| 1 |
; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s |
| 2 |
|
| 3 |
; Test for correct placement of 'lvl' instructions |
| 4 |
|
| 5 |
; Function Attrs: nounwind readonly |
| 6 |
declare <256 x double> @llvm.ve.vl.vld.vssl(i64, ptr, i32) |
| 7 |
declare void @llvm.ve.vl.vst.vssl(<256 x double>, i64, ptr, i32) |
| 8 |
|
| 9 |
; Check that the backend can handle constant VL as well as parametric VL |
| 10 |
; sources. |
| 11 |
|
| 12 |
; Function Attrs: nounwind |
| 13 |
define void @switching_vl(i32 %evl, i32 %evl2, ptr %P, ptr %Q) { |
| 14 |
; CHECK-LABEL: switching_vl: |
| 15 |
; CHECK: # %bb.0: |
| 16 |
; CHECK-NEXT: lea %s4, 256 |
| 17 |
; CHECK-NEXT: lvl %s4 |
| 18 |
; CHECK-NEXT: vld %v0, 8, %s2 |
| 19 |
; CHECK-NEXT: and %s0, %s0, (32)0 |
| 20 |
; CHECK-NEXT: lvl %s0 |
| 21 |
; CHECK-NEXT: vst %v0, 16, %s3 |
| 22 |
; CHECK-NEXT: lea %s4, 128 |
| 23 |
; CHECK-NEXT: lvl %s4 |
| 24 |
; CHECK-NEXT: vld %v0, 16, %s2 |
| 25 |
; CHECK-NEXT: and %s1, %s1, (32)0 |
| 26 |
; CHECK-NEXT: lvl %s1 |
| 27 |
; CHECK-NEXT: vst %v0, 16, %s3 |
| 28 |
; CHECK-NEXT: lvl %s4 |
| 29 |
; CHECK-NEXT: vld %v0, 8, %s2 |
| 30 |
; CHECK-NEXT: lvl %s0 |
| 31 |
; CHECK-NEXT: vst %v0, 16, %s3 |
| 32 |
; CHECK-NEXT: b.l.t (, %s10) |
| 33 |
%l0 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 8, ptr %P, i32 256) |
| 34 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l0, i64 16, ptr %Q, i32 %evl) |
| 35 |
%l1 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 16, ptr %P, i32 128) |
| 36 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l1, i64 16, ptr %Q, i32 %evl2) |
| 37 |
%l2 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 8, ptr %P, i32 128) |
| 38 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l2, i64 16, ptr %Q, i32 %evl) |
| 39 |
ret void |
| 40 |
} |
| 41 |
|
| 42 |
; Check that no redundant 'lvl' is inserted when vector length does not change |
| 43 |
; in a basic block. |
| 44 |
|
| 45 |
; Function Attrs: nounwind |
| 46 |
define void @stable_vl(i32 %evl, ptr %P, ptr %Q) { |
| 47 |
; CHECK-LABEL: stable_vl: |
| 48 |
; CHECK: # %bb.0: |
| 49 |
; CHECK-NEXT: and %s0, %s0, (32)0 |
| 50 |
; CHECK-NEXT: lvl %s0 |
| 51 |
; CHECK-NEXT: vld %v0, 8, %s1 |
| 52 |
; CHECK-NEXT: vst %v0, 16, %s2 |
| 53 |
; CHECK-NEXT: vld %v0, 16, %s1 |
| 54 |
; CHECK-NEXT: vst %v0, 16, %s2 |
| 55 |
; CHECK-NEXT: vld %v0, 8, %s1 |
| 56 |
; CHECK-NEXT: vst %v0, 16, %s2 |
| 57 |
; CHECK-NEXT: b.l.t (, %s10) |
| 58 |
%l0 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 8, ptr %P, i32 %evl) |
| 59 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l0, i64 16, ptr %Q, i32 %evl) |
| 60 |
%l1 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 16, ptr %P, i32 %evl) |
| 61 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l1, i64 16, ptr %Q, i32 %evl) |
| 62 |
%l2 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 8, ptr %P, i32 %evl) |
| 63 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l2, i64 16, ptr %Q, i32 %evl) |
| 64 |
ret void |
| 65 |
} |
| 66 |
|
| 67 |
;;; Check the case we have a call in the middle of vector instructions. |
| 68 |
|
| 69 |
; Function Attrs: nounwind |
| 70 |
define void @call_invl(i32 %evl, ptr %P, ptr %Q) { |
| 71 |
; CHECK-LABEL: call_invl: |
| 72 |
; CHECK: .LBB{{[0-9]+}}_2: |
| 73 |
; CHECK-NEXT: st %s18, 288(, %s11) # 8-byte Folded Spill |
| 74 |
; CHECK-NEXT: st %s19, 296(, %s11) # 8-byte Folded Spill |
| 75 |
; CHECK-NEXT: st %s20, 304(, %s11) # 8-byte Folded Spill |
| 76 |
; CHECK-NEXT: or %s18, 0, %s1 |
| 77 |
; CHECK-NEXT: and %s20, %s0, (32)0 |
| 78 |
; CHECK-NEXT: lvl %s20 |
| 79 |
; CHECK-NEXT: vld %v0, 8, %s1 |
| 80 |
; CHECK-NEXT: or %s19, 0, %s2 |
| 81 |
; CHECK-NEXT: vst %v0, 16, %s2 |
| 82 |
; CHECK-NEXT: lea %s0, fun@lo |
| 83 |
; CHECK-NEXT: and %s0, %s0, (32)0 |
| 84 |
; CHECK-NEXT: lea.sl %s12, fun@hi(, %s0) |
| 85 |
; CHECK-NEXT: bsic %s10, (, %s12) |
| 86 |
; CHECK-NEXT: lvl %s20 |
| 87 |
; CHECK-NEXT: vld %v0, 16, %s18 |
| 88 |
; CHECK-NEXT: vst %v0, 16, %s19 |
| 89 |
; CHECK-NEXT: vld %v0, 8, %s18 |
| 90 |
; CHECK-NEXT: vst %v0, 16, %s19 |
| 91 |
; CHECK-NEXT: ld %s20, 304(, %s11) # 8-byte Folded Reload |
| 92 |
; CHECK-NEXT: ld %s19, 296(, %s11) # 8-byte Folded Reload |
| 93 |
; CHECK-NEXT: ld %s18, 288(, %s11) # 8-byte Folded Reload |
| 94 |
; CHECK-NEXT: or %s11, 0, %s9 |
| 95 |
%l0 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 8, ptr %P, i32 %evl) |
| 96 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l0, i64 16, ptr %Q, i32 %evl) |
| 97 |
call void @fun() |
| 98 |
%l1 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 16, ptr %P, i32 %evl) |
| 99 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l1, i64 16, ptr %Q, i32 %evl) |
| 100 |
%l2 = tail call <256 x double> @llvm.ve.vl.vld.vssl(i64 8, ptr %P, i32 %evl) |
| 101 |
tail call void @llvm.ve.vl.vst.vssl(<256 x double> %l2, i64 16, ptr %Q, i32 %evl) |
| 102 |
ret void |
| 103 |
} |
| 104 |
|
| 105 |
declare void @fun() |