RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 11:42:57 +01:00
Code Issues Projects Releases Wiki Activity

Remove alignment restrictions from FMA load folding.

Browse Source 
llvm-svn: 191136
This commit is contained in:
Craig Topper 2013-09-21 05:58:59 +00:00
parent 4f8b0cf48b
commit ef2cf025cd
1 changed files with 31 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
lib/Target/X86/X86InstrFMA.td
View File

@ -74,43 +74,43 @@ let neverHasSideEffects = 1 in {
// Fused Multiply-Add
let ExeDomain = SSEPackedSingle in {
defm VFMADDPS : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "ps", memopv4f32,
memopv8f32, X86Fmadd, v4f32, v8f32>;
defm VFMSUBPS : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "ps", memopv4f32,
memopv8f32, X86Fmsub, v4f32, v8f32>;
defm VFMADDPS : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "ps", loadv4f32,
loadv8f32, X86Fmadd, v4f32, v8f32>;
defm VFMSUBPS : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "ps", loadv4f32,
loadv8f32, X86Fmsub, v4f32, v8f32>;
defm VFMADDSUBPS : fma3p_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "ps",
memopv4f32, memopv8f32, X86Fmaddsub,
loadv4f32, loadv8f32, X86Fmaddsub,
v4f32, v8f32>;
defm VFMSUBADDPS : fma3p_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "ps",
memopv4f32, memopv8f32, X86Fmsubadd,
loadv4f32, loadv8f32, X86Fmsubadd,
v4f32, v8f32>;
}
let ExeDomain = SSEPackedDouble in {
defm VFMADDPD : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "pd", memopv2f64,
memopv4f64, X86Fmadd, v2f64, v4f64>, VEX_W;
defm VFMSUBPD : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "pd", memopv2f64,
memopv4f64, X86Fmsub, v2f64, v4f64>, VEX_W;
defm VFMADDPD : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "pd", loadv2f64,
loadv4f64, X86Fmadd, v2f64, v4f64>, VEX_W;
defm VFMSUBPD : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "pd", loadv2f64,
loadv4f64, X86Fmsub, v2f64, v4f64>, VEX_W;
defm VFMADDSUBPD : fma3p_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "pd",
memopv2f64, memopv4f64, X86Fmaddsub,
loadv2f64, loadv4f64, X86Fmaddsub,
v2f64, v4f64>, VEX_W;
defm VFMSUBADDPD : fma3p_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "pd",
memopv2f64, memopv4f64, X86Fmsubadd,
loadv2f64, loadv4f64, X86Fmsubadd,
v2f64, v4f64>, VEX_W;
}
// Fused Negative Multiply-Add
let ExeDomain = SSEPackedSingle in {
defm VFNMADDPS : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "ps", memopv4f32,
memopv8f32, X86Fnmadd, v4f32, v8f32>;
defm VFNMSUBPS : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "ps", memopv4f32,
memopv8f32, X86Fnmsub, v4f32, v8f32>;
defm VFNMADDPS : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "ps", loadv4f32,
loadv8f32, X86Fnmadd, v4f32, v8f32>;
defm VFNMSUBPS : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "ps", loadv4f32,
loadv8f32, X86Fnmsub, v4f32, v8f32>;
}
let ExeDomain = SSEPackedDouble in {
defm VFNMADDPD : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "pd", memopv2f64,
memopv4f64, X86Fnmadd, v2f64, v4f64>, VEX_W;
defm VFNMADDPD : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "pd", loadv2f64,
loadv4f64, X86Fnmadd, v2f64, v4f64>, VEX_W;
defm VFNMSUBPD : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "pd",
memopv2f64, memopv4f64, X86Fnmsub, v2f64,
loadv2f64, loadv4f64, X86Fnmsub, v2f64,
v4f64>, VEX_W;
}
@ -338,31 +338,31 @@ defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
let ExeDomain = SSEPackedSingle in {
defm VFMADDPS4 : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
memopv4f32, memopv8f32>;
loadv4f32, loadv8f32>;
defm VFMSUBPS4 : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
memopv4f32, memopv8f32>;
loadv4f32, loadv8f32>;
defm VFNMADDPS4 : fma4p<0x78, "vfnmaddps", X86Fnmadd, v4f32, v8f32,
memopv4f32, memopv8f32>;
loadv4f32, loadv8f32>;
defm VFNMSUBPS4 : fma4p<0x7C, "vfnmsubps", X86Fnmsub, v4f32, v8f32,
memopv4f32, memopv8f32>;
loadv4f32, loadv8f32>;
defm VFMADDSUBPS4 : fma4p<0x5C, "vfmaddsubps", X86Fmaddsub, v4f32, v8f32,
memopv4f32, memopv8f32>;
loadv4f32, loadv8f32>;
defm VFMSUBADDPS4 : fma4p<0x5E, "vfmsubaddps", X86Fmsubadd, v4f32, v8f32,
memopv4f32, memopv8f32>;
loadv4f32, loadv8f32>;
}
let ExeDomain = SSEPackedDouble in {
defm VFMADDPD4 : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
memopv2f64, memopv4f64>;
loadv2f64, loadv4f64>;
defm VFMSUBPD4 : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,
memopv2f64, memopv4f64>;
loadv2f64, loadv4f64>;
defm VFNMADDPD4 : fma4p<0x79, "vfnmaddpd", X86Fnmadd, v2f64, v4f64,
memopv2f64, memopv4f64>;
loadv2f64, loadv4f64>;
defm VFNMSUBPD4 : fma4p<0x7D, "vfnmsubpd", X86Fnmsub, v2f64, v4f64,
memopv2f64, memopv4f64>;
loadv2f64, loadv4f64>;
defm VFMADDSUBPD4 : fma4p<0x5D, "vfmaddsubpd", X86Fmaddsub, v2f64, v4f64,
memopv2f64, memopv4f64>;
loadv2f64, loadv4f64>;
defm VFMSUBADDPD4 : fma4p<0x5F, "vfmsubaddpd", X86Fmsubadd, v2f64, v4f64,
memopv2f64, memopv4f64>;
loadv2f64, loadv4f64>;
}