1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -reassociate -S | FileCheck %s
3
4; Check that a*c+b*c is turned into (a+b)*c
5
6define <4 x float> @test1(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
7; CHECK-LABEL: @test1(
8; CHECK-NEXT:    [[REASS_ADD:%.*]] = fadd fast <4 x float> [[B:%.*]], [[A:%.*]]
9; CHECK-NEXT:    [[REASS_MUL:%.*]] = fmul fast <4 x float> [[REASS_ADD]], [[C:%.*]]
10; CHECK-NEXT:    ret <4 x float> [[REASS_MUL]]
11;
12  %mul = fmul fast <4 x float> %a, %c
13  %mul1 = fmul fast <4 x float> %b, %c
14  %add = fadd fast <4 x float> %mul, %mul1
15  ret <4 x float> %add
16}
17
18; Check that a*c+b*c is turned into (a+b)*c - minimum FMF subset version
19
20define <4 x float> @test1_reassoc(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
21; CHECK-LABEL: @test1_reassoc(
22; CHECK-NEXT:    [[MUL:%.*]] = fmul reassoc <4 x float> [[A:%.*]], [[C:%.*]]
23; CHECK-NEXT:    [[MUL1:%.*]] = fmul reassoc <4 x float> [[B:%.*]], [[C]]
24; CHECK-NEXT:    [[ADD:%.*]] = fadd reassoc <4 x float> [[MUL]], [[MUL1]]
25; CHECK-NEXT:    ret <4 x float> [[ADD]]
26;
27  %mul = fmul reassoc <4 x float> %a, %c
28  %mul1 = fmul reassoc <4 x float> %b, %c
29  %add = fadd reassoc <4 x float> %mul, %mul1
30  ret <4 x float> %add
31}
32
33; Check that a*a*b+a*a*c is turned into a*(a*(b+c)).
34
35define <2 x float> @test2(<2 x float> %a, <2 x float> %b, <2 x float> %c) {
36; CHECK-LABEL: @test2(
37; CHECK-NEXT:    [[REASS_ADD1:%.*]] = fadd fast <2 x float> [[C:%.*]], [[B:%.*]]
38; CHECK-NEXT:    [[REASS_MUL2:%.*]] = fmul fast <2 x float> [[A:%.*]], [[A]]
39; CHECK-NEXT:    [[REASS_MUL:%.*]] = fmul fast <2 x float> [[REASS_MUL2]], [[REASS_ADD1]]
40; CHECK-NEXT:    ret <2 x float> [[REASS_MUL]]
41;
42  %t0 = fmul fast <2 x float> %a, %b
43  %t1 = fmul fast <2 x float> %a, %t0
44  %t2 = fmul fast <2 x float> %a, %c
45  %t3 = fmul fast <2 x float> %a, %t2
46  %t4 = fadd fast <2 x float> %t1, %t3
47  ret <2 x float> %t4
48}
49
50; Check that a*a*b+a*a*c is turned into a*(a*(b+c)) - minimum FMF subset version
51
52define <2 x float> @test2_reassoc(<2 x float> %a, <2 x float> %b, <2 x float> %c) {
53; CHECK-LABEL: @test2_reassoc(
54; CHECK-NEXT:    [[T0:%.*]] = fmul reassoc <2 x float> [[A:%.*]], [[B:%.*]]
55; CHECK-NEXT:    [[T1:%.*]] = fmul reassoc <2 x float> [[A]], [[T0]]
56; CHECK-NEXT:    [[T2:%.*]] = fmul reassoc <2 x float> [[A]], [[C:%.*]]
57; CHECK-NEXT:    [[T3:%.*]] = fmul reassoc <2 x float> [[A]], [[T2]]
58; CHECK-NEXT:    [[T4:%.*]] = fadd reassoc <2 x float> [[T1]], [[T3]]
59; CHECK-NEXT:    ret <2 x float> [[T4]]
60;
61  %t0 = fmul reassoc <2 x float> %a, %b
62  %t1 = fmul reassoc <2 x float> %a, %t0
63  %t2 = fmul reassoc <2 x float> %a, %c
64  %t3 = fmul reassoc <2 x float> %a, %t2
65  %t4 = fadd reassoc <2 x float> %t1, %t3
66  ret <2 x float> %t4
67}
68
69; Check that a*b+a*c+d is turned into a*(b+c)+d.
