[AMDGPU] improve fragile test for divergent branches

[sameerds]

Summary

The affected LIT test intends to test the correct use of divergence
analysis to detect a divergent branch with a uniform predicate. The
passes involved are LLVM IR passes, but the test runs llc and tries to
match against generated ISA, which makes it hard to demonstrate that
the intended behavior was really tested. Replaced this with a test
that invokes opt on the required passes and then checks for the
appropriate changes in the LLVM IR.

[arsenm]

llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll

8

I think this should include ISA generated checks, and we should probably just generate checks for all control flow tests

[sameerds]

llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll

8

I am not sure I understand what you mean. Do you want ISA checks in addition to the checks that I am proposing? Also, what is the set of "all control flow tests"?

38

Note that this block does not even exist in the original test. It is generated internally by the structurizer. Any changes to the structurizer can potentially stop generating the branch that is being implied in the original test.

[arsenm]

llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll

8

Yes. I mean in general we should just generate checks for any control flow test, which this is one of

[sameerds]

llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll

8

Yeah, I see the point in general. But the problem with this test is that there is no line in the original input that explicitly captures what is being tested. (See my inline comment elsewhere about the block Flow2.) The new version is a dump of the LLVM IR after structurizer to actually show the branch being tested. Using this version to generate ISA would pass it through the structurizer yet again, which would need a very strong promise on the structurizer being idempotent. It's also possible that something else in the whole codegen flow might make the output insensitive to whether this specific divergent branch is correctly recognized.

Essentially, there are just too many things happening from input to output. As it stands, the new version of the test makes it a proper "unit test" that explicitly shows what is being tested. It's not the code generation that is being examined here, but the handling of divergent branches by specific LLVM IR passes before code generation.

[sameerds]

fixed the incorrect use of NOT and NEXT directives in FileCheck

llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll

-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck %s
+; RUN: opt --amdgpu-annotate-uniform -S %s |  FileCheck %s -check-prefix=UNIFORM
+; RUN: opt --amdgpu-annotate-uniform --si-annotate-control-flow -S %s |  FileCheck %s -check-prefix=CONTROLFLOW
 
-; This module creates a divergent branch. The branch is marked as divergent by
-; the divergence analysis but the condition is not. This test ensures that the
-; divergence of the branch is tested, not its condition, so that branch is
-; correctly emitted as divergent.
+; This module creates a divergent branch in block Flow2. The branch is
+; marked as divergent by the divergence analysis but the condition is
+; not. This test ensures that the divergence of the branch is tested,
+; not its condition, so that branch is correctly emitted as divergent.
 
 target triple = "amdgcn-mesa-mesa3d"
 
-define amdgpu_ps void @main(i32, float) {
-; CHECK-LABEL: main:
-; CHECK:       ; %bb.0: ; %start
-; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
-; CHECK-NEXT:    s_mov_b32 m0, s0
-; CHECK-NEXT:    s_mov_b32 s0, 0
-; CHECK-NEXT:    v_interp_p1_f32_e32 v0, v1, attr0.x
-; CHECK-NEXT:    v_cmp_nlt_f32_e32 vcc, 0, v0
-; CHECK-NEXT:    s_mov_b64 s[2:3], 0
-; CHECK-NEXT:    ; implicit-def: $sgpr6_sgpr7
-; CHECK-NEXT:    ; implicit-def: $sgpr4_sgpr5
-; CHECK-NEXT:    s_branch BB0_3
-; CHECK-NEXT:  BB0_1: ; %Flow1
-; CHECK-NEXT:    ; in Loop: Header=BB0_3 Depth=1
-; CHECK-NEXT:    s_or_b64 exec, exec, s[8:9]
-; CHECK-NEXT:    s_mov_b64 s[8:9], 0
-; CHECK-NEXT:  BB0_2: ; %Flow
-; CHECK-NEXT:    ; in Loop: Header=BB0_3 Depth=1
-; CHECK-NEXT:    s_and_b64 s[10:11], exec, s[6:7]
-; CHECK-NEXT:    s_or_b64 s[2:3], s[10:11], s[2:3]
-; CHECK-NEXT:    s_andn2_b64 s[4:5], s[4:5], exec
-; CHECK-NEXT:    s_and_b64 s[8:9], s[8:9], exec
-; CHECK-NEXT:    s_or_b64 s[4:5], s[4:5], s[8:9]
-; CHECK-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; CHECK-NEXT:    s_cbranch_execz BB0_6
-; CHECK-NEXT:  BB0_3: ; %loop
-; CHECK-NEXT:    ; =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    s_or_b64 s[6:7], s[6:7], exec
-; CHECK-NEXT:    s_cmp_lt_u32 s0, 32
-; CHECK-NEXT:    s_mov_b64 s[8:9], -1
-; CHECK-NEXT:    s_cbranch_scc0 BB0_2
-; CHECK-NEXT:  ; %bb.4: ; %endif1
-; CHECK-NEXT:    ; in Loop: Header=BB0_3 Depth=1
-; CHECK-NEXT:    s_mov_b64 s[6:7], -1
-; CHECK-NEXT:    s_and_saveexec_b64 s[8:9], vcc
-; CHECK-NEXT:    s_xor_b64 s[8:9], exec, s[8:9]
-; CHECK-NEXT:    s_cbranch_execz BB0_1
-; CHECK-NEXT:  ; %bb.5: ; %endif2
-; CHECK-NEXT:    ; in Loop: Header=BB0_3 Depth=1
-; CHECK-NEXT:    s_add_i32 s0, s0, 1
-; CHECK-NEXT:    s_xor_b64 s[6:7], exec, -1
-; CHECK-NEXT:    s_branch BB0_1
-; CHECK-NEXT:  BB0_6: ; %Flow2
-; CHECK-NEXT:    s_or_b64 exec, exec, s[2:3]
-; CHECK-NEXT:    v_mov_b32_e32 v1, 0
-; CHECK-NEXT:    s_and_saveexec_b64 s[0:1], s[4:5]
-; CHECK-NEXT:  ; %bb.7: ; %if1
-; CHECK-NEXT:    v_sqrt_f32_e32 v1, v0
-; CHECK-NEXT:  ; %bb.8: ; %endloop
-; CHECK-NEXT:    s_or_b64 exec, exec, s[0:1]
-; CHECK-NEXT:    exp mrt0 v1, v1, v1, v1 done vm
-; CHECK-NEXT:    s_endpgm
-
-; this is the divergent branch with the condition not marked as divergent
+define amdgpu_ps void @main(i32 %0, float %1) {
 start:
   %v0 = call float @llvm.amdgcn.interp.p1(float %1, i32 0, i32 0, i32 %0)
   br label %loop
 
