maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/si-spill-cf.ll

504 lines
25 KiB
LLVM
Raw Normal View History

Enable FeatureFlatForGlobal on Volcanic Islands This switches to the workaround that HSA defaults to for the mesa path. This should be applied to the 4.0 branch. Patch by Vedran Miletić <vedran@miletic.net> llvm-svn: 292982
2017-01-25 06:02:15 +08:00
; RUN: llc -march=amdgcn < %s -verify-machineinstrs | FileCheck -check-prefix=SI %s
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
; RUN: llc -march=amdgcn -mcpu=tonga < %s -verify-machineinstrs | FileCheck -check-prefix=SI %s
; If this occurs it is likely due to reordering and the restore was
; originally supposed to happen before SI_END_CF.
AMDGPU: Remove more unused intrinsics Replace tests with lrp with basic IR expansion llvm-svn: 258612
2016-01-23 13:42:38 +08:00
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
; SI: s_or_b64 exec, exec, [[SAVED:s\[[0-9]+:[0-9]+\]|[a-z]+]]
; SI-NOT: v_readlane_b32 [[SAVED]]
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
AMDGPU: Remove superfluous string attributes from tests Also fix v_mac.ll not testing right thing for fneg llvm-svn: 275129
2016-07-12 07:35:48 +08:00
define amdgpu_ps void @main() #0 {
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
main_body:
AMDGPU: Convert tests away from llvm.SI.load.const llvm-svn: 351494
2019-01-18 06:47:26 +08:00
%tmp = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 16, i32 0)
%tmp1 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 32, i32 0)
%tmp2 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 80, i32 0)
%tmp3 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 84, i32 0)
%tmp4 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 88, i32 0)
%tmp5 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 96, i32 0)
%tmp6 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 100, i32 0)
%tmp7 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 104, i32 0)
%tmp8 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 112, i32 0)
%tmp9 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 116, i32 0)
%tmp10 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 120, i32 0)
%tmp11 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 128, i32 0)
%tmp12 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 132, i32 0)
%tmp13 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 136, i32 0)
%tmp14 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 144, i32 0)
%tmp15 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 148, i32 0)
%tmp16 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 152, i32 0)
%tmp17 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 160, i32 0)
%tmp18 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 164, i32 0)
%tmp19 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 168, i32 0)
%tmp20 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 176, i32 0)
%tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 180, i32 0)
%tmp22 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 184, i32 0)
%tmp23 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 192, i32 0)
%tmp24 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 196, i32 0)
%tmp25 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 200, i32 0)
%tmp26 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 208, i32 0)
%tmp27 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 212, i32 0)
%tmp28 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 216, i32 0)
%tmp29 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 224, i32 0)
%tmp30 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 228, i32 0)
