134 <+28>: adrp x29, 0xffffb66fb000 0x0000ffffb66d6138 <+32>: ldr x25, [x25, #3880] 0x0000ffffb66d613c <+36>: ldr x26, [x26, #3888] 0x0000ffffb66d6140 <+40>: ldr x27, [x27, #3896] 0x0000ffffb66d6144 <+44>: ldr x29, [x29, #3904] 0x0000ffffb66d6148 <+48>: sub x24, x19, #0x18 0x0000ffffb66d614c <+52>: ldr x8, [x22] 0x0000ffffb66d6150 <+56>: cmp x8, x25 ... }}} Why do we `adrp` four times? Couldn't this be reused, with each `ldr` result going to a separate register? This was generated by LLVM 7. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16298> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler