steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-14 08:00:50 +02:00
Code Issues Packages Projects Releases 67 Wiki Activity
9,555 Commits 11 Branches 71 Tags
e95c8d703ce2d103bb081cc1eebf50c46926d5c2
Commit Graph
1343 Commits
Author SHA1 Message Date
Ryan Houdek e95c8d703c Arm64: Implement support for emulated masked vector loadstores 
In order to support `vmaskmov{ps,pd}` without SVE128 this is required.
It's pretty gnarly but they aren't often used so that's fine from a
compatibility perspective.

Example SVE128 implementation:
```json
    "vmaskmovps ymm0, ymm1, [rax]": {
      "ExpectedInstructionCount": 9,
      "Comment": [
        "Map 2 0b01 0x2c 256-bit"
      ],
      "ExpectedArm64ASM": [
        "ldr q2, [x28, #32]",
        "mrs x20, nzcv",
        "cmplt p0.s, p6/z, z17.s, #0",
        "ld1w {z16.s}, p0/z, [x4]",
        "add x21, x4, #0x10 (16)",
        "cmplt p0.s, p6/z, z2.s, #0",
        "ld1w {z2.s}, p0/z, [x21]",
        "str q2, [x28, #16]",
        "msr nzcv, x20"
      ]
    },
```

Example ASIMD implementation
```json
    "vmaskmovps ymm0, ymm1, [rax]": {
      "ExpectedInstructionCount": 37,
      "Comment": [
        "Map 2 0b01 0x2c 256-bit"
      ],
      "ExpectedArm64ASM": [
        "ldr q2, [x28, #32]",
        "mrs x20, nzcv",
        "movi v0.2d, #0x0",
        "mov x1, x4",
        "mov x0, v17.d[0]",
        "tbz x0, #63, #+0x8",
        "ld1 {v0.s}[0], [x1]",
        "add x1, x1, #0x4 (4)",
        "tbz w0, #31, #+0x8",
        "ld1 {v0.s}[1], [x1]",
        "add x1, x1, #0x4 (4)",
        "mov x0, v17.d[1]",
        "tbz x0, #63, #+0x8",
        "ld1 {v0.s}[2], [x1]",
        "add x1, x1, #0x4 (4)",
        "tbz w0, #31, #+0x8",
        "ld1 {v0.s}[3], [x1]",
        "mov v16.16b, v0.16b",
        "add x21, x4, #0x10 (16)",
        "movi v0.2d, #0x0",
        "mov x1, x21",
        "mov x0, v2.d[0]",
        "tbz x0, #63, #+0x8",
        "ld1 {v0.s}[0], [x1]",
        "add x1, x1, #0x4 (4)",
        "tbz w0, #31, #+0x8",
        "ld1 {v0.s}[1], [x1]",
        "add x1, x1, #0x4 (4)",
        "mov x0, v2.d[1]",
        "tbz x0, #63, #+0x8",
        "ld1 {v0.s}[2], [x1]",
        "add x1, x1, #0x4 (4)",
        "tbz w0, #31, #+0x8",
        "ld1 {v0.s}[3], [x1]",
        "mov v2.16b, v0.16b",
        "str q2, [x28, #16]",
        "msr nzcv, x20"
      ]
    },
```

There's a little bit of an improvement where nzcv isn't needed to get
touched on the ASIMD implementation, but I'll leave that for a future
improvement.
 2024-06-21 08:21:32 -07:00
Ryan Houdek 3c293b9aed Arm64: Loosen restrictions on V{Load,Store}VectorMasked to allow 128-bit operation 2024-06-21 04:26:09 -07:00
Ryan Houdek f489135b1d Merge pull request #3734 from Sonicadvance1/avx_8 
AVX128: Move moves!
 2024-06-21 00:53:41 -07:00
Ryan Houdek 3f232e631e Merge pull request #3730 from Sonicadvance1/avx_4 
Vector: Helper refactorings
 2024-06-21 00:31:14 -07:00
Ryan Houdek 6e3643c3ef Merge pull request #3714 from pmatos/FSTstiTagSet 
Set tag properly in X87 FST(reg)
 2024-06-21 00:27:24 -07:00
Ryan Houdek c28824f94d AVX128: Implements support for vbroadcast* 2024-06-20 09:43:10 -07:00
Ryan Houdek 664d766b45 AVX128: Implement support for vmovshdup 2024-06-20 09:43:10 -07:00
Ryan Houdek fce694ed92 AVX128: Implement support for vmovsldup 2024-06-20 09:43:10 -07:00
Ryan Houdek 96aafb4f07 AVX128: Implement support for vmovddup 
This instruction is a little weird.
When accessing memory, the 128-bit operating size of the instruction
only loads 64-bits.
Meanwhile the 256-bit operating size of the instruction fetches a full
256-bits.

Theoretically the hardware could get away with two 64-bit loads or a
wacky 24-byte load, but it looks like to simplify hardware they just
spec'd it that the 256-bit version will always load the full range.
 2024-06-20 09:43:10 -07:00
Ryan Houdek dbaf95a8f3 AVX128: Implement support for vmovhps/d 2024-06-20 06:53:21 -07:00
Ryan Houdek e67df96ad9 AVX128: Implement support for movlps/d 2024-06-20 06:53:17 -07:00
Ryan Houdek 56de94578d AVX128: Implement support for vmovq 2024-06-20 06:53:13 -07:00
Ryan Houdek 06fc2f5ef0 AVX128: Implement support for non-temporal moves. 2024-06-20 06:53:09 -07:00
Ryan Houdek b3ba315cbd AVX128: Implements unary/binary lambda helper 2024-06-20 06:53:05 -07:00
Ryan Houdek e5a531e683 Vector: Refactor MPSADBWOpImpl so AVX128 can use it. 2024-06-20 06:43:57 -07:00
Ryan Houdek e2de57bd04 Vector: Refactor PSADBWOpImpl so AVX128 can use it. 2024-06-20 06:43:57 -07:00
Ryan Houdek 4eebca93e3 Vector: Refactor PSHUFBOpImpl. This will be reused for AVX128 2024-06-20 06:33:27 -07:00
Ryan Houdek 3919ec9692 Vector: Expose VBLENDOpImpl in the OpcodeDispatcher. It will be reused by AVX128 2024-06-20 06:33:21 -07:00
Ryan Houdek 02aeb0ac1a Vector: Restructure PMADDWDOpImpl. It's going to get reused for AVX128 2024-06-20 06:33:15 -07:00
Ryan Houdek 206544ad09 Vector: Reconfigure PMADDUBSWOpImpl, it's going to get reused for AVX128 2024-06-20 06:33:08 -07:00
Ryan Houdek 3854cd2b2f Vector: Restruture SHUFOpImpl. AVX128 is going to reuse it. 2024-06-20 06:32:58 -07:00
Ryan Houdek acbd920c9a OpcodeDispatcher: Adds initial groundwork for decomposed AVX operations 
Only installs the tables if SVE256 isn't supported yet AVX is explicitly
enabled with HostFeatures, to protect accidental enablement early.

- Only implements 85 instructions starting out
- Basic vector moves
- Basic vector unary operations
- Basic vector binary operations
- VZeroUpper/VZeroAll

The bulk of the implementation is currently the handling for loading and
storing the halves of the registers from the context or from memory.

This means the load/store helpers must always return a pair unless only
requesting the bottom half of the register, which occurs with 128-bit
AVX operations. The store side then needing to consume the named zero
register if it occurs since those cases will zero the upper bits.

This implementation approach has a few benefits.
- I can pound this out extremely quickly
- SSE implementations are unaffected and don't need to deal with the
  insert behaviour of SVE256.
- We still keep the SVE256 implementation for the inevitable future when
  hardware vendors actually do implement it (Give it 8 years or
  something).
- We can actually unit test this path in CI once it is complete.
- We can partially optimize some paths with SVE128 (Gathers) and support
  a full ASIMD path if necessary.

One downside is that I can't enable this in CI yet because it can't pass
all unittests. but that's a non-issue since it is going to be in heavy
flux as I'm hammering out the implementation. It'll get switched on at
the end when it's passing all 1265 AVX unittests. Currently at 1001 on
this.
 2024-06-20 08:44:14 -04:00
Alyssa Rosenzweig db0bdd48e5 Merge pull request #3729 from alyssarosenzweig/refactor/address-modes 
OpcodeDispatcher: Refactor address modes
 2024-06-20 08:18:33 -04:00
Ryan Houdek da21ee3cda Merge pull request #3692 from pmatos/AFP_RPRES_fix 
Fixes AFP.NEP handling on scalar insertions
 2024-06-19 19:23:49 -07:00
Ryan Houdek d2baef2b36 Merge pull request #3727 from Sonicadvance1/vaes 
VAES support
 2024-06-19 19:22:56 -07:00
Ryan Houdek df96bc83cc Merge pull request #3726 from Sonicadvance1/oryon_errata 
HostFeatures: Work around Qualcomm Oryon RNG errata
 2024-06-19 19:21:14 -07:00
Alyssa Rosenzweig ec03831a21 OpcodeDispatcher: plumb A.NonTSO deeper 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-19 08:52:07 -04:00
Alyssa Rosenzweig 9ca821316a OpcodeDispatcher: factor out DecodeAddress 
this is the common guts of the load/store routines.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-19 08:52:07 -04:00
Alyssa Rosenzweig 025a060337 OpcodeDispatcher: extract IsNonTSOReg 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-19 08:52:07 -04:00
Alyssa Rosenzweig 371d6f0730 OpcodeDispatcher: extract IsOperandMem 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-19 08:52:07 -04:00
Ryan Houdek 643bc10d52 CPUID: Expose VAES if supported 2024-06-19 05:51:47 -07:00
Ryan Houdek 542ed8b6ad Implement support for querying AES256 support 
This is a different feature flag than regular AES as the default AES+AVX
only operates on 128-bit wide vectors.

