Tom Stellard | 1397d49 | 2016-02-11 21:45:07 +0000 | [diff] [blame] | 1 | ; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck %s |
| 2 | |
| 3 | ; FIXME: Move this to sgpr-copy.ll when this is fixed on VI. |
| 4 | ; Make sure that when we split an smrd instruction in order to move it to |
| 5 | ; the VALU, we are also moving its users to the VALU. |
| 6 | ; CHECK-LABEL: {{^}}split_smrd_add_worklist: |
Nikolay Haustov | 2f684f1 | 2016-02-26 09:51:05 +0000 | [diff] [blame] | 7 | ; CHECK: image_sample v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}] dmask:0x1 |
Tom Stellard | 1397d49 | 2016-02-11 21:45:07 +0000 | [diff] [blame] | 8 | |
Nicolai Haehnle | df3a20c | 2016-04-06 19:40:20 +0000 | [diff] [blame^] | 9 | define amdgpu_ps void @split_smrd_add_worklist([34 x <8 x i32>] addrspace(2)* byval %arg) #0 { |
Tom Stellard | 1397d49 | 2016-02-11 21:45:07 +0000 | [diff] [blame] | 10 | bb: |
| 11 | %tmp = call float @llvm.SI.load.const(<16 x i8> undef, i32 96) |
| 12 | %tmp1 = bitcast float %tmp to i32 |
| 13 | br i1 undef, label %bb2, label %bb3 |
| 14 | |
| 15 | bb2: ; preds = %bb |
| 16 | unreachable |
| 17 | |
| 18 | bb3: ; preds = %bb |
| 19 | %tmp4 = bitcast float %tmp to i32 |
| 20 | %tmp5 = add i32 %tmp4, 4 |
| 21 | %tmp6 = sext i32 %tmp5 to i64 |
| 22 | %tmp7 = getelementptr [34 x <8 x i32>], [34 x <8 x i32>] addrspace(2)* %arg, i64 0, i64 %tmp6 |
| 23 | %tmp8 = load <8 x i32>, <8 x i32> addrspace(2)* %tmp7, align 32, !tbaa !0 |
| 24 | %tmp9 = call <4 x float> @llvm.SI.image.sample.v2i32(<2 x i32> <i32 1061158912, i32 1048576000>, <8 x i32> %tmp8, <4 x i32> undef, i32 15, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0) |
| 25 | %tmp10 = extractelement <4 x float> %tmp9, i32 0 |
| 26 | %tmp12 = call i32 @llvm.SI.packf16(float %tmp10, float undef) |
| 27 | %tmp13 = bitcast i32 %tmp12 to float |
| 28 | call void @llvm.SI.export(i32 15, i32 1, i32 1, i32 0, i32 1, float undef, float %tmp13, float undef, float undef) |
| 29 | ret void |
| 30 | } |
| 31 | |
| 32 | ; Function Attrs: nounwind readnone |
| 33 | declare float @llvm.SI.load.const(<16 x i8>, i32) #1 |
| 34 | |
| 35 | declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float) |
| 36 | |
| 37 | declare <4 x float> @llvm.SI.image.sample.v2i32(<2 x i32>, <8 x i32>, <4 x i32>, i32, i32, i32, i32, i32, i32, i32, i32) #1 |
| 38 | |
| 39 | declare i32 @llvm.SI.packf16(float, float) #1 |
| 40 | |
Nicolai Haehnle | df3a20c | 2016-04-06 19:40:20 +0000 | [diff] [blame^] | 41 | attributes #0 = { "unsafe-fp-math"="true" } |
Tom Stellard | 1397d49 | 2016-02-11 21:45:07 +0000 | [diff] [blame] | 42 | attributes #1 = { nounwind readnone } |
| 43 | |
| 44 | !0 = !{!1, !1, i64 0, i32 1} |
| 45 | !1 = !{!"const", null} |
| 46 | !2 = !{!1, !1, i64 0} |