%tmp31 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 232, i32 0)
%tmp32 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 240, i32 0)
%tmp33 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 244, i32 0)
%tmp34 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 248, i32 0)
%tmp35 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 256, i32 0)
%tmp36 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 260, i32 0)
%tmp37 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 264, i32 0)
%tmp38 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 272, i32 0)
%tmp39 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 276, i32 0)
%tmp40 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 280, i32 0)
%tmp41 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 288, i32 0)
%tmp42 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 292, i32 0)
%tmp43 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 296, i32 0)
%tmp44 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 304, i32 0)
%tmp45 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 308, i32 0)
%tmp46 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 312, i32 0)
%tmp47 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 320, i32 0)
%tmp48 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 324, i32 0)
%tmp49 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 328, i32 0)
%tmp50 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 336, i32 0)
%tmp51 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 340, i32 0)
%tmp52 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 344, i32 0)
%tmp53 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 352, i32 0)
%tmp54 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 356, i32 0)
%tmp55 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 360, i32 0)
%tmp56 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 368, i32 0)
%tmp57 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 372, i32 0)
%tmp58 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 376, i32 0)
%tmp59 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 384, i32 0)
%tmp60 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 388, i32 0)
%tmp61 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 392, i32 0)
%tmp62 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 400, i32 0)
%tmp63 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 404, i32 0)
%tmp64 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 408, i32 0)
%tmp65 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 416, i32 0)
%tmp66 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> undef, i32 420, i32 0)
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
br label %LOOP
LOOP: ; preds = %ENDIF2795, %main_body
%temp894.0 = phi float [ 0.000000e+00, %main_body ], [ %temp894.1, %ENDIF2795 ]
%temp18.0 = phi float [ undef, %main_body ], [ %temp18.1, %ENDIF2795 ]
AMDGPU: Remove superfluous string attributes from tests Also fix v_mac.ll not testing right thing for fneg llvm-svn: 275129
2016-07-12 07:35:48 +08:00
%tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp67 = icmp sgt i32 %tid, 4
br i1 %tmp67, label %ENDLOOP, label %ENDIF
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ENDLOOP: ; preds = %ELSE2566, %LOOP
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%one.sub.a.i = fsub float 1.000000e+00, %tmp
AMDGPU: Remove more unused intrinsics Replace tests with lrp with basic IR expansion llvm-svn: 258612
2016-01-23 13:42:38 +08:00
%one.sub.ac.i = fmul float %one.sub.a.i, undef
%result.i = fadd float fmul (float undef, float undef), %one.sub.ac.i
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float undef, float %result.i, float undef, float 1.000000e+00, i1 true, i1 true) #0
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ret void
ENDIF: ; preds = %LOOP
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp68 = fsub float %tmp2, undef
%tmp69 = fsub float %tmp3, undef
%tmp70 = fsub float %tmp4, undef
%tmp71 = fmul float %tmp68, 0.000000e+00
%tmp72 = fmul float %tmp69, undef
%tmp73 = fmul float %tmp70, undef
%tmp74 = fsub float %tmp6, undef
%tmp75 = fsub float %tmp7, undef
%tmp76 = fmul float %tmp74, undef
%tmp77 = fmul float %tmp75, 0.000000e+00
%tmp78 = call float @llvm.minnum.f32(float %tmp73, float %tmp77)
%tmp79 = call float @llvm.maxnum.f32(float %tmp71, float 0.000000e+00)
%tmp80 = call float @llvm.maxnum.f32(float %tmp72, float %tmp76)
%tmp81 = call float @llvm.maxnum.f32(float undef, float %tmp78)
%tmp82 = call float @llvm.minnum.f32(float %tmp79, float %tmp80)
%tmp83 = call float @llvm.minnum.f32(float %tmp82, float undef)
%tmp84 = fsub float %tmp14, undef
%tmp85 = fsub float %tmp15, undef
%tmp86 = fsub float %tmp16, undef
%tmp87 = fmul float %tmp84, undef
%tmp88 = fmul float %tmp85, undef
%tmp89 = fmul float %tmp86, undef
%tmp90 = fsub float %tmp17, undef
%tmp91 = fsub float %tmp18, undef
%tmp92 = fsub float %tmp19, undef
%tmp93 = fmul float %tmp90, 0.000000e+00
%tmp94 = fmul float %tmp91, undef
%tmp95 = fmul float %tmp92, undef
%tmp96 = call float @llvm.minnum.f32(float %tmp88, float %tmp94)
%tmp97 = call float @llvm.maxnum.f32(float %tmp87, float %tmp93)
%tmp98 = call float @llvm.maxnum.f32(float %tmp89, float %tmp95)
%tmp99 = call float @llvm.maxnum.f32(float undef, float %tmp96)
%tmp100 = call float @llvm.maxnum.f32(float %tmp99, float undef)
%tmp101 = call float @llvm.minnum.f32(float %tmp97, float undef)
%tmp102 = call float @llvm.minnum.f32(float %tmp101, float %tmp98)
%tmp103 = fsub float %tmp30, undef
%tmp104 = fsub float %tmp31, undef
%tmp105 = fmul float %tmp103, 0.000000e+00
%tmp106 = fmul float %tmp104, 0.000000e+00
%tmp107 = call float @llvm.minnum.f32(float undef, float %tmp105)
%tmp108 = call float @llvm.maxnum.f32(float undef, float %tmp106)
%tmp109 = call float @llvm.maxnum.f32(float undef, float %tmp107)
%tmp110 = call float @llvm.maxnum.f32(float %tmp109, float undef)
%tmp111 = call float @llvm.minnum.f32(float undef, float %tmp108)
%tmp112 = fsub float %tmp32, undef
%tmp113 = fsub float %tmp33, undef
%tmp114 = fsub float %tmp34, undef
%tmp115 = fmul float %tmp112, 0.000000e+00
%tmp116 = fmul float %tmp113, undef
%tmp117 = fmul float %tmp114, undef
%tmp118 = fsub float %tmp35, undef
%tmp119 = fsub float %tmp36, undef
%tmp120 = fsub float %tmp37, undef
%tmp121 = fmul float %tmp118, undef
%tmp122 = fmul float %tmp119, undef
%tmp123 = fmul float %tmp120, undef
%tmp124 = call float @llvm.minnum.f32(float %tmp115, float %tmp121)
%tmp125 = call float @llvm.minnum.f32(float %tmp116, float %tmp122)
%tmp126 = call float @llvm.minnum.f32(float %tmp117, float %tmp123)
%tmp127 = call float @llvm.maxnum.f32(float %tmp124, float %tmp125)
%tmp128 = call float @llvm.maxnum.f32(float %tmp127, float %tmp126)
%tmp129 = fsub float %tmp38, undef
%tmp130 = fsub float %tmp39, undef
%tmp131 = fsub float %tmp40, undef
%tmp132 = fmul float %tmp129, 0.000000e+00
%tmp133 = fmul float %tmp130, undef
%tmp134 = fmul float %tmp131, undef
%tmp135 = fsub float %tmp41, undef
%tmp136 = fsub float %tmp42, undef
%tmp137 = fsub float %tmp43, undef
%tmp138 = fmul float %tmp135, undef
%tmp139 = fmul float %tmp136, undef
%tmp140 = fmul float %tmp137, undef