70
71define <2 x double> @test3(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x double> %d) {
72; CHECK-LABEL: @test3(
73; CHECK-NEXT:    [[REASS_ADD:%.*]] = fadd fast <2 x double> [[C:%.*]], [[B:%.*]]
74; CHECK-NEXT:    [[REASS_MUL:%.*]] = fmul fast <2 x double> [[REASS_ADD]], [[A:%.*]]
75; CHECK-NEXT:    [[T3:%.*]] = fadd fast <2 x double> [[REASS_MUL]], [[D:%.*]]
76; CHECK-NEXT:    ret <2 x double> [[T3]]
77;
78  %t0 = fmul fast <2 x double> %a, %b
79  %t1 = fmul fast <2 x double> %a, %c
80  %t2 = fadd fast <2 x double> %t1, %d
81  %t3 = fadd fast <2 x double> %t0, %t2
82  ret <2 x double> %t3
83}
84
85; Check that a*b+a*c+d is turned into a*(b+c)+d - minimum FMF subset version
86
87define <2 x double> @test3_reassoc(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x double> %d) {
88; CHECK-LABEL: @test3_reassoc(
89; CHECK-NEXT:    [[T0:%.*]] = fmul reassoc <2 x double> [[A:%.*]], [[B:%.*]]
90; CHECK-NEXT:    [[T1:%.*]] = fmul reassoc <2 x double> [[A]], [[C:%.*]]
91; CHECK-NEXT:    [[T2:%.*]] = fadd reassoc <2 x double> [[T1]], [[D:%.*]]
92; CHECK-NEXT:    [[T3:%.*]] = fadd reassoc <2 x double> [[T0]], [[T2]]
93; CHECK-NEXT:    ret <2 x double> [[T3]]
94;
95  %t0 = fmul reassoc <2 x double> %a, %b
96  %t1 = fmul reassoc <2 x double> %a, %c
97  %t2 = fadd reassoc <2 x double> %t1, %d
98  %t3 = fadd reassoc <2 x double> %t0, %t2
99  ret <2 x double> %t3
100}
101
102; No fast-math.
103
104define <2 x float> @test4(<2 x float> %A) {
105; CHECK-LABEL: @test4(
106; CHECK-NEXT:    [[X:%.*]] = fadd <2 x float> [[A:%.*]], <float 1.000000e+00, float 1.000000e+00>
107; CHECK-NEXT:    [[Y:%.*]] = fadd <2 x float> [[A]], <float 1.000000e+00, float 1.000000e+00>
108; CHECK-NEXT:    [[R:%.*]] = fsub <2 x float> [[X]], [[Y]]
109; CHECK-NEXT:    ret <2 x float> [[R]]
110;
111  %X = fadd <2 x float> %A, < float 1.000000e+00, float 1.000000e+00 >
112  %Y = fadd <2 x float> %A, < float 1.000000e+00, float 1.000000e+00 >
113  %R = fsub <2 x float> %X, %Y
114  ret <2 x float> %R
115}
116
117; Check 47*X + 47*X -> 94*X.
118
119define <2 x float> @test5(<2 x float> %X) {
120; CHECK-LABEL: @test5(
121; CHECK-NEXT:    [[FACTOR:%.*]] = fmul fast <2 x float> [[X:%.*]], <float 9.400000e+01, float 9.400000e+01>
122; CHECK-NEXT:    ret <2 x float> [[FACTOR]]
123;
124  %Y = fmul fast <2 x float> %X, <float 4.700000e+01, float 4.700000e+01>
125  %Z = fadd fast <2 x float> %Y, %Y
126  ret <2 x float> %Z
127}
128
129; Check 47*X + 47*X -> 94*X - minimum FMF subset version
130
131define <2 x float> @test5_reassoc(<2 x float> %X) {
132; CHECK-LABEL: @test5_reassoc(
133; CHECK-NEXT:    [[Y:%.*]] = fmul reassoc <2 x float> [[X:%.*]], <float 4.700000e+01, float 4.700000e+01>
134; CHECK-NEXT:    [[Z:%.*]] = fadd reassoc <2 x float> [[Y]], [[Y]]
135; CHECK-NEXT:    ret <2 x float> [[Z]]
136;
137  %Y = fmul reassoc <2 x float> %X, <float 4.700000e+01, float 4.700000e+01>
138  %Z = fadd reassoc <2 x float> %Y, %Y
139  ret <2 x float> %Z
140}
141
142; Check X+X+X -> 3*X.
