mirror of
https://github.com/FEX-Emu/FEX.git
synced 2026-10-06 08:00:15 +02:00
107 lines
2.6 KiB
JSON
107 lines
2.6 KiB
JSON
{
|
|
"Features": {
|
|
"Bitness": 32,
|
|
"EnabledHostFeatures": [
|
|
"TSO",
|
|
"LRCPC",
|
|
"LRCPC2",
|
|
"FLAGM",
|
|
"FLAGM2"
|
|
],
|
|
"DisabledHostFeatures": [
|
|
"SVE128",
|
|
"SVE256",
|
|
"RPRES",
|
|
"AFP"
|
|
],
|
|
"BinaryCacheVersion": 26
|
|
},
|
|
"Comment": [
|
|
"These are instruction combinations that could be more optimal if FEX optimized for them"
|
|
],
|
|
"Instructions": {
|
|
"Load variables from structs": {
|
|
"x86InstructionCount": 7,
|
|
"ExpectedInstructionCount": 16,
|
|
"Comment": [
|
|
"Saw this in 32-bit libvulkan_freedreno.so:tu_cs_begin_sub_stream_aligned",
|
|
"Loads a bunch of values from structs passed as arguments",
|
|
"Loads failed to use LRCPC2/ldapur with small immediate offset when possible"
|
|
],
|
|
"x86Insts": [
|
|
"mov edi, [ecx + 8]",
|
|
"mov edx, [ecx + 4]",
|
|
"mov ebx, [ecx]",
|
|
"mov esi, [ecx + 0xc]",
|
|
"imul edx, edi",
|
|
"mov eax, [ebx + 0xc]",
|
|
"sub eax, [ebx + 4]"
|
|
],
|
|
"ExpectedArm64ASM": [
|
|
"ldapur w11, [x7, #8]",
|
|
"nop",
|
|
"ldapur w5, [x7, #4]",
|
|
"nop",
|
|
"ldapur w6, [x7]",
|
|
"nop",
|
|
"ldapur w10, [x7, #12]",
|
|
"nop",
|
|
"mul w5, w5, w11",
|
|
"ldapur w4, [x6, #12]",
|
|
"nop",
|
|
"ldapur w20, [x6, #4]",
|
|
"nop",
|
|
"eor x27, x4, x20",
|
|
"subs w26, w4, w20",
|
|
"mov x4, x26"
|
|
]
|
|
},
|
|
"Load variables from memory": {
|
|
"x86InstructionCount": 4,
|
|
"ExpectedInstructionCount": 9,
|
|
"Comment": [
|
|
"Just to ensure small atomic offset loads are using LRCPC2"
|
|
],
|
|
"x86Insts": [
|
|
"mov edi, [ecx]",
|
|
"mov edx, [ecx + 4]",
|
|
"mov ax, [ecx + 24]",
|
|
"mov bl, [ecx + 26]"
|
|
],
|
|
"ExpectedArm64ASM": [
|
|
"ldapur w11, [x7]",
|
|
"nop",
|
|
"ldapur w5, [x7, #4]",
|
|
"nop",
|
|
"ldapurh w20, [x7, #24]",
|
|
"nop",
|
|
"bfxil x4, x20, #0, #16",
|
|
"ldapurb w20, [x7, #26]",
|
|
"bfxil x6, x20, #0, #8"
|
|
]
|
|
},
|
|
"Store variables to memory": {
|
|
"x86InstructionCount": 4,
|
|
"ExpectedInstructionCount": 7,
|
|
"Comment": [
|
|
"Just to ensure small atomic offset stores are using LRCPC2"
|
|
],
|
|
"x86Insts": [
|
|
"mov [ecx], edi",
|
|
"mov [ecx + 4], edx",
|
|
"mov [ecx + 24], ax",
|
|
"mov [ecx + 26], bl"
|
|
],
|
|
"ExpectedArm64ASM": [
|
|
"nop",
|
|
"stlur w11, [x7]",
|
|
"nop",
|
|
"stlur w5, [x7, #4]",
|
|
"nop",
|
|
"stlurh w4, [x7, #24]",
|
|
"stlurb w6, [x7, #26]"
|
|
]
|
|
}
|
|
}
|
|
}
|