1 ; RUN: llc -mtriple=x86_64-linux-gnu %s -o - | FileCheck %s
2 ; RUN: llc -mtriple=x86_64-linux-gnu %s -o - -O0 | FileCheck --check-prefix=NOOPT %s
6 define void @basic(i32 %x) {
8 switch i32 %x, label %return [
14 bb0: tail call void @g(i32 0) br label %return
15 bb1: tail call void @g(i32 1) br label %return
16 bb2: tail call void @g(i32 1) br label %return
19 ; Lowered as a jump table, both with and without optimization.
34 define void @simple_ranges(i32 %x) {
36 switch i32 %x, label %return [
46 bb0: tail call void @g(i32 0) br label %return
47 bb1: tail call void @g(i32 1) br label %return
50 ; Should be lowered to two range checks.
51 ; CHECK-LABEL: simple_ranges
58 ; We do this even at -O0, because it's cheap and makes codegen faster.
59 ; NOOPT-LABEL: simple_ranges
68 define void @jt_is_better(i32 %x) {
70 switch i32 %x, label %return [
82 bb0: tail call void @g(i32 0) br label %return
83 bb1: tail call void @g(i32 1) br label %return
84 bb2: tail call void @g(i32 2) br label %return
85 bb3: tail call void @g(i32 3) br label %return
86 bb4: tail call void @g(i32 4) br label %return
89 ; Cases 0-5 could be lowered with two bit tests,
90 ; but with 6-8, the whole switch is suitable for a jump table.
91 ; CHECK-LABEL: jt_is_better
98 define void @bt_is_better(i32 %x) {
100 switch i32 %x, label %return [
112 bb0: tail call void @g(i32 0) br label %return
113 bb1: tail call void @g(i32 1) br label %return
114 bb2: tail call void @g(i32 2) br label %return
117 ; This could be lowered as a jump table, but bit tests is more efficient.
118 ; CHECK-LABEL: bt_is_better
119 ; 73 = 2^0 + 2^3 + 2^6
122 ; 146 = 2^1 + 2^4 + 2^7
125 ; 292 = 2^2 + 2^5 + 2^8
131 define void @optimal_pivot1(i32 %x) {
133 switch i32 %x, label %return [
142 bb0: tail call void @g(i32 0) br label %return
143 bb1: tail call void @g(i32 1) br label %return
146 ; Should pivot around 400 for two subtrees of equal size.
147 ; CHECK-LABEL: optimal_pivot1
153 define void @optimal_pivot2(i32 %x) {
155 switch i32 %x, label %return [
156 i32 100, label %bb0 i32 101, label %bb1 i32 102, label %bb2 i32 103, label %bb3
157 i32 200, label %bb0 i32 201, label %bb1 i32 202, label %bb2 i32 203, label %bb3
158 i32 300, label %bb0 i32 301, label %bb1 i32 302, label %bb2 i32 303, label %bb3
159 i32 400, label %bb0 i32 401, label %bb1 i32 402, label %bb2 i32 403, label %bb3
162 bb0: tail call void @g(i32 0) br label %return
163 bb1: tail call void @g(i32 1) br label %return
164 bb2: tail call void @g(i32 2) br label %return
165 bb3: tail call void @g(i32 3) br label %return
168 ; Should pivot around 300 for two subtrees with two jump tables each.
169 ; CHECK-LABEL: optimal_pivot2
179 define void @optimal_jump_table1(i32 %x) {
181 switch i32 %x, label %return [
189 bb0: tail call void @g(i32 0) br label %return
190 bb1: tail call void @g(i32 1) br label %return
191 bb2: tail call void @g(i32 2) br label %return
192 bb3: tail call void @g(i32 3) br label %return
193 bb4: tail call void @g(i32 4) br label %return
194 bb5: tail call void @g(i32 5) br label %return
197 ; Splitting in the largest gap (between 6 and 12) would yield suboptimal result.
198 ; Expecting a jump table from 5 to 15.
199 ; CHECK-LABEL: optimal_jump_table1
204 ; At -O0, we don't build jump tables for only parts of a switch.