143
144define <2 x float> @test6(<2 x float> %X) {
145; CHECK-LABEL: @test6(
146; CHECK-NEXT:    [[FACTOR:%.*]] = fmul fast <2 x float> [[X:%.*]], <float 3.000000e+00, float 3.000000e+00>
147; CHECK-NEXT:    ret <2 x float> [[FACTOR]]
148;
149  %Y = fadd fast <2 x float> %X ,%X
150  %Z = fadd fast <2 x float> %Y, %X
151  ret <2 x float> %Z
152}
153
154; Check X+X+X -> 3*X - minimum FMF subset version
155
156define <2 x float> @test6_reassoc(<2 x float> %X) {
157; CHECK-LABEL: @test6_reassoc(
158; CHECK-NEXT:    [[Y:%.*]] = fadd reassoc <2 x float> [[X:%.*]], [[X]]
159; CHECK-NEXT:    [[Z:%.*]] = fadd reassoc <2 x float> [[X]], [[Y]]
160; CHECK-NEXT:    ret <2 x float> [[Z]]
161;
162  %Y = fadd reassoc <2 x float> %X ,%X
163  %Z = fadd reassoc <2 x float> %Y, %X
164  ret <2 x float> %Z
165}
166
167; Check 127*W+50*W -> 177*W.
168
169define <2 x double> @test7(<2 x double> %W) {
170; CHECK-LABEL: @test7(
171; CHECK-NEXT:    [[REASS_MUL:%.*]] = fmul fast <2 x double> [[W:%.*]], <double 1.770000e+02, double 1.770000e+02>
172; CHECK-NEXT:    ret <2 x double> [[REASS_MUL]]
173;
174  %X = fmul fast <2 x double> %W, <double 127.0, double 127.0>
175  %Y = fmul fast <2 x double> %W, <double 50.0, double 50.0>
176  %Z = fadd fast <2 x double> %Y, %X
177  ret <2 x double> %Z
178}
179
180; Check 127*W+50*W -> 177*W - minimum FMF subset version
181
182define <2 x double> @test7_reassoc(<2 x double> %W) {
183; CHECK-LABEL: @test7_reassoc(
184; CHECK-NEXT:    [[X:%.*]] = fmul reassoc <2 x double> [[W:%.*]], <double 1.270000e+02, double 1.270000e+02>
185; CHECK-NEXT:    [[Y:%.*]] = fmul reassoc <2 x double> [[W]], <double 5.000000e+01, double 5.000000e+01>
186; CHECK-NEXT:    [[Z:%.*]] = fadd reassoc <2 x double> [[Y]], [[X]]
187; CHECK-NEXT:    ret <2 x double> [[Z]]
188;
189  %X = fmul reassoc <2 x double> %W, <double 127.0, double 127.0>
190  %Y = fmul reassoc <2 x double> %W, <double 50.0, double 50.0>
191  %Z = fadd reassoc <2 x double> %Y, %X
192  ret <2 x double> %Z
193}
194
195; Check X*12*12 -> X*144.
