Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / 118e53fd6370c165dcd297c5a16ef3b8fd3b469f / . / llvm / test / CodeGen / X86 / smul_fix.ll
blob: 2e69a2666f9ab9e85ff2fc7dfcc8e54f0004c7e4 [file] [log] [blame]
Leonard Chan118e53f2018-12-12 06:29:14 +0000[diff] [blame^]1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s --check-prefix=X64
3; RUN: llc < %s -mtriple=i686 -mattr=cmov | FileCheck %s --check-prefix=X86
4
5declare i4 @llvm.smul.fix.i4 (i4, i4, i32)
6declare i32 @llvm.smul.fix.i32 (i32, i32, i32)
7declare i64 @llvm.smul.fix.i64 (i64, i64, i32)
8declare <4 x i32> @llvm.smul.fix.v4i32(<4 x i32>, <4 x i32>, i32)
9
10define i32 @func(i32 %x, i32 %y) nounwind {
11; X64-LABEL: func:
12; X64: # %bb.0:
13; X64-NEXT: movslq %esi, %rax
14; X64-NEXT: movslq %edi, %rcx
15; X64-NEXT: imulq %rax, %rcx
16; X64-NEXT: movq %rcx, %rax
17; X64-NEXT: shrq $32, %rax
18; X64-NEXT: shldl $30, %ecx, %eax
19; X64-NEXT: # kill: def $eax killed $eax killed $rax
20; X64-NEXT: retq
21;
22; X86-LABEL: func:
23; X86: # %bb.0:
24; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
25; X86-NEXT: imull {{[0-9]+}}(%esp)
26; X86-NEXT: shrdl $2, %edx, %eax
27; X86-NEXT: retl
28 %tmp = call i32 @llvm.smul.fix.i32(i32 %x, i32 %y, i32 2);
29 ret i32 %tmp;
30}
31
32define i64 @func2(i64 %x, i64 %y) {
33; X64-LABEL: func2:
34; X64: # %bb.0:
35; X64-NEXT: movq %rdi, %rax
36; X64-NEXT: imulq %rsi
37; X64-NEXT: shrdq $2, %rdx, %rax
38; X64-NEXT: retq
39;
40; X86-LABEL: func2:
41; X86: # %bb.0:
42; X86-NEXT: pushl %ebp
43; X86-NEXT: .cfi_def_cfa_offset 8
44; X86-NEXT: pushl %ebx
45; X86-NEXT: .cfi_def_cfa_offset 12
46; X86-NEXT: pushl %edi
47; X86-NEXT: .cfi_def_cfa_offset 16
48; X86-NEXT: pushl %esi
49; X86-NEXT: .cfi_def_cfa_offset 20
50; X86-NEXT: .cfi_offset %esi, -20
51; X86-NEXT: .cfi_offset %edi, -16
52; X86-NEXT: .cfi_offset %ebx, -12
53; X86-NEXT: .cfi_offset %ebp, -8
54; X86-NEXT: movl {{[0-9]+}}(%esp), %ebx
55; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
56; X86-NEXT: movl %ebx, %eax
57; X86-NEXT: mull {{[0-9]+}}(%esp)
58; X86-NEXT: movl %edx, %esi
59; X86-NEXT: movl %eax, %edi
60; X86-NEXT: movl %ebx, %eax
61; X86-NEXT: mull %ecx
62; X86-NEXT: movl %eax, %ebx
63; X86-NEXT: movl %edx, %ebp
64; X86-NEXT: addl %edi, %ebp
65; X86-NEXT: movl {{[0-9]+}}(%esp), %edi
66; X86-NEXT: adcl $0, %esi
67; X86-NEXT: movl %edi, %eax
68; X86-NEXT: mull %ecx
69; X86-NEXT: addl %ebp, %eax
70; X86-NEXT: adcl %esi, %edx
71; X86-NEXT: movl %edi, %esi
72; X86-NEXT: imull {{[0-9]+}}(%esp), %esi
73; X86-NEXT: addl %edx, %esi
74; X86-NEXT: movl %esi, %ebp
75; X86-NEXT: subl %ecx, %ebp
76; X86-NEXT: testl %edi, %edi
77; X86-NEXT: cmovnsl %esi, %ebp
78; X86-NEXT: movl %ebp, %edx
79; X86-NEXT: subl {{[0-9]+}}(%esp), %edx
80; X86-NEXT: cmpl $0, {{[0-9]+}}(%esp)
81; X86-NEXT: cmovnsl %ebp, %edx
82; X86-NEXT: shldl $30, %eax, %edx
83; X86-NEXT: shldl $30, %ebx, %eax
84; X86-NEXT: popl %esi
85; X86-NEXT: .cfi_def_cfa_offset 16
86; X86-NEXT: popl %edi
87; X86-NEXT: .cfi_def_cfa_offset 12
