[llvm] [LoopUnroll] Introduce parallel accumulators when unrolling FP reductions. (PR #166630)

[Florian Hahn via llvm-commits]

================
@@ -220,6 +220,137 @@ exit:
   ret i32 %res
 }
 
+define float @test_fadd_reduction(ptr %a, i64 %n) {
+; CHECK-LABEL: define float @test_fadd_reduction(
+; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
+; CHECK-NEXT:    [[XTRAITER:%.*]] = and i64 [[N]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 1
+; CHECK-NEXT:    br i1 [[TMP1]], label %[[LOOP_EPIL_PREHEADER:.*]], label %[[ENTRY_NEW:.*]]
+; CHECK:       [[ENTRY_NEW]]:
+; CHECK-NEXT:    [[UNROLL_ITER:%.*]] = sub i64 [[N]], [[XTRAITER]]
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, %[[ENTRY_NEW]] ], [ [[IV_NEXT_1:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[RDX_1:%.*]] = phi float [ -0.000000e+00, %[[ENTRY_NEW]] ], [ [[RDX_NEXT_1:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[RDX:%.*]] = phi float [ 0.000000e+00, %[[ENTRY_NEW]] ], [ [[RDX_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, %[[ENTRY_NEW]] ], [ [[NITER_NEXT_1:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[GEP_A:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[IV]]
+; CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr [[GEP_A]], align 16
+; CHECK-NEXT:    [[RDX_NEXT]] = fadd reassoc float [[RDX]], [[TMP2]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT:    [[GEP_A_1:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[IV_NEXT]]
+; CHECK-NEXT:    [[TMP3:%.*]] = load float, ptr [[GEP_A_1]], align 16
+; CHECK-NEXT:    [[RDX_NEXT_1]] = fadd reassoc float [[RDX_1]], [[TMP3]]
+; CHECK-NEXT:    [[IV_NEXT_1]] = add nuw nsw i64 [[IV]], 2
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER]], 2
+; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NEXT_1]], [[UNROLL_ITER]]
+; CHECK-NEXT:    br i1 [[NITER_NCMP_1]], label %[[EXIT_UNR_LCSSA:.*]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK:       [[EXIT_UNR_LCSSA]]:
+; CHECK-NEXT:    [[RES_PH:%.*]] = phi float [ [[RDX_NEXT_1]], %[[LOOP]] ]
+; CHECK-NEXT:    [[IV_UNR:%.*]] = phi i64 [ [[IV_NEXT_1]], %[[LOOP]] ]
+; CHECK-NEXT:    [[RDX_UNR:%.*]] = phi float [ [[RDX_NEXT_1]], %[[LOOP]] ]
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = fadd reassoc float [[RDX_NEXT_1]], [[RDX_NEXT]]
+; CHECK-NEXT:    [[LCMP_MOD:%.*]] = icmp ne i64 [[XTRAITER]], 0
+; CHECK-NEXT:    br i1 [[LCMP_MOD]], label %[[LOOP_EPIL_PREHEADER]], label %[[EXIT:.*]]
+; CHECK:       [[LOOP_EPIL_PREHEADER]]:
+; CHECK-NEXT:    [[IV_EPIL_INIT:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_UNR]], %[[EXIT_UNR_LCSSA]] ]
+; CHECK-NEXT:    [[RDX_EPIL_INIT:%.*]] = phi float [ 0.000000e+00, %[[ENTRY]] ], [ [[BIN_RDX]], %[[EXIT_UNR_LCSSA]] ]
+; CHECK-NEXT:    [[LCMP_MOD2:%.*]] = icmp ne i64 [[XTRAITER]], 0
+; CHECK-NEXT:    call void @llvm.assume(i1 [[LCMP_MOD2]])
+; CHECK-NEXT:    br label %[[LOOP_EPIL:.*]]
+; CHECK:       [[LOOP_EPIL]]:
+; CHECK-NEXT:    [[GEP_A_EPIL:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[IV_EPIL_INIT]]
+; CHECK-NEXT:    [[TMP4:%.*]] = load float, ptr [[GEP_A_EPIL]], align 16
+; CHECK-NEXT:    [[RDX_NEXT_EPIL:%.*]] = fadd reassoc float [[RDX_EPIL_INIT]], [[TMP4]]
+; CHECK-NEXT:    br label %[[EXIT]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    [[RES:%.*]] = phi float [ [[BIN_RDX]], %[[EXIT_UNR_LCSSA]] ], [ [[RDX_NEXT_EPIL]], %[[LOOP_EPIL]] ]
+; CHECK-NEXT:    ret float [[RES]]
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+  %rdx = phi float [ 0.0, %entry ], [ %rdx.next, %loop ]
+  %gep.a = getelementptr inbounds nuw float, ptr %a, i64 %iv
+  %1 = load float, ptr %gep.a, align 16
+  %rdx.next = fadd reassoc float %rdx, %1
----------------
fhahn wrote:

Could you also add a test with a different fast-math flag, to make sure we don't introduce additional accumulators in that case?

https://github.com/llvm/llvm-project/pull/166630