196
197define <2 x float> @test8(<2 x float> %arg) {
198; CHECK-LABEL: @test8(
199; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast <2 x float> [[ARG:%.*]], <float 1.440000e+02, float 1.440000e+02>
200; CHECK-NEXT:    ret <2 x float> [[TMP2]]
201;
202  %tmp1 = fmul fast <2 x float> <float 1.200000e+01, float 1.200000e+01>, %arg
203  %tmp2 = fmul fast <2 x float> %tmp1, <float 1.200000e+01, float 1.200000e+01>
204  ret <2 x float> %tmp2
205}
206
207; Check X*12*12 -> X*144 - minimum FMF subset version
208
209define <2 x float> @test8_reassoc(<2 x float> %arg) {
210; CHECK-LABEL: @test8_reassoc(
211; CHECK-NEXT:    [[TMP1:%.*]] = fmul reassoc <2 x float> [[ARG:%.*]], <float 1.200000e+01, float 1.200000e+01>
212; CHECK-NEXT:    [[TMP2:%.*]] = fmul reassoc <2 x float> [[TMP1]], <float 1.200000e+01, float 1.200000e+01>
213; CHECK-NEXT:    ret <2 x float> [[TMP2]]
214;
215  %tmp1 = fmul reassoc <2 x float> <float 1.200000e+01, float 1.200000e+01>, %arg
216  %tmp2 = fmul reassoc <2 x float> %tmp1, <float 1.200000e+01, float 1.200000e+01>
217  ret <2 x float> %tmp2
218}
219
220; Check (b+(a+1234))+-a -> b+1234.
221
222define <2 x double> @test9(<2 x double> %b, <2 x double> %a) {
223; CHECK-LABEL: @test9(
224; CHECK-NEXT:    [[TMP1:%.*]] = fsub fast <2 x double> zeroinitializer, [[A:%.*]]
225; CHECK-NEXT:    [[TMP2:%.*]] = fadd fast <2 x double> [[B:%.*]], <double 1.234000e+03, double 1.234000e+03>
226; CHECK-NEXT:    ret <2 x double> [[TMP2]]
227;
228  %1 = fadd fast <2 x double> %a, <double 1.234000e+03, double 1.234000e+03>
229  %2 = fadd fast <2 x double> %b, %1
230  %3 = fsub fast <2 x double> <double 0.000000e+00, double 0.000000e+00>, %a
231  %4 = fadd fast <2 x double> %2, %3
232  ret <2 x double> %4
233}
234
235; Check (b+(a+1234))+-a -> b+1234 - minimum FMF subset version
236
237define <2 x double> @test9_reassoc(<2 x double> %b, <2 x double> %a) {
238; CHECK-LABEL: @test9_reassoc(
239; CHECK-NEXT:    [[TMP1:%.*]] = fadd reassoc <2 x double> [[A:%.*]], <double 1.234000e+03, double 1.234000e+03>
240; CHECK-NEXT:    [[TMP2:%.*]] = fadd reassoc <2 x double> [[B:%.*]], [[TMP1]]
241; CHECK-NEXT:    [[TMP3:%.*]] = fsub reassoc <2 x double> zeroinitializer, [[A]]
242; CHECK-NEXT:    [[TMP4:%.*]] = fadd reassoc <2 x double> [[TMP3]], [[TMP2]]
243; CHECK-NEXT:    ret <2 x double> [[TMP4]]
244;
245  %1 = fadd reassoc <2 x double> %a, <double 1.234000e+03, double 1.234000e+03>
246  %2 = fadd reassoc <2 x double> %b, %1
247  %3 = fsub reassoc <2 x double> <double 0.000000e+00, double 0.000000e+00>, %a
248  %4 = fadd reassoc <2 x double> %2, %3
249  ret <2 x double> %4
250}
251
252; Check -(-(z*40)*a) -> a*40*z.