88; X86-NEXT: popl %ebx
89; X86-NEXT: .cfi_def_cfa_offset 8
90; X86-NEXT: popl %ebp
91; X86-NEXT: .cfi_def_cfa_offset 4
92; X86-NEXT: retl
93 %tmp = call i64 @llvm.smul.fix.i64(i64 %x, i64 %y, i32 2);
94 ret i64 %tmp;
95}
96
97define i4 @func3(i4 %x, i4 %y) nounwind {
98; X64-LABEL: func3:
99; X64: # %bb.0:
100; X64-NEXT: shlb $4, %dil
101; X64-NEXT: sarb $4, %dil
102; X64-NEXT: shlb $4, %sil
103; X64-NEXT: sarb $4, %sil
104; X64-NEXT: movsbl %sil, %ecx
105; X64-NEXT: movsbl %dil, %eax
106; X64-NEXT: imull %ecx, %eax
107; X64-NEXT: movl %eax, %ecx
108; X64-NEXT: shrb $2, %cl
109; X64-NEXT: shrl $8, %eax
110; X64-NEXT: shlb $6, %al
111; X64-NEXT: orb %cl, %al
112; X64-NEXT: # kill: def $al killed $al killed $eax
113; X64-NEXT: retq
114;
115; X86-LABEL: func3:
116; X86: # %bb.0:
117; X86-NEXT: movb {{[0-9]+}}(%esp), %al
118; X86-NEXT: shlb $4, %al
119; X86-NEXT: sarb $4, %al
120; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
121; X86-NEXT: shlb $4, %cl
122; X86-NEXT: sarb $4, %cl
123; X86-NEXT: movsbl %cl, %ecx
124; X86-NEXT: movsbl %al, %eax
125; X86-NEXT: imull %ecx, %eax
126; X86-NEXT: shlb $6, %ah
127; X86-NEXT: shrb $2, %al
128; X86-NEXT: orb %ah, %al
129; X86-NEXT: # kill: def $al killed $al killed $eax
130; X86-NEXT: retl
131 %tmp = call i4 @llvm.smul.fix.i4(i4 %x, i4 %y, i32 2);
132 ret i4 %tmp;
133}
134
135define <4 x i32> @vec(<4 x i32> %x, <4 x i32> %y) nounwind {
136; X64-LABEL: vec:
137; X64: # %bb.0:
138; X64-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,1,2,3]
139; X64-NEXT: movd %xmm2, %eax
140; X64-NEXT: cltq
141; X64-NEXT: pshufd {{.*#+}} xmm2 = xmm0[3,1,2,3]
142; X64-NEXT: movd %xmm2, %ecx
143; X64-NEXT: movslq %ecx, %rcx
144; X64-NEXT: imulq %rax, %rcx
145; X64-NEXT: movq %rcx, %rax
146; X64-NEXT: shrq $32, %rax
147; X64-NEXT: shldl $30, %ecx, %eax
148; X64-NEXT: movd %eax, %xmm2
149; X64-NEXT: pshufd {{.*#+}} xmm3 = xmm1[2,3,0,1]
150; X64-NEXT: movd %xmm3, %eax
151; X64-NEXT: cltq
152; X64-NEXT: pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
153; X64-NEXT: movd %xmm3, %ecx
154; X64-NEXT: movslq %ecx, %rcx
155; X64-NEXT: imulq %rax, %rcx
156; X64-NEXT: movq %rcx, %rax
157; X64-NEXT: shrq $32, %rax
158; X64-NEXT: shldl $30, %ecx, %eax
159; X64-NEXT: movd %eax, %xmm3
160; X64-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
161; X64-NEXT: movd %xmm1, %eax
162; X64-NEXT: cltq
163; X64-NEXT: movd %xmm0, %ecx
164; X64-NEXT: movslq %ecx, %rcx
165; X64-NEXT: imulq %rax, %rcx
166; X64-NEXT: movq %rcx, %rax
167; X64-NEXT: shrq $32, %rax
168; X64-NEXT: shldl $30, %ecx, %eax
169; X64-NEXT: movd %eax, %xmm2
170; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
171; X64-NEXT: movd %xmm1, %eax
172; X64-NEXT: cltq
173; X64-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
174; X64-NEXT: movd %xmm0, %ecx
175; X64-NEXT: movslq %ecx, %rcx
176; X64-NEXT: imulq %rax, %rcx
177; X64-NEXT: movq %rcx, %rax
178; X64-NEXT: shrq $32, %rax
179; X64-NEXT: shldl $30, %ecx, %eax