%tmp141 = call float @llvm.minnum.f32(float %tmp132, float %tmp138)
%tmp142 = call float @llvm.minnum.f32(float %tmp133, float %tmp139)
%tmp143 = call float @llvm.minnum.f32(float %tmp134, float %tmp140)
%tmp144 = call float @llvm.maxnum.f32(float %tmp141, float %tmp142)
%tmp145 = call float @llvm.maxnum.f32(float %tmp144, float %tmp143)
%tmp146 = fsub float %tmp44, undef
%tmp147 = fsub float %tmp45, undef
%tmp148 = fsub float %tmp46, undef
%tmp149 = fmul float %tmp146, 0.000000e+00
%tmp150 = fmul float %tmp147, 0.000000e+00
%tmp151 = fmul float %tmp148, undef
%tmp152 = fsub float %tmp47, undef
%tmp153 = fsub float %tmp48, undef
%tmp154 = fsub float %tmp49, undef
%tmp155 = fmul float %tmp152, undef
%tmp156 = fmul float %tmp153, 0.000000e+00
%tmp157 = fmul float %tmp154, undef
%tmp158 = call float @llvm.minnum.f32(float %tmp149, float %tmp155)
%tmp159 = call float @llvm.minnum.f32(float %tmp150, float %tmp156)
%tmp160 = call float @llvm.minnum.f32(float %tmp151, float %tmp157)
%tmp161 = call float @llvm.maxnum.f32(float %tmp158, float %tmp159)
%tmp162 = call float @llvm.maxnum.f32(float %tmp161, float %tmp160)
%tmp163 = fsub float %tmp50, undef
%tmp164 = fsub float %tmp51, undef
%tmp165 = fsub float %tmp52, undef
%tmp166 = fmul float %tmp163, undef
%tmp167 = fmul float %tmp164, 0.000000e+00
%tmp168 = fmul float %tmp165, 0.000000e+00
%tmp169 = fsub float %tmp53, undef
%tmp170 = fsub float %tmp54, undef
%tmp171 = fsub float %tmp55, undef
%tmp172 = fdiv float 1.000000e+00, %temp18.0
%tmp173 = fmul float %tmp169, undef
%tmp174 = fmul float %tmp170, undef
%tmp175 = fmul float %tmp171, %tmp172
%tmp176 = call float @llvm.minnum.f32(float %tmp166, float %tmp173)
%tmp177 = call float @llvm.minnum.f32(float %tmp167, float %tmp174)
%tmp178 = call float @llvm.minnum.f32(float %tmp168, float %tmp175)
%tmp179 = call float @llvm.maxnum.f32(float %tmp176, float %tmp177)
%tmp180 = call float @llvm.maxnum.f32(float %tmp179, float %tmp178)
%tmp181 = fsub float %tmp62, undef
%tmp182 = fsub float %tmp63, undef
%tmp183 = fsub float %tmp64, undef
%tmp184 = fmul float %tmp181, 0.000000e+00
%tmp185 = fmul float %tmp182, undef
%tmp186 = fmul float %tmp183, undef
%tmp187 = fsub float %tmp65, undef
%tmp188 = fsub float %tmp66, undef
%tmp189 = fmul float %tmp187, undef
%tmp190 = fmul float %tmp188, undef
%tmp191 = call float @llvm.maxnum.f32(float %tmp184, float %tmp189)
%tmp192 = call float @llvm.maxnum.f32(float %tmp185, float %tmp190)
%tmp193 = call float @llvm.maxnum.f32(float %tmp186, float undef)
%tmp194 = call float @llvm.minnum.f32(float %tmp191, float %tmp192)
%tmp195 = call float @llvm.minnum.f32(float %tmp194, float %tmp193)
%.temp292.7 = select i1 undef, float %tmp162, float undef
%temp292.9 = select i1 false, float %tmp180, float %.temp292.7
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
%.temp292.9 = select i1 undef, float undef, float %temp292.9
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp196 = fcmp ogt float undef, 0.000000e+00
%tmp197 = fcmp olt float undef, %tmp195
%tmp198 = and i1 %tmp196, %tmp197
%tmp199 = fcmp olt float undef, %.temp292.9
%tmp200 = and i1 %tmp198, %tmp199
%temp292.11 = select i1 %tmp200, float undef, float %.temp292.9
%tid0 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
AMDGPU/SI: Detect uniform branches and emit s_cbranch instructions Reviewers: arsenm Subscribers: mareko, MatzeB, qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D16603 llvm-svn: 260765
2016-02-13 07:45:29 +08:00
%cmp0 = icmp eq i32 %tid0, 0
br i1 %cmp0, label %IF2565, label %ELSE2566
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
IF2565: ; preds = %ENDIF
