Merge branch 'main' into users/mingmingl-llvm/samplefdo-profile-formatusers/mingmingl-llvm/samplefdo-profile-format

1 files changed, 188 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll b/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll
index 24853ebafefc..d504d418c150 100644
--- a/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll
+++ b/llvm/test/CodeGen/RISCV/rv64xtheadbb.ll
@@ -1122,3 +1122,191 @@ define i64 @bswap_i64(i64 %a) {
   %1 = call i64 @llvm.bswap.i64(i64 %a)
   ret i64 %1
 }
+
+define void @sextw_removal_ext(i32 signext %arg, i32 signext %arg1) nounwind {
+; RV64I-LABEL: sextw_removal_ext:
+; RV64I:       # %bb.0: # %bb
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sraw a0, a0, a1
+; RV64I-NEXT:  .LBB36_1: # %bb2
+; RV64I-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64I-NEXT:    call foo
+; RV64I-NEXT:    slli a0, a0, 16
+; RV64I-NEXT:    srai a0, a0, 32
+; RV64I-NEXT:    bnez a0, .LBB36_1
+; RV64I-NEXT:  # %bb.2: # %bb7
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+;
+; RV64XTHEADBB-LABEL: sextw_removal_ext:
+; RV64XTHEADBB:       # %bb.0: # %bb
+; RV64XTHEADBB-NEXT:    addi sp, sp, -16
+; RV64XTHEADBB-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64XTHEADBB-NEXT:    sraw a0, a0, a1
+; RV64XTHEADBB-NEXT:  .LBB36_1: # %bb2
+; RV64XTHEADBB-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64XTHEADBB-NEXT:    call foo
+; RV64XTHEADBB-NEXT:    th.ext a0, a0, 47, 16
+; RV64XTHEADBB-NEXT:    bnez a0, .LBB36_1
+; RV64XTHEADBB-NEXT:  # %bb.2: # %bb7
+; RV64XTHEADBB-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64XTHEADBB-NEXT:    addi sp, sp, 16
+; RV64XTHEADBB-NEXT:    ret
+bb:
+  %i = ashr i32 %arg, %arg1
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %bb
+  %i3 = phi i32 [ %i, %bb ], [ %i7, %bb2 ]
+  %i4 = tail call i64 @foo(i32 signext %i3)
+  %i5 = shl i64 %i4, 16
+  %i6 = ashr i64 %i5, 32
+  %i7 = trunc i64 %i6 to i32
+  %i8 = icmp eq i32 %i7, 0
+  br i1 %i8, label %bb7, label %bb2
+
+bb7:                                              ; preds = %bb2
+  ret void
+}
+
+declare i64 @foo(i32 signext)
+
+define void @sextw_removal_extu(i32 signext %arg, i32 signext %arg1) nounwind {
+; RV64I-LABEL: sextw_removal_extu:
+; RV64I:       # %bb.0: # %bb
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sraw a0, a0, a1
+; RV64I-NEXT:  .LBB37_1: # %bb2
+; RV64I-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64I-NEXT:    call foo
+; RV64I-NEXT:    slli a0, a0, 16
+; RV64I-NEXT:    srli a0, a0, 33
+; RV64I-NEXT:    bnez a0, .LBB37_1
+; RV64I-NEXT:  # %bb.2: # %bb7
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+;
+; RV64XTHEADBB-LABEL: sextw_removal_extu:
+; RV64XTHEADBB:       # %bb.0: # %bb
+; RV64XTHEADBB-NEXT:    addi sp, sp, -16
+; RV64XTHEADBB-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64XTHEADBB-NEXT:    sraw a0, a0, a1
+; RV64XTHEADBB-NEXT:  .LBB37_1: # %bb2
+; RV64XTHEADBB-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64XTHEADBB-NEXT:    call foo
+; RV64XTHEADBB-NEXT:    th.extu a0, a0, 47, 17
+; RV64XTHEADBB-NEXT:    bnez a0, .LBB37_1
+; RV64XTHEADBB-NEXT:  # %bb.2: # %bb7
+; RV64XTHEADBB-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64XTHEADBB-NEXT:    addi sp, sp, 16
+; RV64XTHEADBB-NEXT:    ret
+bb:
+  %i = ashr i32 %arg, %arg1
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %bb
