diff options
| author | Vince Weaver <vince@csl.cornell.edu> | 2009-10-30 12:49:37 -0400 |
|---|---|---|
| committer | Vince Weaver <vince@csl.cornell.edu> | 2009-10-30 12:49:37 -0400 |
| commit | a12557439ba0297fefb558b34ecc761d2bc34017 (patch) | |
| tree | 598ace115720808989fca84d89716f7e282142e8 /src/arch/x86 | |
| parent | 5873ec2238e521f34f7e70e6aa1b35a611225d80 (diff) | |
| download | gem5-a12557439ba0297fefb558b34ecc761d2bc34017.tar.xz | |
X86: Add support for x86 psrldq and pslldq instructions
These are complicated instructions and the micro-code might be suboptimal.
This has been tested with some small sample programs (attached)
The psrldq instruction is needed by various spec2k programs.
Diffstat (limited to 'src/arch/x86')
3 files changed, 78 insertions, 4 deletions
diff --git a/src/arch/x86/isa/decoder/two_byte_opcodes.isa b/src/arch/x86/isa/decoder/two_byte_opcodes.isa index d52dfb26e..288c5e5a8 100644 --- a/src/arch/x86/isa/decoder/two_byte_opcodes.isa +++ b/src/arch/x86/isa/decoder/two_byte_opcodes.isa @@ -673,9 +673,9 @@ //0x3: group14_pshimq(); 0x3: decode MODRM_REG { 0x2: PSRLQ(VRo,Ib); - 0x3: WarnUnimpl::psrldq_VRo_Ib(); + 0x3: PSRLDQ(VRo,Ib); 0x6: PSLLQ(VRo,Ib); - 0x7: WarnUnimpl::pslldq_VRo_Ib(); + 0x7: PSLLDQ(VRo,Ib); default: UD2(); } 0x4: PCMPEQB(Vo,Wo); diff --git a/src/arch/x86/isa/insts/simd128/integer/shift/left_logical_shift.py b/src/arch/x86/isa/insts/simd128/integer/shift/left_logical_shift.py index 617033bc0..c13c7064c 100644 --- a/src/arch/x86/isa/insts/simd128/integer/shift/left_logical_shift.py +++ b/src/arch/x86/isa/insts/simd128/integer/shift/left_logical_shift.py @@ -122,5 +122,43 @@ def macroop PSLLQ_XMM_I { mslli xmml, xmml, imm, size=8, ext=0 mslli xmmh, xmmh, imm, size=8, ext=0 }; + +def macroop PSLLDQ_XMM_I { + + limm t2, 8 + subi t1, t2, imm, flags=(ECF,), dataSize=1 + br label("pslldq_less_8"), flags=(nCECF,) + + # Greater than 8 + + limm t2, 16 + subi t1, t2, imm, flags=(ECF,), dataSize=1 + br label("pslldq_less_16"), flags=(nCECF,) + + # Greater than 16 + + lfpimm xmml, 0 + lfpimm xmmh, 0 + br label("pslldq_end") + +pslldq_less_16: + + # Between 8 and 16 + + mslli xmmh, xmml, "(IMMEDIATE-8)<<3", size=8, ext=0 + lfpimm xmml, 0 + br label("pslldq_end") + +pslldq_less_8: + + # Less than 8 + + msrli ufp1, xmml, "(8-IMMEDIATE) << 3", size=8, ext=0 + mslli xmmh, xmmh, "IMMEDIATE << 3", size=8, ext=0 + mslli xmml, xmml, "IMMEDIATE << 3", size=8, ext=0 + mor xmmh, xmmh, ufp1 + +pslldq_end: + fault "NoFault" +}; ''' -# PSLLDQ diff --git a/src/arch/x86/isa/insts/simd128/integer/shift/right_logical_shift.py b/src/arch/x86/isa/insts/simd128/integer/shift/right_logical_shift.py index c904eaf50..61efe1a5d 100644 --- a/src/arch/x86/isa/insts/simd128/integer/shift/right_logical_shift.py +++ b/src/arch/x86/isa/insts/simd128/integer/shift/right_logical_shift.py @@ -122,5 +122,41 @@ def macroop PSRLQ_XMM_I { msrli xmml, xmml, imm, size=8, ext=0 msrli xmmh, xmmh, imm, size=8, ext=0 }; + +def macroop PSRLDQ_XMM_I { + limm t2, 8 + subi t1, t2, imm, flags=(ECF,), dataSize=1 + br label("psrldq_less_8"), flags=(nCECF,) + # Greater than 8 + + limm t2, 16 + subi t1, t2, imm, flags=(ECF,), dataSize=1 + br label("psrldq_less_16"), flags=(nCECF,) + + # Greater than 16 + + lfpimm xmml, 0 + lfpimm xmmh, 0 + br label("psrldq_end") + +psrldq_less_16: + + # Between 8 and 16 + + msrli xmml, xmmh, "(IMMEDIATE-8)<<3", size=8, ext=0 + lfpimm xmmh, 0 + br label("psrldq_end") + +psrldq_less_8: + + # Less than 8 + + mslli ufp1, xmmh, "(8-IMMEDIATE) << 3", size=8, ext=0 + msrli xmml, xmml, "IMMEDIATE << 3", size=8, ext=0 + msrli xmmh, xmmh, "IMMEDIATE << 3", size=8, ext=0 + mor xmml, xmml, ufp1 + +psrldq_end: + fault "NoFault" +}; ''' -# PSRLDQ |