205 ; NOOPT-LABEL: optimal_jump_table1
206 ; NOOPT: testl %edi, %edi
208 ; NOOPT: subl $5, %eax
210 ; NOOPT: subl $6, %eax
212 ; NOOPT: subl $12, %eax
214 ; NOOPT: subl $13, %eax
216 ; NOOPT: subl $15, %eax
221 define void @optimal_jump_table2(i32 %x) {
223 switch i32 %x, label %return [
231 bb0: tail call void @g(i32 0) br label %return
232 bb1: tail call void @g(i32 1) br label %return
233 bb2: tail call void @g(i32 2) br label %return
234 bb3: tail call void @g(i32 3) br label %return
235 bb4: tail call void @g(i32 4) br label %return
236 bb5: tail call void @g(i32 5) br label %return
239 ; Partitioning the cases to the minimum number of dense sets is not good enough.
240 ; This can be partitioned as {0,1,2,9},{14,15} or {0,1,2},{9,14,15}. The former
241 ; should be preferred. Expecting a table from 0-9.
242 ; CHECK-LABEL: optimal_jump_table2
248 define void @optimal_jump_table3(i32 %x) {
250 switch i32 %x, label %return [
261 bb0: tail call void @g(i32 0) br label %return
262 bb1: tail call void @g(i32 1) br label %return
263 bb2: tail call void @g(i32 2) br label %return
264 bb3: tail call void @g(i32 3) br label %return
265 bb4: tail call void @g(i32 4) br label %return
268 ; Splitting to maximize left-right density sum and gap size would split this
269 ; between 3 and 10, and then between 20 and 25. It's better to build a table
271 ; CHECK-LABEL: optimal_jump_table3
277 %struct.S = type { %struct.S*, i32 }
278 define void @phi_node_trouble(%struct.S* %s) {
282 %ptr = phi %struct.S* [ %s, %entry ], [ %next, %loop ]
283 %bool = icmp eq %struct.S* %ptr, null
284 br i1 %bool, label %exit, label %loop
286 %nextptr = getelementptr inbounds %struct.S, %struct.S* %ptr, i64 0, i32 0
287 %next = load %struct.S*, %struct.S** %nextptr
288 %xptr = getelementptr inbounds %struct.S, %struct.S* %next, i64 0, i32 1
289 %x = load i32, i32* %xptr
290 switch i32 %x, label %exit [
301 ; This will be lowered to a comparison with 4 and then bit tests. Make sure
302 ; that the phi node in %header gets a value from the comparison block.
303 ; CHECK-LABEL: phi_node_trouble
304 ; CHECK: movq (%[[REG1:[a-z]+]]), %[[REG1]]
305 ; CHECK: movl 8(%[[REG1]]), %[[REG2:[a-z]+]]
306 ; CHECK: cmpl $4, %[[REG2]]
310 define void @default_only(i32 %x) {
316 switch i32 %x, label %return [
319 ; Branch directly to the default.
320 ; (In optimized builds the switch is removed earlier.)
321 ; NOOPT-LABEL: default_only
322 ; NOOPT: .[[L:[A-Z0-9_]+]]:
328 define void @int_max_table_cluster(i8 %x) {
330 switch i8 %x, label %return [
331 i8 0, label %bb0 i8 1, label %bb0 i8 2, label %bb0 i8 3, label %bb0
332 i8 4, label %bb0 i8 5, label %bb0 i8 6, label %bb0 i8 7, label %bb0
333 i8 8, label %bb0 i8 9, label %bb0 i8 10, label %bb0 i8 11, label %bb0
334 i8 12, label %bb0 i8 13, label %bb0 i8 14, label %bb0 i8 15, label %bb0
335 i8 16, label %bb0 i8 17, label %bb0 i8 18, label %bb0 i8 19, label %bb0
336 i8 20, label %bb0 i8 21, label %bb0 i8 22, label %bb0 i8 23, label %bb0
337 i8 24, label %bb0 i8 25, label %bb0 i8 26, label %bb0 i8 27, label %bb0
338 i8 28, label %bb0 i8 29, label %bb0 i8 30, label %bb0 i8 31, label %bb0
339 i8 32, label %bb0 i8 33, label %bb0 i8 34, label %bb0 i8 35, label %bb0
