| Nadav Rotem | b7bb72e | 2012-06-04 11:27:21 +0000 | [diff] [blame] | 1 | ; RUN: llc < %s -march=x86-64 -mcpu=corei7 -mattr=+sse41 | FileCheck %s |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 2 | |
| 3 | ;CHECK: load_2_i8 |
| 4 | ; A single 16-bit load |
| Michael Liao | 1be96bb | 2012-10-23 17:34:00 +0000 | [diff] [blame] | 5 | ;CHECK: pmovzxbq |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 6 | ;CHECK: paddq |
| 7 | ;CHECK: pshufb |
| 8 | ; A single 16-bit store |
| 9 | ;CHECK: movw |
| 10 | ;CHECK: ret |
| 11 | |
| 12 | define void @load_2_i8(<2 x i8>* %A) { |
| 13 | %T = load <2 x i8>* %A |
| 14 | %G = add <2 x i8> %T, <i8 9, i8 7> |
| 15 | store <2 x i8> %G, <2 x i8>* %A |
| 16 | ret void |
| 17 | } |
| 18 | |
| 19 | ;CHECK: load_2_i16 |
| 20 | ; Read 32-bits |
| Michael Liao | 1be96bb | 2012-10-23 17:34:00 +0000 | [diff] [blame] | 21 | ;CHECK: pmovzxwq |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 22 | ;CHECK: paddq |
| 23 | ;CHECK: pshufb |
| 24 | ;CHECK: movd |
| 25 | ;CHECK: ret |
| 26 | define void @load_2_i16(<2 x i16>* %A) { |
| 27 | %T = load <2 x i16>* %A |
| 28 | %G = add <2 x i16> %T, <i16 9, i16 7> |
| 29 | store <2 x i16> %G, <2 x i16>* %A |
| 30 | ret void |
| 31 | } |
| 32 | |
| 33 | ;CHECK: load_2_i32 |
| Michael Liao | 1be96bb | 2012-10-23 17:34:00 +0000 | [diff] [blame] | 34 | ;CHECK: pmovzxdq |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 35 | ;CHECK: paddq |
| 36 | ;CHECK: pshufd |
| 37 | ;CHECK: ret |
| 38 | define void @load_2_i32(<2 x i32>* %A) { |
| 39 | %T = load <2 x i32>* %A |
| 40 | %G = add <2 x i32> %T, <i32 9, i32 7> |
| 41 | store <2 x i32> %G, <2 x i32>* %A |
| 42 | ret void |
| 43 | } |
| 44 | |
| 45 | ;CHECK: load_4_i8 |
| Michael Liao | 1be96bb | 2012-10-23 17:34:00 +0000 | [diff] [blame] | 46 | ;CHECK: pmovzxbd |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 47 | ;CHECK: paddd |
| 48 | ;CHECK: pshufb |
| 49 | ;CHECK: ret |
| 50 | define void @load_4_i8(<4 x i8>* %A) { |
| 51 | %T = load <4 x i8>* %A |
| 52 | %G = add <4 x i8> %T, <i8 1, i8 4, i8 9, i8 7> |
| 53 | store <4 x i8> %G, <4 x i8>* %A |
| 54 | ret void |
| 55 | } |
| 56 | |
| 57 | ;CHECK: load_4_i16 |
| Michael Liao | 1be96bb | 2012-10-23 17:34:00 +0000 | [diff] [blame] | 58 | ;CHECK: pmovzxwd |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 59 | ;CHECK: paddd |
| 60 | ;CHECK: pshufb |
| 61 | ;CHECK: ret |
| 62 | define void @load_4_i16(<4 x i16>* %A) { |
| 63 | %T = load <4 x i16>* %A |
| 64 | %G = add <4 x i16> %T, <i16 1, i16 4, i16 9, i16 7> |
| 65 | store <4 x i16> %G, <4 x i16>* %A |
| 66 | ret void |
| 67 | } |
| 68 | |
| 69 | ;CHECK: load_8_i8 |
| Michael Liao | 1be96bb | 2012-10-23 17:34:00 +0000 | [diff] [blame] | 70 | ;CHECK: pmovzxbw |
| Nadav Rotem | 7ae1127 | 2011-09-18 10:39:32 +0000 | [diff] [blame] | 71 | ;CHECK: paddw |
| 72 | ;CHECK: pshufb |
| 73 | ;CHECK: ret |
| 74 | define void @load_8_i8(<8 x i8>* %A) { |
| 75 | %T = load <8 x i8>* %A |
| 76 | %G = add <8 x i8> %T, %T |
| 77 | store <8 x i8> %G, <8 x i8>* %A |
| 78 | ret void |
| 79 | } |