+  %i3 = phi i32 [ %i, %bb ], [ %i7, %bb2 ]
+  %i4 = tail call i64 @foo(i32 signext %i3)
+  %i5 = shl i64 %i4, 16
+  %i6 = lshr i64 %i5, 33
+  %i7 = trunc i64 %i6 to i32
+  %i8 = icmp eq i32 %i7, 0
+  br i1 %i8, label %bb7, label %bb2
+
+bb7:                                              ; preds = %bb2
+  ret void
+}
+
+define signext i32 @hasAllNBitUsers_extu(i64 %arg1, i64 %arg2, i64 %arg3)  {
+; RV64I-LABEL: hasAllNBitUsers_extu:
+; RV64I:       # %bb.0: # %entry
+; RV64I-NEXT:    addi a2, a2, -1
+; RV64I-NEXT:    li a3, 256
+; RV64I-NEXT:  .LBB38_1: # %bb2
+; RV64I-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64I-NEXT:    slli a0, a0, 47
+; RV64I-NEXT:    srli a0, a0, 62
+; RV64I-NEXT:    addi a2, a2, 1
+; RV64I-NEXT:    addw a0, a0, a1
+; RV64I-NEXT:    bltu a2, a3, .LBB38_1
+; RV64I-NEXT:  # %bb.2: # %bb7
+; RV64I-NEXT:    ret
+;
+; RV64XTHEADBB-LABEL: hasAllNBitUsers_extu:
+; RV64XTHEADBB:       # %bb.0: # %entry
+; RV64XTHEADBB-NEXT:    addi a2, a2, -1
+; RV64XTHEADBB-NEXT:    li a3, 256
+; RV64XTHEADBB-NEXT:  .LBB38_1: # %bb2
+; RV64XTHEADBB-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64XTHEADBB-NEXT:    th.extu a0, a0, 16, 15
+; RV64XTHEADBB-NEXT:    addi a2, a2, 1
+; RV64XTHEADBB-NEXT:    addw a0, a0, a1
+; RV64XTHEADBB-NEXT:    bltu a2, a3, .LBB38_1
+; RV64XTHEADBB-NEXT:  # %bb.2: # %bb7
+; RV64XTHEADBB-NEXT:    ret
+entry:
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %entry
+  %i1 = phi i64 [ %arg1, %entry ], [ %i5, %bb2 ]
+  %i2 = phi i64 [ %arg3, %entry ], [ %i3, %bb2 ]
+  %i3 = add i64 %i2, 1
+  %i4 = lshr i64 %i1, 15
+  %i4b = and i64 %i4, 3
+  %i5 = add i64 %i4b, %arg2
+  %i6 = icmp ugt i64 %i2, 255
+  br i1 %i6, label %bb7, label %bb2
+
+bb7:                                              ; preds = %bb2
+  %i7 = trunc i64 %i5 to i32
+  ret i32 %i7
+}
+
+define signext i32 @hasAllNBitUsers_ext(i64 %arg1, i64 %arg2, i64 %arg3)  {
+; RV64I-LABEL: hasAllNBitUsers_ext:
+; RV64I:       # %bb.0: # %entry
+; RV64I-NEXT:    addi a2, a2, -1
+; RV64I-NEXT:    li a3, 256
+; RV64I-NEXT:  .LBB39_1: # %bb2
+; RV64I-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64I-NEXT:    slli a0, a0, 47
+; RV64I-NEXT:    srli a0, a0, 62
+; RV64I-NEXT:    addi a2, a2, 1
+; RV64I-NEXT:    addw a0, a0, a1
+; RV64I-NEXT:    bltu a2, a3, .LBB39_1
+; RV64I-NEXT:  # %bb.2: # %bb7
+; RV64I-NEXT:    ret
+;
+; RV64XTHEADBB-LABEL: hasAllNBitUsers_ext:
+; RV64XTHEADBB:       # %bb.0: # %entry
+; RV64XTHEADBB-NEXT:    addi a2, a2, -1
+; RV64XTHEADBB-NEXT:    li a3, 256
+; RV64XTHEADBB-NEXT:  .LBB39_1: # %bb2
+; RV64XTHEADBB-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV64XTHEADBB-NEXT:    th.extu a0, a0, 16, 15
+; RV64XTHEADBB-NEXT:    addi a2, a2, 1
+; RV64XTHEADBB-NEXT:    addw a0, a0, a1
+; RV64XTHEADBB-NEXT:    bltu a2, a3, .LBB39_1
+; RV64XTHEADBB-NEXT:  # %bb.2: # %bb7
+; RV64XTHEADBB-NEXT:    ret
+entry:
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %entry
+  %i1 = phi i64 [ %arg1, %entry ], [ %i5, %bb2 ]
+  %i2 = phi i64 [ %arg3, %entry ], [ %i3, %bb2 ]
+  %i3 = add i64 %i2, 1
+  %i4 = ashr i64 %i1, 15
+  %i4b = and i64 %i4, 3
+  %i5 = add i64 %i4b, %arg2
+  %i6 = icmp ugt i64 %i2, 255
+  br i1 %i6, label %bb7, label %bb2
+
+bb7:                                              ; preds = %bb2
+  %i7 = trunc i64 %i5 to i32
+  ret i32 %i7
+}