Mon P Wang | f4646d9 | 2009-01-28 08:13:56 +0000 | [diff] [blame] | 1 | ; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 -disable-mmx -o %t -f |
| 2 | ; RUN: grep psrad %t | count 2 |
| 3 | ; RUN: grep psraw %t | count 2 |
| 4 | |
| 5 | ; test vector shifts converted to proper SSE2 vector shifts when the shift |
| 6 | ; amounts are the same. |
| 7 | |
| 8 | ; Note that x86 does have ashr |
| 9 | define void @shift1a(<2 x i64> %val, <2 x i64>* %dst) nounwind { |
| 10 | entry: |
| 11 | %ashr = ashr <2 x i64> %val, < i64 32, i64 32 > |
| 12 | store <2 x i64> %ashr, <2 x i64>* %dst |
| 13 | ret void |
| 14 | } |
| 15 | |
| 16 | define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind { |
| 17 | entry: |
| 18 | %ashr = ashr <4 x i32> %val, < i32 5, i32 5, i32 5, i32 5 > |
| 19 | store <4 x i32> %ashr, <4 x i32>* %dst |
| 20 | ret void |
| 21 | } |
| 22 | |
| 23 | define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind { |
| 24 | entry: |
| 25 | %0 = insertelement <4 x i32> undef, i32 %amt, i32 0 |
| 26 | %1 = insertelement <4 x i32> %0, i32 %amt, i32 1 |
| 27 | %2 = insertelement <4 x i32> %1, i32 %amt, i32 2 |
| 28 | %3 = insertelement <4 x i32> %2, i32 %amt, i32 3 |
| 29 | %ashr = ashr <4 x i32> %val, %3 |
| 30 | store <4 x i32> %ashr, <4 x i32>* %dst |
| 31 | ret void |
| 32 | } |
| 33 | |
| 34 | define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind { |
| 35 | entry: |
| 36 | %ashr = ashr <8 x i16> %val, < i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5 > |
| 37 | store <8 x i16> %ashr, <8 x i16>* %dst |
| 38 | ret void |
| 39 | } |
| 40 | |
| 41 | define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind { |
| 42 | entry: |
| 43 | %0 = insertelement <8 x i16> undef, i16 %amt, i32 0 |
| 44 | %1 = insertelement <8 x i16> %0, i16 %amt, i32 1 |
| 45 | %2 = insertelement <8 x i16> %0, i16 %amt, i32 2 |
| 46 | %3 = insertelement <8 x i16> %0, i16 %amt, i32 3 |
| 47 | %4 = insertelement <8 x i16> %0, i16 %amt, i32 4 |
| 48 | %5 = insertelement <8 x i16> %0, i16 %amt, i32 5 |
| 49 | %6 = insertelement <8 x i16> %0, i16 %amt, i32 6 |
| 50 | %7 = insertelement <8 x i16> %0, i16 %amt, i32 7 |
| 51 | %ashr = ashr <8 x i16> %val, %7 |
| 52 | store <8 x i16> %ashr, <8 x i16>* %dst |
| 53 | ret void |
| 54 | } |