[llvm] [LV] Add support for extended fadd reductions (PR #178447)

[Graham Hunter via llvm-commits]

================
@@ -750,6 +750,113 @@ for.exit:                        ; preds = %for.body
   ret float %add
 }
 
+define float @extended_reduce_fadd(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: define float @extended_reduce_fadd(
+; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 4
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]]
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP3:%.*]] = shl nuw i64 [[TMP2]], 3
+; CHECK-NEXT:    [[TMP4:%.*]] = shl nuw i64 [[TMP3]], 1
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP4]]
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x float> [ insertelement (<vscale x 4 x float> splat (float -0.000000e+00), float 0.000000e+00, i32 0), %[[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <vscale x 4 x float> [ insertelement (<vscale x 4 x float> splat (float -0.000000e+00), float -0.000000e+00, i32 0), %[[VECTOR_PH]] ], [ [[PARTIAL_REDUCE3:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr half, ptr [[A]], i64 [[INDEX]]
+; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr half, ptr [[TMP5]], i64 [[TMP3]]
+; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 8 x half>, ptr [[TMP5]], align 1
+; CHECK-NEXT:    [[WIDE_LOAD2:%.*]] = load <vscale x 8 x half>, ptr [[TMP6]], align 1
+; CHECK-NEXT:    [[TMP7:%.*]] = fpext <vscale x 8 x half> [[WIDE_LOAD]] to <vscale x 8 x float>
+; CHECK-NEXT:    [[PARTIAL_REDUCE]] = call reassoc contract <vscale x 4 x float> @llvm.vector.partial.reduce.fadd.nxv4f32.nxv8f32(<vscale x 4 x float> [[VEC_PHI]], <vscale x 8 x float> [[TMP7]])
+; CHECK-NEXT:    [[TMP8:%.*]] = fpext <vscale x 8 x half> [[WIDE_LOAD2]] to <vscale x 8 x float>
+; CHECK-NEXT:    [[PARTIAL_REDUCE3]] = call reassoc contract <vscale x 4 x float> @llvm.vector.partial.reduce.fadd.nxv4f32.nxv8f32(<vscale x 4 x float> [[VEC_PHI1]], <vscale x 8 x float> [[TMP8]])
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = fadd reassoc contract <vscale x 4 x float> [[PARTIAL_REDUCE3]], [[PARTIAL_REDUCE]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call reassoc contract float @llvm.vector.reduce.fadd.nxv4f32(float -0.000000e+00, <vscale x 4 x float> [[BIN_RDX]])
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], [[FOR_EXIT:label %.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %accum = phi float [ 0.0, %entry ], [ %add, %for.body ]
+  %gep.a = getelementptr half, ptr %a, i64 %iv
+  %load.a = load half, ptr %gep.a, align 1
+  %ext.a = fpext half %load.a to float
+  %add = fadd reassoc contract float %ext.a, %accum
+  %iv.next = add i64 %iv, 1
+  %exitcond.not = icmp eq i64 %iv.next, 1024
+  br i1 %exitcond.not, label %for.exit, label %for.body, !loop !0
----------------
huntergr-arm wrote:

Should be `!llvm.loop !0`  ;)

https://github.com/llvm/llvm-project/pull/178447