[PATCH] D123458: [LSR][RISCV] Improve test coverage for LSR in RISC-V

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: llvm/test/CodeGen/RISCV/loop-strength-reduce-add-cheaper-than-mul.ll:74
+; RV64-NEXT:    mv a1, s0
+; RV64-NEXT:    call __muldi3 at plt
+; RV64-NEXT:    addw a0, a0, s2
----------------
Please use the -mattr=+m to avoid this libcall


================
Comment at: llvm/test/CodeGen/RISCV/loop-strength-reduce-ivusers.ll:48
+; RV64-NEXT:  .LBB0_2: # %bb29.i38
+; RV64-NEXT:    # Parent Loop BB0_1 Depth=1
+; RV64-NEXT:    # => This Inner Loop Header: Depth=2
----------------
Why is this loop substantially simpler than the RV32 version? I suspect LSR doesn't run on RV64, but that means that LSR is making the RV32 loop worse than we would get without it?


================
Comment at: llvm/test/CodeGen/RISCV/loop-strength-reduce-ivusers.ll:57
+bb4:		; preds = %bb4, %entry
+	br i1 true, label %bb5, label %bb4
+
----------------
Are these branch on true needed?


================
Comment at: llvm/test/CodeGen/RISCV/loop-strength-reduce-ivusers.ll:74
+	%1 = getelementptr [4 x [21 x double]], [4 x [21 x double]]* null, i32 0, i32 %0, i32 %sfb.314.i		; <double*> [#uses=1]
+	%2 = load double, double* %1, align 8		; <double> [#uses=0]
+	br i1 false, label %bb30.i41, label %bb33.i47
----------------
This load is dead. This whole test looks overly reduced.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123458/new/

https://reviews.llvm.org/D123458