253
254define <2 x float> @test10(<2 x float> %a, <2 x float> %b, <2 x float> %z) {
255; CHECK-LABEL: @test10(
256; CHECK-NEXT:    [[TMP1:%.*]] = fsub fast <2 x float> zeroinitializer, zeroinitializer
257; CHECK-NEXT:    [[E:%.*]] = fmul fast <2 x float> [[A:%.*]], <float 4.000000e+01, float 4.000000e+01>
258; CHECK-NEXT:    [[F:%.*]] = fmul fast <2 x float> [[E]], [[Z:%.*]]
259; CHECK-NEXT:    ret <2 x float> [[F]]
260;
261  %d = fmul fast <2 x float> %z, <float 4.000000e+01, float 4.000000e+01>
262  %c = fsub fast <2 x float> <float 0.000000e+00, float 0.000000e+00>, %d
263  %e = fmul fast <2 x float> %a, %c
264  %f = fsub fast <2 x float> <float 0.000000e+00, float 0.000000e+00>, %e
265  ret <2 x float> %f
266}
267
268; Check -(-(z*40)*a) -> a*40*z - minimum FMF subset version
269
270define <2 x float> @test10_reassoc(<2 x float> %a, <2 x float> %b, <2 x float> %z) {
271; CHECK-LABEL: @test10_reassoc(
272; CHECK-NEXT:    [[D:%.*]] = fmul reassoc <2 x float> [[Z:%.*]], <float 4.000000e+01, float 4.000000e+01>
273; CHECK-NEXT:    [[C:%.*]] = fsub reassoc <2 x float> zeroinitializer, [[D]]
274; CHECK-NEXT:    [[E:%.*]] = fmul reassoc <2 x float> [[A:%.*]], [[C]]
275; CHECK-NEXT:    [[F:%.*]] = fsub reassoc <2 x float> zeroinitializer, [[E]]
276; CHECK-NEXT:    ret <2 x float> [[F]]
277;
278  %d = fmul reassoc <2 x float> %z, <float 4.000000e+01, float 4.000000e+01>
279  %c = fsub reassoc <2 x float> <float 0.000000e+00, float 0.000000e+00>, %d
280  %e = fmul reassoc <2 x float> %a, %c
281  %f = fsub reassoc <2 x float> <float 0.000000e+00, float 0.000000e+00>, %e
282  ret <2 x float> %f
283}
284
285; Check x*y+y*x -> x*y*2.
286
287define <2 x double> @test11(<2 x double> %x, <2 x double> %y) {
288; CHECK-LABEL: @test11(
289; CHECK-NEXT:    [[FACTOR:%.*]] = fmul fast <2 x double> [[X:%.*]], [[Y:%.*]]
290; CHECK-NEXT:    [[REASS_MUL:%.*]] = fmul fast <2 x double> [[FACTOR]], <double 2.000000e+00, double 2.000000e+00>
291; CHECK-NEXT:    ret <2 x double> [[REASS_MUL]]
292;
293  %1 = fmul fast <2 x double> %x, %y
294  %2 = fmul fast <2 x double> %y, %x
295  %3 = fadd fast <2 x double> %1, %2
296  ret <2 x double> %3
297}
298
299; Check x*y+y*x -> x*y*2 - minimum FMF subset version
300
301define <2 x double> @test11_reassoc(<2 x double> %x, <2 x double> %y) {
302; CHECK-LABEL: @test11_reassoc(
303; CHECK-NEXT:    [[TMP1:%.*]] = fmul reassoc <2 x double> [[X:%.*]], [[Y:%.*]]
304; CHECK-NEXT:    [[TMP2:%.*]] = fmul reassoc <2 x double> [[X]], [[Y]]
305; CHECK-NEXT:    [[TMP3:%.*]] = fadd reassoc <2 x double> [[TMP1]], [[TMP2]]
306; CHECK-NEXT:    ret <2 x double> [[TMP3]]
307;
308  %1 = fmul reassoc <2 x double> %x, %y
309  %2 = fmul reassoc <2 x double> %y, %x
310  %3 = fadd reassoc <2 x double> %1, %2
311  ret <2 x double> %3
312}
313
314; FIXME: shifts should be converted to mul to assist further reassociation.
315
316define <2 x i64> @test12(<2 x i64> %b, <2 x i64> %c) {
317; CHECK-LABEL: @test12(
318; CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i64> [[C:%.*]], [[B:%.*]]
319; CHECK-NEXT:    [[SHL:%.*]] = shl <2 x i64> [[MUL]], <i64 5, i64 5>
320; CHECK-NEXT:    ret <2 x i64> [[SHL]]
321;
322  %mul = mul <2 x i64> %c, %b
323  %shl = shl <2 x i64> %mul, <i64 5, i64 5>
324  ret <2 x i64> %shl
325}
326
327; FIXME: expressions with a negative const should be canonicalized to assist
328; further reassociation.