180; X64-NEXT: movd %eax, %xmm0
181; X64-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
182; X64-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
183; X64-NEXT: movdqa %xmm2, %xmm0
184; X64-NEXT: retq
185;
186; X86-LABEL: vec:
187; X86: # %bb.0:
188; X86-NEXT: pushl %ebp
189; X86-NEXT: pushl %ebx
190; X86-NEXT: pushl %edi
191; X86-NEXT: pushl %esi
192; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
193; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
194; X86-NEXT: movl {{[0-9]+}}(%esp), %edi
195; X86-NEXT: movl {{[0-9]+}}(%esp), %ebx
196; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
197; X86-NEXT: imull {{[0-9]+}}(%esp)
198; X86-NEXT: movl %edx, %ebp
199; X86-NEXT: shldl $30, %eax, %ebp
200; X86-NEXT: movl %ebx, %eax
201; X86-NEXT: imull {{[0-9]+}}(%esp)
202; X86-NEXT: movl %edx, %ebx
203; X86-NEXT: shldl $30, %eax, %ebx
204; X86-NEXT: movl %edi, %eax
205; X86-NEXT: imull {{[0-9]+}}(%esp)
206; X86-NEXT: movl %edx, %edi
207; X86-NEXT: shldl $30, %eax, %edi
208; X86-NEXT: movl %esi, %eax
209; X86-NEXT: imull {{[0-9]+}}(%esp)
210; X86-NEXT: shldl $30, %eax, %edx
211; X86-NEXT: movl %edx, 12(%ecx)
212; X86-NEXT: movl %edi, 8(%ecx)
213; X86-NEXT: movl %ebx, 4(%ecx)
214; X86-NEXT: movl %ebp, (%ecx)
215; X86-NEXT: movl %ecx, %eax
216; X86-NEXT: popl %esi
217; X86-NEXT: popl %edi
218; X86-NEXT: popl %ebx
219; X86-NEXT: popl %ebp
220; X86-NEXT: retl $4
221 %tmp = call <4 x i32> @llvm.smul.fix.v4i32(<4 x i32> %x, <4 x i32> %y, i32 2);
222 ret <4 x i32> %tmp;
223}
224
225; These result in regular integer multiplication
226define i32 @func4(i32 %x, i32 %y) nounwind {
227; X64-LABEL: func4:
228; X64: # %bb.0:
229; X64-NEXT: movl %edi, %eax
230; X64-NEXT: imull %esi, %eax
231; X64-NEXT: retq
232;
233; X86-LABEL: func4:
234; X86: # %bb.0:
235; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
236; X86-NEXT: imull {{[0-9]+}}(%esp), %eax
237; X86-NEXT: retl
238 %tmp = call i32 @llvm.smul.fix.i32(i32 %x, i32 %y, i32 0);
239 ret i32 %tmp;
240}
241
242define i64 @func5(i64 %x, i64 %y) {
243; X64-LABEL: func5:
244; X64: # %bb.0:
245; X64-NEXT: movq %rdi, %rax
246; X64-NEXT: imulq %rsi, %rax
247; X64-NEXT: retq
248;
249; X86-LABEL: func5:
250; X86: # %bb.0:
251; X86-NEXT: pushl %esi
252; X86-NEXT: .cfi_def_cfa_offset 8
253; X86-NEXT: .cfi_offset %esi, -8
254; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
255; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
256; X86-NEXT: movl %ecx, %eax
257; X86-NEXT: mull %esi
258; X86-NEXT: imull {{[0-9]+}}(%esp), %ecx
259; X86-NEXT: addl %ecx, %edx
260; X86-NEXT: imull {{[0-9]+}}(%esp), %esi
261; X86-NEXT: addl %esi, %edx
262; X86-NEXT: popl %esi
263; X86-NEXT: .cfi_def_cfa_offset 4
264; X86-NEXT: retl
265 %tmp = call i64 @llvm.smul.fix.i64(i64 %x, i64 %y, i32 0);
266 ret i64 %tmp;
267}
268
269define i4 @func6(i4 %x, i4 %y) nounwind {
270; X64-LABEL: func6:
271; X64: # %bb.0:
272; X64-NEXT: movl %edi, %eax
273; X64-NEXT: shlb $4, %al
274; X64-NEXT: sarb $4, %al
275; X64-NEXT: shlb $4, %sil