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tid1 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
AMDGPU/SI: Detect uniform branches and emit s_cbranch instructions Reviewers: arsenm Subscribers: mareko, MatzeB, qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D16603 llvm-svn: 260765
2016-02-13 07:45:29 +08:00
%cmp1 = icmp eq i32 %tid1, 0
br i1 %cmp1, label %ENDIF2582, label %ELSE2584
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2566: ; preds = %ENDIF
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tid2 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
AMDGPU/SI: Detect uniform branches and emit s_cbranch instructions Reviewers: arsenm Subscribers: mareko, MatzeB, qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D16603 llvm-svn: 260765
2016-02-13 07:45:29 +08:00
%tidf = bitcast i32 %tid2 to float
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp201 = fcmp oeq float %temp292.11, %tidf
br i1 %tmp201, label %ENDLOOP, label %ELSE2593
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ENDIF2564: ; preds = %ENDIF2594, %ENDIF2588
%temp894.1 = phi float [ undef, %ENDIF2588 ], [ %temp894.2, %ENDIF2594 ]
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%temp18.1 = phi float [ %tmp218, %ENDIF2588 ], [ undef, %ENDIF2594 ]
%tmp202 = fsub float %tmp5, undef
%tmp203 = fmul float %tmp202, undef
%tmp204 = call float @llvm.maxnum.f32(float undef, float %tmp203)
%tmp205 = call float @llvm.minnum.f32(float %tmp204, float undef)
%tmp206 = call float @llvm.minnum.f32(float %tmp205, float undef)
%tmp207 = fcmp ogt float undef, 0.000000e+00
%tmp208 = fcmp olt float undef, 1.000000e+00
%tmp209 = and i1 %tmp207, %tmp208
%tid3 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
AMDGPU/SI: Detect uniform branches and emit s_cbranch instructions Reviewers: arsenm Subscribers: mareko, MatzeB, qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D16603 llvm-svn: 260765
2016-02-13 07:45:29 +08:00
%tidf3 = bitcast i32 %tid3 to float
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp210 = fcmp olt float %tidf3, %tmp206
%tmp211 = and i1 %tmp209, %tmp210
br i1 %tmp211, label %ENDIF2795, label %ELSE2797
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2584: ; preds = %IF2565
br label %ENDIF2582
ENDIF2582: ; preds = %ELSE2584, %IF2565
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp212 = fadd float %tmp1, undef
%tmp213 = fadd float 0.000000e+00, %tmp212
%floor = call float @llvm.floor.f32(float %tmp213)
%tmp214 = fsub float %tmp213, %floor
%tid4 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
AMDGPU/SI: Detect uniform branches and emit s_cbranch instructions Reviewers: arsenm Subscribers: mareko, MatzeB, qcolombet, arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D16603 llvm-svn: 260765
2016-02-13 07:45:29 +08:00
%cmp4 = icmp eq i32 %tid4, 0
br i1 %cmp4, label %IF2589, label %ELSE2590
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
IF2589: ; preds = %ENDIF2582
br label %ENDIF2588
ELSE2590: ; preds = %ENDIF2582
br label %ENDIF2588
ENDIF2588: ; preds = %ELSE2590, %IF2589
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp215 = fsub float 1.000000e+00, %tmp214
%tmp216 = call float @llvm.sqrt.f32(float %tmp215)
%tmp217 = fmul float %tmp216, undef
%tmp218 = fadd float %tmp217, undef
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
br label %ENDIF2564
ELSE2593: ; preds = %ELSE2566
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp219 = fcmp oeq float %temp292.11, %tmp81
%tmp220 = fcmp olt float %tmp81, %tmp83
%tmp221 = and i1 %tmp219, %tmp220
br i1 %tmp221, label %ENDIF2594, label %ELSE2596