-loop:
-  %v1 = phi i32 [ 0, %start ], [ %v5, %endif2 ]
+loop:                                             ; preds = %Flow, %start
+  %v1 = phi i32 [ 0, %start ], [ %6, %Flow ]
   %v2 = icmp ugt i32 %v1, 31
-  br i1 %v2, label %if1, label %endif1
+  %2 = xor i1 %v2, true
+  br i1 %2, label %endif1, label %Flow
+
+Flow1:                                            ; preds = %endif2, %endif1
+  %3 = phi i32 [ %v5, %endif2 ], [ undef, %endif1 ]
+  %4 = phi i1 [ false, %endif2 ], [ true, %endif1 ]
+  br label %Flow
+
+; UNIFORM-LABEL: Flow2:
+; UNIFORM-NEXT: br i1 %8, label %if1, label %endloop
+; UNIFORM-NOT: !amdgpu.uniform
+; UNIFORM: if1:
+
+; CONTROLFLOW-LABEL: Flow2:
+; CONTROLFLOW-NEXT:  call void @llvm.amdgcn.end.cf.i64(i64 %{{.*}})
+; CONTROLFLOW-NEXT:  [[IF:%.*]] = call { i1, i64 } @llvm.amdgcn.if.i64(i1 %{{.*}})
+; CONTROLFLOW-NEXT:  [[COND:%.*]] = extractvalue { i1, i64 } [[IF]], 0
+; CONTROLFLOW-NEXT:  %{{.*}} = extractvalue { i1, i64 } [[IF]], 1
+; CONTROLFLOW-NEXT:  br i1 [[COND]], label %if1, label %endloop
 
-if1:
+Flow2:                                            ; preds = %Flow
+  br i1 %8, label %if1, label %endloop
+
+if1:                                              ; preds = %Flow2
   %v3 = call float @llvm.sqrt.f32(float %v0)
   br label %endloop
 
-endif1:
+endif1:                                           ; preds = %loop
   %v4 = fcmp ogt float %v0, 0.000000e+00
-  br i1 %v4, label %endloop, label %endif2
+  %5 = xor i1 %v4, true
+  br i1 %5, label %endif2, label %Flow1
+
+Flow:                                             ; preds = %Flow1, %loop
+  %6 = phi i32 [ %3, %Flow1 ], [ undef, %loop ]
+  %7 = phi i1 [ %4, %Flow1 ], [ true, %loop ]
+  %8 = phi i1 [ false, %Flow1 ], [ true, %loop ]
+  br i1 %7, label %Flow2, label %loop
 
-endif2:
+endif2:                                           ; preds = %endif1
   %v5 = add i32 %v1, 1
-  br label %loop
+  br label %Flow1
 
-endloop:
-  %v6 = phi float [ %v3, %if1 ], [ 0.0, %endif1 ]
-  call void @llvm.amdgcn.exp.v4f32(i32 0, i32 15, float %v6, float %v6, float %v6, float %v6, i1 true, i1 true)
+endloop:                                          ; preds = %if1, %Flow2
+  %v6 = phi float [ 0.000000e+00, %Flow2 ], [ %v3, %if1 ]
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %v6, float %v6, float %v6, float %v6, i1 true, i1 true)
   ret void
 }
 
-declare float @llvm.sqrt.f32(float) #1
-declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32) #1
-declare void @llvm.amdgcn.exp.v4f32(i32, i32, float, float, float, float, i1, i1) #0
+; Function Attrs: nounwind readnone speculatable willreturn
+declare float @llvm.sqrt.f32(float) #0
+
+; Function Attrs: nounwind readnone speculatable
+declare float @llvm.amdgcn.interp.p1(float, i32 immarg, i32 immarg, i32) #1
+
+; Function Attrs: inaccessiblememonly nounwind writeonly
+declare void @llvm.amdgcn.exp.f32(i32 immarg, i32 immarg, float, float, float, float, i1 immarg, i1 immarg) #2
 
-attributes #0 = { nounwind }
-attributes #1 = { nounwind readnone }
+attributes #0 = { nounwind readnone speculatable willreturn }
+attributes #1 = { nounwind readnone speculatable }
+attributes #2 = { inaccessiblememonly nounwind writeonly }