[all-commits] [llvm/llvm-project] e6f313: [Codegen] Alter the default promotion for saturati...

[David Green via All-commits]

  Branch: refs/heads/master
  Home:   https://github.com/llvm/llvm-project
  Commit: e6f313b3807d23017d188aa7060b8cad09b3d095
      https://github.com/llvm/llvm-project/commit/e6f313b3807d23017d188aa7060b8cad09b3d095
  Author: David Green <david.green at arm.com>
  Date:   2019-10-18 (Fri, 18 Oct 2019)

  Changed paths:
    M llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
    M llvm/test/CodeGen/AArch64/sadd_sat.ll
    M llvm/test/CodeGen/AArch64/sadd_sat_plus.ll
    M llvm/test/CodeGen/AArch64/sadd_sat_vec.ll
    M llvm/test/CodeGen/AArch64/ssub_sat.ll
    M llvm/test/CodeGen/AArch64/ssub_sat_plus.ll
    M llvm/test/CodeGen/AArch64/ssub_sat_vec.ll
    M llvm/test/CodeGen/AArch64/uadd_sat.ll
    M llvm/test/CodeGen/AArch64/uadd_sat_plus.ll
    M llvm/test/CodeGen/AArch64/uadd_sat_vec.ll
    M llvm/test/CodeGen/AArch64/usub_sat.ll
    M llvm/test/CodeGen/AArch64/usub_sat_plus.ll
    M llvm/test/CodeGen/AArch64/usub_sat_vec.ll
    M llvm/test/CodeGen/ARM/sadd_sat.ll
    M llvm/test/CodeGen/ARM/sadd_sat_plus.ll
    M llvm/test/CodeGen/ARM/ssub_sat.ll
    M llvm/test/CodeGen/ARM/ssub_sat_plus.ll
    M llvm/test/CodeGen/ARM/uadd_sat.ll
    M llvm/test/CodeGen/ARM/uadd_sat_plus.ll
    M llvm/test/CodeGen/ARM/usub_sat.ll
    M llvm/test/CodeGen/ARM/usub_sat_plus.ll
    M llvm/test/CodeGen/X86/sadd_sat.ll
    M llvm/test/CodeGen/X86/sadd_sat_plus.ll
    M llvm/test/CodeGen/X86/ssub_sat.ll
    M llvm/test/CodeGen/X86/ssub_sat_plus.ll
    M llvm/test/CodeGen/X86/uadd_sat.ll
    M llvm/test/CodeGen/X86/uadd_sat_plus.ll
    M llvm/test/CodeGen/X86/usub_sat.ll
    M llvm/test/CodeGen/X86/usub_sat_plus.ll

  Log Message:
  -----------
  [Codegen] Alter the default promotion for saturating adds and subs

The default promotion for the add_sat/sub_sat nodes currently does:
    ANY_EXTEND iN to iM
    SHL by M-N
    [US][ADD|SUB]SAT
    L/ASHR by M-N

If the promoted add_sat or sub_sat node is not legal, this can produce code
that effectively does a lot of shifting (and requiring large constants to be
materialised) just to use the overflow flag. It is simpler to just do the
saturation manually, using the higher bitwidth addition and a min/max against
the saturating bounds. That is what this patch attempts to do.

Differential Revision: https://reviews.llvm.org/D68926

llvm-svn: 375211