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2596: ; preds = %ELSE2593
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp222 = fcmp oeq float %temp292.11, %tmp100
%tmp223 = fcmp olt float %tmp100, %tmp102
%tmp224 = and i1 %tmp222, %tmp223
br i1 %tmp224, label %ENDIF2594, label %ELSE2632
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ENDIF2594: ; preds = %ELSE2788, %ELSE2785, %ELSE2782, %ELSE2779, %IF2775, %ELSE2761, %ELSE2758, %IF2757, %ELSE2704, %ELSE2686, %ELSE2671, %ELSE2668, %IF2667, %ELSE2632, %ELSE2596, %ELSE2593
%temp894.2 = phi float [ 0.000000e+00, %IF2667 ], [ 0.000000e+00, %ELSE2671 ], [ 0.000000e+00, %IF2757 ], [ 0.000000e+00, %ELSE2761 ], [ %temp894.0, %ELSE2758 ], [ 0.000000e+00, %IF2775 ], [ 0.000000e+00, %ELSE2779 ], [ 0.000000e+00, %ELSE2782 ], [ %.2848, %ELSE2788 ], [ 0.000000e+00, %ELSE2785 ], [ 0.000000e+00, %ELSE2593 ], [ 0.000000e+00, %ELSE2632 ], [ 0.000000e+00, %ELSE2704 ], [ 0.000000e+00, %ELSE2686 ], [ 0.000000e+00, %ELSE2668 ], [ 0.000000e+00, %ELSE2596 ]
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp225 = fmul float %temp894.2, undef
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
br label %ENDIF2564
ELSE2632: ; preds = %ELSE2596
br i1 undef, label %ENDIF2594, label %ELSE2650
ELSE2650: ; preds = %ELSE2632
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp226 = fcmp oeq float %temp292.11, %tmp110
%tmp227 = fcmp olt float %tmp110, %tmp111
%tmp228 = and i1 %tmp226, %tmp227
br i1 %tmp228, label %IF2667, label %ELSE2668
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
IF2667: ; preds = %ELSE2650
br i1 undef, label %ENDIF2594, label %ELSE2671
ELSE2668: ; preds = %ELSE2650
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp229 = fcmp oeq float %temp292.11, %tmp128
%tmp230 = fcmp olt float %tmp128, undef
%tmp231 = and i1 %tmp229, %tmp230
br i1 %tmp231, label %ENDIF2594, label %ELSE2686
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2671: ; preds = %IF2667
br label %ENDIF2594
ELSE2686: ; preds = %ELSE2668
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp232 = fcmp oeq float %temp292.11, %tmp145
%tmp233 = fcmp olt float %tmp145, undef
%tmp234 = and i1 %tmp232, %tmp233
br i1 %tmp234, label %ENDIF2594, label %ELSE2704
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2704: ; preds = %ELSE2686
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp235 = fcmp oeq float %temp292.11, %tmp180
%tmp236 = fcmp olt float %tmp180, undef
%tmp237 = and i1 %tmp235, %tmp236
br i1 %tmp237, label %ENDIF2594, label %ELSE2740
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2740: ; preds = %ELSE2704
br i1 undef, label %IF2757, label %ELSE2758
IF2757: ; preds = %ELSE2740
br i1 undef, label %ENDIF2594, label %ELSE2761
ELSE2758: ; preds = %ELSE2740
br i1 undef, label %IF2775, label %ENDIF2594
ELSE2761: ; preds = %IF2757
br label %ENDIF2594
IF2775: ; preds = %ELSE2758
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp238 = fcmp olt float undef, undef
br i1 %tmp238, label %ENDIF2594, label %ELSE2779
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2779: ; preds = %IF2775
br i1 undef, label %ENDIF2594, label %ELSE2782
ELSE2782: ; preds = %ELSE2779
br i1 undef, label %ENDIF2594, label %ELSE2785
ELSE2785: ; preds = %ELSE2782
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp239 = fcmp olt float undef, 0.000000e+00
br i1 %tmp239, label %ENDIF2594, label %ELSE2788
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2788: ; preds = %ELSE2785
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp240 = fcmp olt float 0.000000e+00, undef
%.2848 = select i1 %tmp240, float -1.000000e+00, float 1.000000e+00