340 i8 36, label %bb0 i8 37, label %bb0 i8 38, label %bb0 i8 39, label %bb0
341 i8 40, label %bb0 i8 41, label %bb0 i8 42, label %bb0 i8 43, label %bb0
342 i8 44, label %bb0 i8 45, label %bb0 i8 46, label %bb0 i8 47, label %bb0
343 i8 48, label %bb0 i8 49, label %bb0 i8 50, label %bb0 i8 51, label %bb0
344 i8 52, label %bb0 i8 53, label %bb0 i8 54, label %bb0 i8 55, label %bb0
345 i8 56, label %bb0 i8 57, label %bb0 i8 58, label %bb0 i8 59, label %bb0
346 i8 60, label %bb0 i8 61, label %bb0 i8 62, label %bb0 i8 63, label %bb0
347 i8 64, label %bb0 i8 65, label %bb0 i8 66, label %bb0 i8 67, label %bb0
348 i8 68, label %bb0 i8 69, label %bb0 i8 70, label %bb0 i8 71, label %bb0
349 i8 72, label %bb0 i8 73, label %bb0 i8 74, label %bb0 i8 75, label %bb0
350 i8 76, label %bb0 i8 77, label %bb0 i8 78, label %bb0 i8 79, label %bb0
351 i8 80, label %bb0 i8 81, label %bb0 i8 82, label %bb0 i8 83, label %bb0
352 i8 84, label %bb0 i8 85, label %bb0 i8 86, label %bb0 i8 87, label %bb0
353 i8 88, label %bb0 i8 89, label %bb0 i8 90, label %bb0 i8 91, label %bb0
354 i8 92, label %bb0 i8 93, label %bb0 i8 94, label %bb0 i8 95, label %bb0
355 i8 96, label %bb0 i8 97, label %bb0 i8 98, label %bb0 i8 99, label %bb0
356 i8 100, label %bb0 i8 101, label %bb0 i8 102, label %bb0 i8 103, label %bb0
357 i8 104, label %bb0 i8 105, label %bb0 i8 106, label %bb0 i8 107, label %bb0
358 i8 108, label %bb0 i8 109, label %bb0 i8 110, label %bb0 i8 111, label %bb0
359 i8 112, label %bb0 i8 113, label %bb0 i8 114, label %bb0 i8 115, label %bb0
360 i8 116, label %bb0 i8 117, label %bb0 i8 118, label %bb0 i8 119, label %bb0
361 i8 120, label %bb0 i8 121, label %bb0 i8 122, label %bb0 i8 123, label %bb0
362 i8 124, label %bb0 i8 125, label %bb0 i8 126, label %bb0 i8 127, label %bb0
363 i8 -64, label %bb1 i8 -63, label %bb1 i8 -62, label %bb1 i8 -61, label %bb1
364 i8 -60, label %bb1 i8 -59, label %bb1 i8 -58, label %bb1 i8 -57, label %bb1
365 i8 -56, label %bb1 i8 -55, label %bb1 i8 -54, label %bb1 i8 -53, label %bb1
366 i8 -52, label %bb1 i8 -51, label %bb1 i8 -50, label %bb1 i8 -49, label %bb1
367 i8 -48, label %bb1 i8 -47, label %bb1 i8 -46, label %bb1 i8 -45, label %bb1
368 i8 -44, label %bb1 i8 -43, label %bb1 i8 -42, label %bb1 i8 -41, label %bb1
369 i8 -40, label %bb1 i8 -39, label %bb1 i8 -38, label %bb1 i8 -37, label %bb1
370 i8 -36, label %bb1 i8 -35, label %bb1 i8 -34, label %bb1 i8 -33, label %bb1
371 i8 -32, label %bb2 i8 -31, label %bb2 i8 -30, label %bb2 i8 -29, label %bb2
372 i8 -28, label %bb2 i8 -27, label %bb2 i8 -26, label %bb2 i8 -25, label %bb2
373 i8 -24, label %bb2 i8 -23, label %bb2 i8 -22, label %bb2 i8 -21, label %bb2
374 i8 -20, label %bb2 i8 -19, label %bb2 i8 -18, label %bb2 i8 -17, label %bb2
375 i8 -16, label %bb3 i8 -15, label %bb3 i8 -14, label %bb3 i8 -13, label %bb3
376 i8 -12, label %bb3 i8 -11, label %bb3 i8 -10, label %bb3 i8 -9, label %bb3
378 bb0: tail call void @g(i32 0) br label %return
379 bb1: tail call void @g(i32 1) br label %return
380 bb2: tail call void @g(i32 1) br label %return
381 bb3: tail call void @g(i32 1) br label %return
384 ; Don't infloop on jump tables where the upper bound is the max value of the
385 ; input type (in this case 127).
