xref: /llvm-project/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vw-web-simplification.ll (revision 74f985b793bf4005e49736f8c2cef8b5cbf7c1ab)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs %s -o - --riscv-lower-ext-max-web-size=1 | FileCheck %s --check-prefixes=NO_FOLDING
3; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs %s -o - --riscv-lower-ext-max-web-size=1 | FileCheck %s --check-prefixes=NO_FOLDING
4; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs %s -o - --riscv-lower-ext-max-web-size=2 | FileCheck %s --check-prefixes=NO_FOLDING
5; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs %s -o - --riscv-lower-ext-max-web-size=2 | FileCheck %s --check-prefixes=NO_FOLDING
6; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs %s -o - --riscv-lower-ext-max-web-size=3 | FileCheck %s --check-prefixes=FOLDING
7; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs %s -o - --riscv-lower-ext-max-web-size=3 | FileCheck %s --check-prefixes=FOLDING
8; Check that the default value enables the web folding and
9; that it is bigger than 3.
10; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=FOLDING
11; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=FOLDING
12
13
14; Check that the add/sub/mul operations are all promoted into their
15; vw counterpart when the folding of the web size is increased to 3.
16; We need the web size to be at least 3 for the folding to happen, because
17; %c has 3 uses.
18define <2 x i16> @vwmul_v2i16_multiple_users(ptr %x, ptr %y, ptr %z) {
19; NO_FOLDING-LABEL: vwmul_v2i16_multiple_users:
20; NO_FOLDING:       # %bb.0:
21; NO_FOLDING-NEXT:    vsetivli zero, 2, e16, mf4, ta, ma
22; NO_FOLDING-NEXT:    vle8.v v8, (a0)
23; NO_FOLDING-NEXT:    vle8.v v9, (a1)
24; NO_FOLDING-NEXT:    vle8.v v10, (a2)
25; NO_FOLDING-NEXT:    vsext.vf2 v11, v8
26; NO_FOLDING-NEXT:    vsext.vf2 v8, v9
27; NO_FOLDING-NEXT:    vsext.vf2 v9, v10
28; NO_FOLDING-NEXT:    vmul.vv v8, v11, v8
29; NO_FOLDING-NEXT:    vadd.vv v10, v11, v9
30; NO_FOLDING-NEXT:    vsub.vv v9, v11, v9
31; NO_FOLDING-NEXT:    vor.vv v8, v8, v10
32; NO_FOLDING-NEXT:    vor.vv v8, v8, v9
33; NO_FOLDING-NEXT:    ret
34;
35; FOLDING-LABEL: vwmul_v2i16_multiple_users:
36; FOLDING:       # %bb.0:
37; FOLDING-NEXT:    vsetivli zero, 2, e8, mf8, ta, ma
38; FOLDING-NEXT:    vle8.v v8, (a0)
39; FOLDING-NEXT:    vle8.v v9, (a1)
40; FOLDING-NEXT:    vle8.v v10, (a2)
41; FOLDING-NEXT:    vwmul.vv v11, v8, v9
42; FOLDING-NEXT:    vwadd.vv v9, v8, v10
43; FOLDING-NEXT:    vwsub.vv v12, v8, v10
44; FOLDING-NEXT:    vsetvli zero, zero, e16, mf4, ta, ma
45; FOLDING-NEXT:    vor.vv v8, v11, v9
46; FOLDING-NEXT:    vor.vv v8, v8, v12
47; FOLDING-NEXT:    ret
48  %a = load <2 x i8>, ptr %x
49  %b = load <2 x i8>, ptr %y
50  %b2 = load <2 x i8>, ptr %z
51  %c = sext <2 x i8> %a to <2 x i16>
52  %d = sext <2 x i8> %b to <2 x i16>
53  %d2 = sext <2 x i8> %b2 to <2 x i16>
54  %e = mul <2 x i16> %c, %d
55  %f = add <2 x i16> %c, %d2
56  %g = sub <2 x i16> %c, %d2
57  %h = or <2 x i16> %e, %f
58  %i = or <2 x i16> %h, %g
59  ret <2 x i16> %i
60}
61