diff options
| author | Evan Cheng <evan.cheng@apple.com> | 2006-04-10 21:42:19 +0000 |
|---|---|---|
| committer | Evan Cheng <evan.cheng@apple.com> | 2006-04-10 21:42:19 +0000 |
| commit | 76112c3cb8cf30e1bc88bc8a6d7518c353f26bcd (patch) | |
| tree | b75ef560d191b74dd34a3e9c74f4b887765a7077 | |
| parent | 664fcba5faaa3bc29f7623b45891d4c105c1def1 (diff) | |
| download | bcm5719-llvm-76112c3cb8cf30e1bc88bc8a6d7518c353f26bcd.tar.gz bcm5719-llvm-76112c3cb8cf30e1bc88bc8a6d7518c353f26bcd.zip | |
Added some missing shuffle patterns.
llvm-svn: 27564
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrSSE.td | 27 |
1 files changed, 22 insertions, 5 deletions
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td index e8667b52988..4664440445f 100644 --- a/llvm/lib/Target/X86/X86InstrSSE.td +++ b/llvm/lib/Target/X86/X86InstrSSE.td @@ -1365,7 +1365,7 @@ def PSHUFHWmi : Ii8<0x70, MRMSrcMem, (ops VR128:$dst, i128mem:$src1, i8imm:$src2), "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (v8i16 (vector_shuffle - (bc_v8i16 (loadv2i64 addr:$src1)), (undef), + (load addr:$src1), (undef), PSHUFHW_shuffle_mask:$src2)))]>, XS, Requires<[HasSSE2]>; @@ -1381,7 +1381,7 @@ def PSHUFLWmi : Ii8<0x70, MRMSrcMem, (ops VR128:$dst, i128mem:$src1, i32i8imm:$src2), "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (v8i16 (vector_shuffle - (bc_v8i16 (loadv2i64 addr:$src1)), (undef), + (load addr:$src1), (undef), PSHUFLW_shuffle_mask:$src2)))]>, XD, Requires<[HasSSE2]>; @@ -1823,11 +1823,28 @@ def : Pat<(vector_shuffle (v4i32 VR128:$src1), (load addr:$src2), (v4i32 (SHUFPSrm VR128:$src1, addr:$src2, SHUFP_int_shuffle_mask:$sm))>, Requires<[HasSSE2]>; -// Shuffle v4f32 with PSHUF* if others do not match. +// Special pshuf* cases: folding (bit_convert (loadv2i64 addr)). +def : Pat<(v4i32 (vector_shuffle (bc_v4i32 (loadv2i64 addr:$src1)), (undef), + PSHUFD_shuffle_mask:$src2)), + (PSHUFDmi addr:$src1, PSHUFD_shuffle_mask:$src2)>, + Requires<[HasSSE2]>; +def : Pat<(v8i16 (vector_shuffle (bc_v8i16 (loadv2i64 addr:$src1)), (undef), + PSHUFHW_shuffle_mask:$src2)), + (PSHUFHWmi addr:$src1, PSHUFHW_shuffle_mask:$src2)>, + Requires<[HasSSE2]>; +def : Pat<(v8i16 (vector_shuffle (bc_v8i16 (loadv2i64 addr:$src1)), (undef), + PSHUFLW_shuffle_mask:$src2)), + (PSHUFLWmi addr:$src1, PSHUFHW_shuffle_mask:$src2)>, + Requires<[HasSSE2]>; + + +// Special SHUFPSrr case: looks like a PSHUFD, like make both operands src1. +// FIXME: when we want non two-address code, then we should use PSHUFD! def : Pat<(vector_shuffle (v4f32 VR128:$src1), (undef), PSHUFD_fp_shuffle_mask:$sm), - (v4f32 (PSHUFDri VR128:$src1, PSHUFD_fp_shuffle_mask:$sm))>, - Requires<[HasSSE2]>; + (v4f32 (SHUFPSrr VR128:$src1, VR128:$src1, PSHUFD_fp_shuffle_mask:$sm))>, + Requires<[HasSSE1]>; +// Shuffle v4f32 with PSHUF* if others do not match. def : Pat<(vector_shuffle (loadv4f32 addr:$src1), (undef), PSHUFD_fp_shuffle_mask:$sm), (v4f32 (PSHUFDmi addr:$src1, PSHUFD_fp_shuffle_mask:$sm))>, |