329; We would expect (-5*b)+a -> a-(5*b) but only the constant operand is commuted.
330
331define <4 x float> @test13(<4 x float> %a, <4 x float> %b) {
332; CHECK-LABEL: @test13(
333; CHECK-NEXT:    [[MUL:%.*]] = fmul fast <4 x float> [[B:%.*]], <float -5.000000e+00, float -5.000000e+00, float -5.000000e+00, float -5.000000e+00>
334; CHECK-NEXT:    [[ADD:%.*]] = fadd fast <4 x float> [[MUL]], [[A:%.*]]
335; CHECK-NEXT:    ret <4 x float> [[ADD]]
336;
337  %mul = fmul fast <4 x float> <float -5.000000e+00, float -5.000000e+00, float -5.000000e+00, float -5.000000e+00>, %b
338  %add = fadd fast <4 x float> %mul, %a
339  ret <4 x float> %add
340}
341
342; Break up subtract to assist further reassociation.
343; Check a+b-c -> a+b+-c.
344
345define <2 x i64> @test14(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c) {
346; CHECK-LABEL: @test14(
347; CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[B:%.*]], [[A:%.*]]
348; CHECK-NEXT:    [[C_NEG:%.*]] = sub <2 x i64> zeroinitializer, [[C:%.*]]
349; CHECK-NEXT:    [[SUB:%.*]] = add <2 x i64> [[ADD]], [[C_NEG]]
350; CHECK-NEXT:    ret <2 x i64> [[SUB]]
351;
352  %add = add <2 x i64> %b, %a
353  %sub = sub <2 x i64> %add, %c
354  ret <2 x i64> %sub
355}
356
357define <2 x i32> @test15(<2 x i32> %x, <2 x i32> %y) {
358; CHECK-LABEL: @test15(
359; CHECK-NEXT:    [[TMP3:%.*]] = and <2 x i32> [[Y:%.*]], [[X:%.*]]
360; CHECK-NEXT:    ret <2 x i32> [[TMP3]]
361;
362  %tmp1 = and <2 x i32> %x, %y
363  %tmp2 = and <2 x i32> %y, %x
364  %tmp3 = and <2 x i32> %tmp1, %tmp2
365  ret <2 x i32> %tmp3
366}
367
368define <2 x i32> @test16(<2 x i32> %x, <2 x i32> %y) {
369; CHECK-LABEL: @test16(
370; CHECK-NEXT:    [[TMP3:%.*]] = or <2 x i32> [[Y:%.*]], [[X:%.*]]
371; CHECK-NEXT:    ret <2 x i32> [[TMP3]]
372;
373  %tmp1 = or <2 x i32> %x, %y
374  %tmp2 = or <2 x i32> %y, %x
375  %tmp3 = or <2 x i32> %tmp1, %tmp2
376  ret <2 x i32> %tmp3
377}
378
379define <2 x i32> @test17(<2 x i32> %x, <2 x i32> %y) {
380; CHECK-LABEL: @test17(
381; CHECK-NEXT:    ret <2 x i32> zeroinitializer
382;
383  %tmp1 = xor <2 x i32> %x, %y
384  %tmp2 = xor <2 x i32> %y, %x
385  %tmp3 = xor <2 x i32> %tmp1, %tmp2
386  ret <2 x i32> %tmp3
387}
388
389define <2 x i32> @test18(<2 x i32> %x, <2 x i32> %y) {
390; CHECK-LABEL: @test18(
391; CHECK-NEXT:    [[TMP5:%.*]] = xor <2 x i32> [[Y:%.*]], [[X:%.*]]
392; CHECK-NEXT:    ret <2 x i32> [[TMP5]]
393;
394  %tmp1 = xor <2 x i32> %x, %y
395  %tmp2 = xor <2 x i32> %y, %x
396  %tmp3 = xor <2 x i32> %x, %y
397  %tmp4 = xor <2 x i32> %tmp1, %tmp2
398  %tmp5 = xor <2 x i32> %tmp4, %tmp3
399  ret <2 x i32> %tmp5
400}
401