386 ; CHECK-LABEL: int_max_table_cluster
391 define void @bt_order_by_weight(i32 %x) {
393 switch i32 %x, label %return [
405 bb0: tail call void @g(i32 0) br label %return
406 bb1: tail call void @g(i32 1) br label %return
407 bb2: tail call void @g(i32 2) br label %return
410 ; Cases 1,4,7 have a very large branch weight (which shouldn't overflow), so
411 ; their bit test should come first. 0,3,6 and 2,5,8,9 both have a weight of 12,
412 ; but the latter set has more cases, so should be tested for earlier.
414 ; CHECK-LABEL: bt_order_by_weight
415 ; 146 = 2^1 + 2^4 + 2^7
418 ; 292 = 2^2 + 2^5 + 2^8 + 2^9
421 ; 73 = 2^0 + 2^3 + 2^6
426 !1 = !{!"branch_weights",
432 i32 4294967295, i32 2, i32 4294967295,
434 i32 3, i32 3, i32 3, i32 3}
436 define void @order_by_weight_and_fallthrough(i32 %x) {
438 switch i32 %x, label %return [
443 bb0: tail call void @g(i32 0) br label %return
444 bb1: tail call void @g(i32 1) br label %return
447 ; Case 200 has the highest weight and should come first. 100 and 300 have the
448 ; same weight, but 300 goes to the 'next' block, so should be last.
449 ; CHECK-LABEL: order_by_weight_and_fallthrough
455 !2 = !{!"branch_weights",
466 define void @zero_weight_tree(i32 %x) {
468 switch i32 %x, label %return [
476 bb0: tail call void @g(i32 0) br label %return
477 bb1: tail call void @g(i32 1) br label %return
478 bb2: tail call void @g(i32 2) br label %return
479 bb3: tail call void @g(i32 3) br label %return
480 bb4: tail call void @g(i32 4) br label %return
481 bb5: tail call void @g(i32 5) br label %return
484 ; Make sure to pick a pivot in the middle also with zero-weight cases.
485 ; CHECK-LABEL: zero_weight_tree
490 !3 = !{!"branch_weights", i32 1, i32 10, i32 0, i32 0, i32 0, i32 0, i32 10}
493 define void @left_leaning_weight_balanced_tree(i32 %x) {
495 switch i32 %x, label %return [
503 bb0: tail call void @g(i32 0) br label %return
504 bb1: tail call void @g(i32 1) br label %return
505 bb2: tail call void @g(i32 2) br label %return
506 bb3: tail call void @g(i32 3) br label %return
507 bb4: tail call void @g(i32 4) br label %return
508 bb5: tail call void @g(i32 5) br label %return
511 ; To balance the tree by weight, the pivot is shifted to the right, moving hot
512 ; cases closer to the root.
513 ; CHECK-LABEL: left_leaning_weight_balanced_tree
518 !4 = !{!"branch_weights", i32 1, i32 10, i32 1, i32 1, i32 1, i32 10, i32 10}
521 define void @jump_table_affects_balance(i32 %x) {
523 switch i32 %x, label %return [
534 bb0: tail call void @g(i32 0) br label %return
535 bb1: tail call void @g(i32 1) br label %return
536 bb2: tail call void @g(i32 2) br label %return
537 bb3: tail call void @g(i32 3) br label %return
540 ; CHECK-LABEL: jump_table_affects_balance
541 ; If the tree were balanced based on number of clusters, {0-3,100} would go on
542 ; the left and {200,300} on the right. However, the jump table weights as much
543 ; as its components, so 100 is selected as the pivot.
549 define void @pr23738(i4 %x) {
551 switch i4 %x, label %bb0 [
556 bb0: tail call void @g(i32 0) br label %return
557 bb1: tail call void @g(i32 1) br label %return
559 ; Don't assert due to truncating the bitwidth (64) to i4 when checking
560 ; that the bit-test range fits in a word.