276; X64-NEXT: sarb $4, %sil
277; X64-NEXT: # kill: def $al killed $al killed $eax
278; X64-NEXT: mulb %sil
279; X64-NEXT: retq
280;
281; X86-LABEL: func6:
282; X86: # %bb.0:
283; X86-NEXT: movb {{[0-9]+}}(%esp), %al
284; X86-NEXT: shlb $4, %al
285; X86-NEXT: sarb $4, %al
286; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
287; X86-NEXT: shlb $4, %cl
288; X86-NEXT: sarb $4, %cl
289; X86-NEXT: mulb %cl
290; X86-NEXT: retl
291 %tmp = call i4 @llvm.smul.fix.i4(i4 %x, i4 %y, i32 0);
292 ret i4 %tmp;
293}
294
295define <4 x i32> @vec2(<4 x i32> %x, <4 x i32> %y) nounwind {
296; X64-LABEL: vec2:
297; X64: # %bb.0:
298; X64-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,1,2,3]
299; X64-NEXT: movd %xmm2, %eax
300; X64-NEXT: pshufd {{.*#+}} xmm2 = xmm0[3,1,2,3]
301; X64-NEXT: movd %xmm2, %ecx
302; X64-NEXT: imull %eax, %ecx
303; X64-NEXT: movd %ecx, %xmm2
304; X64-NEXT: pshufd {{.*#+}} xmm3 = xmm1[2,3,0,1]
305; X64-NEXT: movd %xmm3, %eax
306; X64-NEXT: pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
307; X64-NEXT: movd %xmm3, %ecx
308; X64-NEXT: imull %eax, %ecx
309; X64-NEXT: movd %ecx, %xmm3
310; X64-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
311; X64-NEXT: movd %xmm1, %eax
312; X64-NEXT: movd %xmm0, %ecx
313; X64-NEXT: imull %eax, %ecx
314; X64-NEXT: movd %ecx, %xmm2
315; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
316; X64-NEXT: movd %xmm1, %eax
317; X64-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
318; X64-NEXT: movd %xmm0, %ecx
319; X64-NEXT: imull %eax, %ecx
320; X64-NEXT: movd %ecx, %xmm0
321; X64-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
322; X64-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
323; X64-NEXT: movdqa %xmm2, %xmm0
324; X64-NEXT: retq
325;
326; X86-LABEL: vec2:
327; X86: # %bb.0:
328; X86-NEXT: pushl %edi
329; X86-NEXT: pushl %esi
330; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
331; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
332; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
333; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
334; X86-NEXT: movl {{[0-9]+}}(%esp), %edi
335; X86-NEXT: imull {{[0-9]+}}(%esp), %edi
336; X86-NEXT: imull {{[0-9]+}}(%esp), %esi
337; X86-NEXT: imull {{[0-9]+}}(%esp), %edx
338; X86-NEXT: imull {{[0-9]+}}(%esp), %ecx
339; X86-NEXT: movl %ecx, 12(%eax)
340; X86-NEXT: movl %edx, 8(%eax)
341; X86-NEXT: movl %esi, 4(%eax)
342; X86-NEXT: movl %edi, (%eax)
343; X86-NEXT: popl %esi
344; X86-NEXT: popl %edi
345; X86-NEXT: retl $4
346 %tmp = call <4 x i32> @llvm.smul.fix.v4i32(<4 x i32> %x, <4 x i32> %y, i32 0);
347 ret <4 x i32> %tmp;
348}
349
350define i64 @func7(i64 %x, i64 %y) nounwind {
351; X64-LABEL: func7:
352; X64: # %bb.0:
353; X64-NEXT: movq %rdi, %rax
354; X64-NEXT: imulq %rsi
355; X64-NEXT: shrdq $32, %rdx, %rax
356; X64-NEXT: retq
357;
358; X86-LABEL: func7:
359; X86: # %bb.0:
360; X86-NEXT: pushl %ebp
361; X86-NEXT: pushl %ebx
362; X86-NEXT: pushl %edi
363; X86-NEXT: pushl %esi
364; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
365; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
366; X86-NEXT: movl {{[0-9]+}}(%esp), %ebp
367; X86-NEXT: movl %ecx, %eax
368; X86-NEXT: mull {{[0-9]+}}(%esp)
369; X86-NEXT: movl %edx, %edi
370; X86-NEXT: movl %eax, %ebx
371; X86-NEXT: movl %ecx, %eax
372; X86-NEXT: mull %ebp
373; X86-NEXT: addl %edx, %ebx
374; X86-NEXT: adcl $0, %edi
375; X86-NEXT: movl %esi, %eax
376; X86-NEXT: mull %ebp
377; X86-NEXT: addl %ebx, %eax
378; X86-NEXT: adcl %edi, %edx
379; X86-NEXT: movl %esi, %edi
380; X86-NEXT: imull {{[0-9]+}}(%esp), %edi
381; X86-NEXT: addl %edx, %edi
382; X86-NEXT: movl %edi, %ebx
383; X86-NEXT: subl %ebp, %ebx
384; X86-NEXT: testl %esi, %esi
385; X86-NEXT: cmovnsl %edi, %ebx
386; X86-NEXT: movl %ebx, %edx
387; X86-NEXT: subl %ecx, %edx
388; X86-NEXT: cmpl $0, {{[0-9]+}}(%esp)
389; X86-NEXT: cmovnsl %ebx, %edx
390; X86-NEXT: popl %esi
391; X86-NEXT: popl %edi
392; X86-NEXT: popl %ebx
393; X86-NEXT: popl %ebp
394; X86-NEXT: retl
395 %tmp = call i64 @llvm.smul.fix.i64(i64 %x, i64 %y, i32 32);
396 ret i64 %tmp;
397}
398
399define i64 @func8(i64 %x, i64 %y) nounwind {
400; X64-LABEL: func8:
401; X64: # %bb.0:
402; X64-NEXT: movq %rdi, %rax
403; X64-NEXT: imulq %rsi
404; X64-NEXT: shrdq $63, %rdx, %rax
405; X64-NEXT: retq
406;
407; X86-LABEL: func8:
408; X86: # %bb.0:
409; X86-NEXT: pushl %ebp
410; X86-NEXT: pushl %ebx
411; X86-NEXT: pushl %edi
412; X86-NEXT: pushl %esi
413; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
414; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
415; X86-NEXT: movl %ecx, %eax
416; X86-NEXT: mull {{[0-9]+}}(%esp)
417; X86-NEXT: movl %edx, %edi
418; X86-NEXT: movl %eax, %ebx
419; X86-NEXT: movl %ecx, %eax
420; X86-NEXT: mull {{[0-9]+}}(%esp)
421; X86-NEXT: movl %edx, %ebp
422; X86-NEXT: addl %ebx, %ebp
423; X86-NEXT: adcl $0, %edi
424; X86-NEXT: movl %esi, %eax
425; X86-NEXT: imull {{[0-9]+}}(%esp)
426; X86-NEXT: movl %edx, %ebx
427; X86-NEXT: movl %eax, %ecx
428; X86-NEXT: movl %esi, %eax
429; X86-NEXT: mull {{[0-9]+}}(%esp)
430; X86-NEXT: addl %ebp, %eax
431; X86-NEXT: adcl %edi, %edx
432; X86-NEXT: adcl $0, %ebx
433; X86-NEXT: addl %ecx, %edx
434; X86-NEXT: adcl $0, %ebx
435; X86-NEXT: movl %edx, %ecx
436; X86-NEXT: subl {{[0-9]+}}(%esp), %ecx
437; X86-NEXT: movl %ebx, %esi
438; X86-NEXT: sbbl $0, %esi
439; X86-NEXT: cmpl $0, {{[0-9]+}}(%esp)
440; X86-NEXT: cmovnsl %ebx, %esi
441; X86-NEXT: cmovnsl %edx, %ecx
442; X86-NEXT: movl %ecx, %edi
443; X86-NEXT: subl {{[0-9]+}}(%esp), %edi
444; X86-NEXT: movl %esi, %edx
445; X86-NEXT: sbbl $0, %edx
446; X86-NEXT: cmpl $0, {{[0-9]+}}(%esp)
447; X86-NEXT: cmovnsl %esi, %edx
448; X86-NEXT: cmovnsl %ecx, %edi
449; X86-NEXT: shldl $1, %edi, %edx
450; X86-NEXT: shrdl $31, %edi, %eax
451; X86-NEXT: popl %esi
452; X86-NEXT: popl %edi
453; X86-NEXT: popl %ebx
454; X86-NEXT: popl %ebp
455; X86-NEXT: retl
456 %tmp = call i64 @llvm.smul.fix.i64(i64 %x, i64 %y, i32 63);
457 ret i64 %tmp;
458}