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
br label %ENDIF2594
ELSE2797: ; preds = %ENDIF2564
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp241 = fsub float %tmp8, undef
%tmp242 = fsub float %tmp9, undef
%tmp243 = fsub float %tmp10, undef
%tmp244 = fmul float %tmp241, undef
%tmp245 = fmul float %tmp242, undef
%tmp246 = fmul float %tmp243, undef
%tmp247 = fsub float %tmp11, undef
%tmp248 = fsub float %tmp12, undef
%tmp249 = fsub float %tmp13, undef
%tmp250 = fmul float %tmp247, undef
%tmp251 = fmul float %tmp248, undef
%tmp252 = fmul float %tmp249, undef
%tmp253 = call float @llvm.minnum.f32(float %tmp244, float %tmp250)
%tmp254 = call float @llvm.minnum.f32(float %tmp245, float %tmp251)
%tmp255 = call float @llvm.maxnum.f32(float %tmp246, float %tmp252)
%tmp256 = call float @llvm.maxnum.f32(float %tmp253, float %tmp254)
%tmp257 = call float @llvm.maxnum.f32(float %tmp256, float undef)
%tmp258 = call float @llvm.minnum.f32(float undef, float %tmp255)
%tmp259 = fcmp ogt float %tmp257, 0.000000e+00
%tmp260 = fcmp olt float %tmp257, 1.000000e+00
%tmp261 = and i1 %tmp259, %tmp260
%tmp262 = fcmp olt float %tmp257, %tmp258
%tmp263 = and i1 %tmp261, %tmp262
br i1 %tmp263, label %ENDIF2795, label %ELSE2800
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ENDIF2795: ; preds = %ELSE2824, %ELSE2821, %ELSE2818, %ELSE2815, %ELSE2812, %ELSE2809, %ELSE2806, %ELSE2803, %ELSE2800, %ELSE2797, %ENDIF2564
br label %LOOP
ELSE2800: ; preds = %ELSE2797
br i1 undef, label %ENDIF2795, label %ELSE2803
ELSE2803: ; preds = %ELSE2800
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp264 = fsub float %tmp20, undef
%tmp265 = fsub float %tmp21, undef
%tmp266 = fsub float %tmp22, undef
%tmp267 = fmul float %tmp264, undef
%tmp268 = fmul float %tmp265, undef
%tmp269 = fmul float %tmp266, 0.000000e+00
%tmp270 = fsub float %tmp23, undef
%tmp271 = fsub float %tmp24, undef
%tmp272 = fsub float %tmp25, undef
%tmp273 = fmul float %tmp270, undef
%tmp274 = fmul float %tmp271, undef
%tmp275 = fmul float %tmp272, undef
%tmp276 = call float @llvm.minnum.f32(float %tmp267, float %tmp273)
%tmp277 = call float @llvm.maxnum.f32(float %tmp268, float %tmp274)
%tmp278 = call float @llvm.maxnum.f32(float %tmp269, float %tmp275)
%tmp279 = call float @llvm.maxnum.f32(float %tmp276, float undef)
%tmp280 = call float @llvm.maxnum.f32(float %tmp279, float undef)
%tmp281 = call float @llvm.minnum.f32(float undef, float %tmp277)
%tmp282 = call float @llvm.minnum.f32(float %tmp281, float %tmp278)
%tmp283 = fcmp ogt float %tmp280, 0.000000e+00
%tmp284 = fcmp olt float %tmp280, 1.000000e+00
%tmp285 = and i1 %tmp283, %tmp284
%tmp286 = fcmp olt float %tmp280, %tmp282
%tmp287 = and i1 %tmp285, %tmp286
br i1 %tmp287, label %ENDIF2795, label %ELSE2806
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2806: ; preds = %ELSE2803
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp288 = fsub float %tmp26, undef
%tmp289 = fsub float %tmp27, undef
%tmp290 = fsub float %tmp28, undef
%tmp291 = fmul float %tmp288, undef
%tmp292 = fmul float %tmp289, 0.000000e+00
%tmp293 = fmul float %tmp290, undef
%tmp294 = fsub float %tmp29, undef
%tmp295 = fmul float %tmp294, undef
%tmp296 = call float @llvm.minnum.f32(float %tmp291, float %tmp295)
%tmp297 = call float @llvm.minnum.f32(float %tmp292, float undef)
%tmp298 = call float @llvm.maxnum.f32(float %tmp293, float undef)
%tmp299 = call float @llvm.maxnum.f32(float %tmp296, float %tmp297)
%tmp300 = call float @llvm.maxnum.f32(float %tmp299, float undef)
%tmp301 = call float @llvm.minnum.f32(float undef, float %tmp298)
%tmp302 = fcmp ogt float %tmp300, 0.000000e+00
