| Hal Finkel | c34e511 | 2012-02-01 03:51:43 +0000 | [diff] [blame] | 1 | target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128" |
| Chandler Carruth | 7383bfd6 | 2013-01-07 10:22:36 +0000 | [diff] [blame] | 2 | ; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -bb-vectorize-ignore-target-info -instcombine -gvn -S | FileCheck %s |
| 3 | ; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -bb-vectorize-search-limit=4 -bb-vectorize-ignore-target-info -instcombine -gvn -S | FileCheck %s -check-prefix=CHECK-SL4 |
| Hal Finkel | c34e511 | 2012-02-01 03:51:43 +0000 | [diff] [blame] | 4 | |
| 5 | define double @test1(double %A1, double %A2, double %B1, double %B2) { |
| Stephen Lin | c1c7a13 | 2013-07-14 01:42:54 +0000 | [diff] [blame] | 6 | ; CHECK-LABEL: @test1( |
| 7 | ; CHECK-SL4-LABEL: @test1( |
| Hal Finkel | c34e511 | 2012-02-01 03:51:43 +0000 | [diff] [blame] | 8 | ; CHECK-SL4-NOT: <2 x double> |
| 9 | ; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0 |
| Hal Finkel | c34e511 | 2012-02-01 03:51:43 +0000 | [diff] [blame] | 10 | ; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1 |
| Hal Finkel | 842ad0b | 2012-10-31 15:17:07 +0000 | [diff] [blame] | 11 | ; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0 |
| Hal Finkel | c34e511 | 2012-02-01 03:51:43 +0000 | [diff] [blame] | 12 | ; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1 |
| 13 | %X1 = fsub double %A1, %B1 |
| 14 | %X2 = fsub double %A2, %B2 |
| 15 | ; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2 |
| 16 | %Y1 = fmul double %X1, %A1 |
| 17 | %Y2 = fmul double %X2, %A2 |
| 18 | ; CHECK: %Y1 = fmul <2 x double> %X1, %X1.v.i0.2 |
| 19 | %Z1 = fadd double %Y1, %B1 |
| 20 | ; Here we have a dependency chain: the short search limit will not |
| 21 | ; see past this chain and so will not see the second part of the |
| 22 | ; pair to vectorize. |
| 23 | %mul41 = fmul double %Z1, %Y2 |
| 24 | %sub48 = fsub double %Z1, %mul41 |
| 25 | %mul62 = fmul double %Z1, %sub48 |
| 26 | %sub69 = fsub double %Z1, %mul62 |
| 27 | %mul83 = fmul double %Z1, %sub69 |
| 28 | %sub90 = fsub double %Z1, %mul83 |
| 29 | %mul104 = fmul double %Z1, %sub90 |
| 30 | %sub111 = fsub double %Z1, %mul104 |
| 31 | %mul125 = fmul double %Z1, %sub111 |
| 32 | %sub132 = fsub double %Z1, %mul125 |
| 33 | %mul146 = fmul double %Z1, %sub132 |
| 34 | %sub153 = fsub double %Z1, %mul146 |
| 35 | ; end of chain. |
| 36 | %Z2 = fadd double %Y2, %B2 |
| 37 | ; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2 |
| 38 | %R1 = fdiv double %Z1, %Z2 |
| 39 | %R = fmul double %R1, %sub153 |
| 40 | ; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0 |
| 41 | ; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1 |
| 42 | ; CHECK: %R1 = fdiv double %Z1.v.r1, %Z1.v.r2 |
| 43 | ret double %R |
| 44 | ; CHECK: ret double %R |
| 45 | } |
| 46 | |