6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-02-23 05:29:23 +00:00
Code Issues Projects Releases Wiki Activity

Merge intrinsic instruction definitions for SSE and AVX versions of RCPPS and RSQRTPS.

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171339 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Craig Topper 2013-01-01 03:30:21 +00:00
parent 42ab0d75ab
commit 76f94fd361
1 changed files with 34 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

64
lib/Target/X86/X86InstrSSE.td
View File

@ -3007,7 +3007,32 @@ let Predicates = [HasAVX] in {
/// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
Intrinsic V4F32Int, OpndItins itins> {
Intrinsic V4F32Int, Intrinsic V8F32Int,
OpndItins itins> {
let Predicates = [HasAVX] in {
def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(!strconcat("v", OpcodeStr),
"ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int VR128:$src))],
itins.rr>, VEX;
def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
!strconcat(!strconcat("v", OpcodeStr),
"ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
itins.rm>, VEX;
def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
!strconcat(!strconcat("v", OpcodeStr),
"ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (V8F32Int VR256:$src))],
itins.rr>, VEX, VEX_L;
def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
(ins f256mem:$src),
!strconcat(!strconcat("v", OpcodeStr),
"ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
itins.rm>, VEX, VEX_L;
}
def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int VR128:$src))],
@ -3018,19 +3043,6 @@ multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
itins.rm>;
}
/// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
Intrinsic V4F32Int, OpndItins itins> {
def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (V4F32Int VR256:$src))],
itins.rr>, VEX_L;
def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
itins.rm>, VEX_L;
}
/// sse2_fp_unop_s - SSE2 unops in scalar form.
multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
@ -3103,8 +3115,12 @@ let Predicates = [HasAVX] in {
defm SQRT : sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
defm RSQRT : sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>;
defm RCP : sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>;
defm RSQRT : sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
defm RCP : sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
int_x86_avx_rcp_ps_256, SSE_RCPP>;
let Predicates = [HasAVX] in {
// Square root.
@ -3114,16 +3130,7 @@ let Predicates = [HasAVX] in {
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
defm VRSQRT : sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
SSE_SQRTP>,
sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
SSE_SQRTP>, VEX;
defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
defm VRCP : sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
SSE_RCPP>,
sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
SSE_RCPP>, VEX;
}
def : Pat<(f32 (fsqrt FR32:$src)),
@ -3215,17 +3222,14 @@ multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
SSE_SQRTS>,
sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
SSE_SQRTP>;
SSE_SQRTS>;
let Predicates = [UseSSE1] in {
def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
(RSQRTSSr_Int VR128:$src, VR128:$src)>;
}
defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
SSE_RCPS>,
sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPP>;
SSE_RCPS>;
let Predicates = [UseSSE1] in {
def : Pat<(int_x86_sse_rcp_ss VR128:$src),
(RCPSSr_Int VR128:$src, VR128:$src)>;