With the newer `VAES` extension this is expanded to 256-bit.
 2024-06-19 05:51:47 -07:00
Paulo Matos 2483329ef6 Fixes AFP.NEP handling on scalar insertions 
Fixes #3690

When doing scalar insertions, upper bits come from different arguments
depending on the operation. These are listed in the ARM spec under the
NEP bit documentation.
 2024-06-19 10:02:54 +02:00
Paulo Matos 359221b379 Set tag properly in X87 FST(reg) 2024-06-19 10:02:05 +02:00
Paulo Matos f9b38a1de7 FXCH should set C1 to zero 2024-06-19 08:57:48 +02:00
Ryan Houdek 67e1ac0442 Merge pull request #3725 from alyssarosenzweig/ir/vbic 
IR: rename _VBic -> _VAndn
 2024-06-18 16:34:26 -07:00
Ryan Houdek c57e9e008f Merge pull request #3723 from alyssarosenzweig/fexcore/zero-helper 
OpcodeDispatcher: refactor zero vector loads
 2024-06-18 16:34:15 -07:00
Ryan Houdek b34c23fe3d HostFeatures: Work around Qualcomm Oryon RNG errata 
The Oryon is the first CPU we know of that implemented support for the
RNG extension. It also has an errata where reading the RNDRRS register
never returns success. X86's RDSEED guarantees forward progress with
enough retries.

When an x86 processor messed this up at one point, some Linux systems
would infinite loop (presumably when something in boot was filling an
entropy pool). This required a microcode change to fix that processor.

The rdseed unittest infinite loops on this platform if RNG was exposed.
 2024-06-18 16:29:53 -07:00
Alyssa Rosenzweig 01da5972fc IR: rename _VBic -> _VAndn 
to be consistent with the scalar _Andn opcode, which is specifically named _Andn
and not _Bic.

noticed while reviewing AVX patches

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-18 14:00:01 -04:00
Ryan Houdek bf812aae8f CoreState: Adds avx_high structure for tracking decoupled AVX halves. 
Needed something inbetween the `InlineJITBlockHeader` and `avx_high` in
order to match alignment requirements of 16-byte for avx_high. Chose the
`DeferredSignalRefCount` because we hit it quite frequently and it is
basically the only 64-bit variable that we end up touching
significantly.

In the future the CPUState object is going to need to change its view of
the object depending on if the device supports SVE256 or not, but we
don't need to frontload the work right now. It'll become significantly
easier to support that path once the RCLSE pass gets deleted.
 2024-06-18 12:00:45 -04:00
Ryan Houdek ee165249bc Dispatcher: Fix ARM64EC 
We don't have CI for this and was missed.
 2024-06-18 12:00:45 -04:00
Alyssa Rosenzweig af8cfb79e5 OpcodeDispatcher: refactor zero vector loads 
AVX128 is going to slam this, so make it more ergonomic.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-18 11:44:46 -04:00
Ryan Houdek 1ce27a5e6b FEXCore: Disentangle the SVE256 feature from AVX 
In quite a few locations we are mixing the case that SVE256 == AVX or
that AVX means the guest register size is 256-bit.

While this is true today, this is entanglement is going to change very
quickly and cause confusion in follow-up PRs.

Now we have SVE128, SVE256, and SVE2 HostFeatures to disambiguate the
different features which mean different things.

This PR keeps the alias that `SupportsAVX` = `SupportsSVE256 && SupportsSVE2`
but that alias is going to very quickly change its definition.
 2024-06-17 17:20:32 -07:00
Ryan Houdek 933d622860 Merge pull request #3710 from Sonicadvance1/avx128_1 
CoreState: Move `InlineJITBlockHeader` to the start of the struct
 2024-06-17 17:17:56 -07:00
Alyssa Rosenzweig 799c17eb90 Arm64Emitter: drop out of date comment 
I fixed this when we landed the new RA

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-17 14:58:08 -04:00
Alyssa Rosenzweig 4965344ef5 Merge pull request #3705 from alyssarosenzweig/pre-rclse 
Clean ups from my RCLSE branch
 2024-06-17 14:22:01 -04:00
Alyssa Rosenzweig 46ca53ad0d Merge pull request #3704 from alyssarosenzweig/ra/spill-better 
RA: priorize remat over spilling
 2024-06-17 09:01:50 -04:00
Alyssa Rosenzweig 61ff1b3584 Merge pull request #3712 from alyssarosenzweig/jit/silly-assert 
JIT: delete silly assert
 2024-06-17 08:59:00 -04:00
Alyssa Rosenzweig 7c0c5de4bd JIT: delete silly assert 
noticed in the area.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-06-17 08:51:22 -04:00
Ryan Houdek a9bacc1b6b CoreState: Move InlineJITBlockHeader to the start of the struct 
This currently doesn't do much but soon this will be very important to
ensure the data prefetcher of Cortex keeps the cachelines following this
variable in L1.
 2024-06-17 02:59:56 -07:00