My previous change to the Thumb-2 movsi patterns [1] caused a codesize regression with -Os in large functions. Fix this by using the literal pool offset of the 16-bit literal load so that the literal pool is dumped earlier, reducing the number of 32-bit literal loads.
Bootstrap & regress OK on arm-none-linux-gnueabihf --with-cpu=cortex-a57, committed as obvious. [1] https://gcc.gnu.org/ml/gcc-patches/2019-07/msg01579.html ChangeLog: 2019-07-30 Wilco Dijkstra <wdijk...@arm.com> * config/arm/thumb2.md (thumb2_movsi_insn): Adjust literal offset. * config/arm/vfp.md (thumb2_movsi_vfp): Likewise. --- diff --git a/gcc/config/arm/thumb2.md b/gcc/config/arm/thumb2.md index c124838c3a536244909223bd18735427cef7d969..6ccc875e2b4e7b8ce256e52da966dfe220c6f5d6 100644 --- a/gcc/config/arm/thumb2.md +++ b/gcc/config/arm/thumb2.md @@ -274,7 +274,7 @@ (define_insn "*thumb2_movsi_insn" (set_attr "length" "2,4,2,4,4,4,4") (set_attr "predicable" "yes") (set_attr "predicable_short_it" "yes,no,yes,no,no,no,no") - (set_attr "pool_range" "*,*,*,*,*,4094,*") + (set_attr "pool_range" "*,*,*,*,*,1018,*") (set_attr "neg_pool_range" "*,*,*,*,*,0,*")] ) diff --git a/gcc/config/arm/vfp.md b/gcc/config/arm/vfp.md index 9cc5946c79e027c0132d2b5e4fd6b97f13bca72e..661919e2357d352d12ff1020dc061f0c8d052841 100644 --- a/gcc/config/arm/vfp.md +++ b/gcc/config/arm/vfp.md @@ -297,7 +297,7 @@ (define_insn "*thumb2_movsi_vfp" (set_attr "predicable_short_it" "yes,no,yes,no,no,no,no,no,no,no,no,no") (set_attr "type" "mov_reg,mov_reg,mov_reg,mvn_reg,mov_imm,load_4,store_4,f_mcr,f_mrc,fmov,f_loads,f_stores") (set_attr "length" "2,4,2,4,4,4,4,4,4,4,4,4") - (set_attr "pool_range" "*,*,*,*,*,4094,*,*,*,*,1018,*") + (set_attr "pool_range" "*,*,*,*,*,1018,*,*,*,*,1018,*") (set_attr "neg_pool_range" "*,*,*,*,*, 0,*,*,*,*,1008,*")] )