%tmp303 = fcmp olt float %tmp300, 1.000000e+00
%tmp304 = and i1 %tmp302, %tmp303
%tmp305 = fcmp olt float %tmp300, %tmp301
%tmp306 = and i1 %tmp304, %tmp305
br i1 %tmp306, label %ENDIF2795, label %ELSE2809
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2809: ; preds = %ELSE2806
br i1 undef, label %ENDIF2795, label %ELSE2812
ELSE2812: ; preds = %ELSE2809
br i1 undef, label %ENDIF2795, label %ELSE2815
ELSE2815: ; preds = %ELSE2812
br i1 undef, label %ENDIF2795, label %ELSE2818
ELSE2818: ; preds = %ELSE2815
br i1 undef, label %ENDIF2795, label %ELSE2821
ELSE2821: ; preds = %ELSE2818
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
%tmp307 = fsub float %tmp56, undef
%tmp308 = fsub float %tmp57, undef
%tmp309 = fsub float %tmp58, undef
%tmp310 = fmul float %tmp307, undef
%tmp311 = fmul float %tmp308, 0.000000e+00
%tmp312 = fmul float %tmp309, undef
%tmp313 = fsub float %tmp59, undef
%tmp314 = fsub float %tmp60, undef
%tmp315 = fsub float %tmp61, undef
%tmp316 = fmul float %tmp313, undef
%tmp317 = fmul float %tmp314, undef
%tmp318 = fmul float %tmp315, undef
%tmp319 = call float @llvm.maxnum.f32(float %tmp310, float %tmp316)
%tmp320 = call float @llvm.maxnum.f32(float %tmp311, float %tmp317)
%tmp321 = call float @llvm.maxnum.f32(float %tmp312, float %tmp318)
%tmp322 = call float @llvm.minnum.f32(float %tmp319, float %tmp320)
%tmp323 = call float @llvm.minnum.f32(float %tmp322, float %tmp321)
%tmp324 = fcmp ogt float undef, 0.000000e+00
%tmp325 = fcmp olt float undef, 1.000000e+00
%tmp326 = and i1 %tmp324, %tmp325
%tmp327 = fcmp olt float undef, %tmp323
%tmp328 = and i1 %tmp326, %tmp327
br i1 %tmp328, label %ENDIF2795, label %ELSE2824
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
ELSE2824: ; preds = %ELSE2821
%.2849 = select i1 undef, float 0.000000e+00, float 1.000000e+00
br label %ENDIF2795
}
AMDGPU: Remove superfluous string attributes from tests Also fix v_mac.ll not testing right thing for fneg llvm-svn: 275129
2016-07-12 07:35:48 +08:00
declare float @llvm.floor.f32(float) #1
declare float @llvm.sqrt.f32(float) #1
declare float @llvm.minnum.f32(float, float) #1
declare float @llvm.maxnum.f32(float, float) #1
AMDGPU: Remove some uses of llvm.SI.export in tests Merge some of the old, smaller tests into more complete versions. llvm-svn: 295792
2017-02-22 08:02:21 +08:00
declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #1
declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
AMDGPU: Convert tests away from llvm.SI.load.const llvm-svn: 351494
2019-01-18 06:47:26 +08:00
declare float @llvm.amdgcn.s.buffer.load.f32(<4 x i32>, i32, i32) #1
R600/SI: add pass to mark CF live ranges as non-spillable Spilling can insert instructions almost anywhere, and this can mess up control flow lowering in a multitude of ways, due to instruction reordering. Let's sort this out the easy way: never spill registers involved with control flow, i.e. saved EXEC masks. Unfortunately, this does not work at all with optimizations disabled, as the register allocator ignores spill weights. This should be addressed in a future commit. The test was reduced from the "stacks" shader of [1]. Some issues trigger the machine verifier while another one is checked manually. [1] http://madebyevan.com/webgl-path-tracing/ v2: only insert pass with optimizations enabled, merge test runs. Patch by: Grigori Goronzy llvm-svn: 237152
2015-05-13 01:13:02 +08:00
AMDGPU: Remove superfluous string attributes from tests Also fix v_mac.ll not testing right thing for fneg llvm-svn: 275129
2016-07-12 07:35:48 +08:00
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }