6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-07-23 16:30:00 +00:00
Code Issues Projects Releases Wiki Activity

Corrected Atom latencies for SSE SQRT instructions.

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181346 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Preston Gurd 2013-05-07 19:57:34 +00:00
parent f931f691ee
commit acccd2edc8
3 changed files with 36 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
lib/Target/X86/X86InstrSSE.td
View File

@ -3049,12 +3049,20 @@ let isCodeGenOnly = 1 in {
/// And, we have a special variant form for a full-vector intrinsic form.
let Sched = WriteFSqrt in {
def SSE_SQRTP : OpndItins<
IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
def SSE_SQRTPS : OpndItins<
IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
>;
def SSE_SQRTS : OpndItins<
IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
def SSE_SQRTSS : OpndItins<
IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
>;
def SSE_SQRTPD : OpndItins<
IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
>;
def SSE_SQRTSD : OpndItins<
IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
>;
}
@ -3319,18 +3327,18 @@ let Predicates = [HasAVX] in {
// Square root.
defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
SSE_SQRTS>,
sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
SSE_SQRTSS>,
sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
SSE_SQRTS>,
sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
SSE_SQRTSD>,
sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,

12
lib/Target/X86/X86Schedule.td
View File

@ -266,10 +266,14 @@ def IIC_SSE_PINSRW : InstrItinClass;
def IIC_SSE_PABS_RR : InstrItinClass;
def IIC_SSE_PABS_RM : InstrItinClass;
def IIC_SSE_SQRTP_RR : InstrItinClass;
def IIC_SSE_SQRTP_RM : InstrItinClass;
def IIC_SSE_SQRTS_RR : InstrItinClass;
def IIC_SSE_SQRTS_RM : InstrItinClass;
def IIC_SSE_SQRTPS_RR : InstrItinClass;
def IIC_SSE_SQRTPS_RM : InstrItinClass;
def IIC_SSE_SQRTSS_RR : InstrItinClass;
def IIC_SSE_SQRTSS_RM : InstrItinClass;
def IIC_SSE_SQRTPD_RR : InstrItinClass;
def IIC_SSE_SQRTPD_RM : InstrItinClass;
def IIC_SSE_SQRTSD_RR : InstrItinClass;
def IIC_SSE_SQRTSD_RM : InstrItinClass;
def IIC_SSE_RCPP_RR : InstrItinClass;
def IIC_SSE_RCPP_RM : InstrItinClass;

13
lib/Target/X86/X86ScheduleAtom.td
View File

@ -211,10 +211,15 @@ def AtomItineraries : ProcessorItineraries<
InstrItinData<IIC_SSE_UNPCK, [InstrStage<1, [Port0]>] >,
InstrItinData<IIC_SSE_SQRTP_RR, [InstrStage<13, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTP_RM, [InstrStage<14, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTS_RR, [InstrStage<11, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTS_RM, [InstrStage<12, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTPS_RR, [InstrStage<70, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTPS_RM, [InstrStage<70, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTSS_RR, [InstrStage<34, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTSS_RM, [InstrStage<34, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTPD_RR, [InstrStage<125, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTPD_RM, [InstrStage<125, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTSD_RR, [InstrStage<62, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_SQRTSD_RM, [InstrStage<62, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_RCPP_RR, [InstrStage<9, [Port0, Port1]>] >,
InstrItinData<IIC_SSE_RCPP_RM, [InstrStage<10, [Port0, Port1]>] >,