diff options
| author | LDj3SNuD <35856442+LDj3SNuD@users.noreply.github.com> | 2021-01-04 23:45:54 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2021-01-04 23:45:54 +0100 |
| commit | 430ba6da65a781196db7d723cc88710bb7f5caf8 (patch) | |
| tree | a7ed55f638dde795f4270a324fa5338ffb80ee12 /Ryujinx.Tests/Cpu | |
| parent | a03ab0c4a0bef3c168874dc2105c43c9051e0807 (diff) | |
CPU (A64): Add Pmull_V Inst. with Clmul fast path for the "1/2D -> 1Q" variant & Sse fast path and slow path for both the "8/16B -> 8H" and "1/2D -> 1Q" variants; with Test. (#1817)
* Add Pmull_V Sse fast path only, both "8/16B -> 8H" and "1/2D -> 1Q" variants; with Test.
* Add Clmul fast path for the 128 bits variant.
* Small optimisation (save 60 instructions) for the Sse fast path about the 128 bits variant.
* Add slow path, both variants. Fix V128 Shl/Shr when shift = 0.
* A32: Add Vmull_I P64 variant (slow path); not tested.
* A32: Add Vmull_I_P8_P64 Test and fix P64 variant.
Diffstat (limited to 'Ryujinx.Tests/Cpu')
| -rw-r--r-- | Ryujinx.Tests/Cpu/CpuTestSimdReg.cs | 34 | ||||
| -rw-r--r-- | Ryujinx.Tests/Cpu/CpuTestSimdReg32.cs | 37 |
2 files changed, 71 insertions, 0 deletions
diff --git a/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs b/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs index 828c1bf9..0daeb1d1 100644 --- a/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs +++ b/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs @@ -60,6 +60,13 @@ namespace Ryujinx.Tests.Cpu 0x8080808080808080ul, 0xFFFFFFFFFFFFFFFFul }; } + private static ulong[] _8B1D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, + 0x8080808080808080ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + private static ulong[] _8B4H2S_() { return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, @@ -1977,6 +1984,33 @@ namespace Ryujinx.Tests.Cpu CompareAgainstUnicorn(); } + [Test, Pairwise, Description("PMULL{2} <Vd>.<Ta>, <Vn>.<Tb>, <Vm>.<Tb>")] + public void Pmull_V([Values(0u)] uint rd, + [Values(1u, 0u)] uint rn, + [Values(2u, 0u)] uint rm, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong z0, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong z1, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong a0, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong a1, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong b0, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong b1, + [Values(0b00u, 0b11u)] uint size, // Q0: <8B, 1D> => <8H, 1Q> + [Values(0b0u, 0b1u)] uint q) // Q1: <16B, 2D> => <8H, 1Q> + { + uint opcode = 0x0E20E000; // PMULL V0.8H, V0.8B, V0.8B + opcode |= ((rm & 31) << 16) | ((rn & 31) << 5) | ((rd & 31) << 0); + opcode |= ((size & 3) << 22); + opcode |= ((q & 1) << 30); + + V128 v0 = MakeVectorE0E1(z0, z1); + V128 v1 = MakeVectorE0E1(a0, a1); + V128 v2 = MakeVectorE0E1(b0, b1); + + SingleOpcode(opcode, v0: v0, v1: v1, v2: v2); + + CompareAgainstUnicorn(); + } + [Test, Pairwise, Description("RADDHN{2} <Vd>.<Tb>, <Vn>.<Ta>, <Vm>.<Ta>")] public void Raddhn_V_8H8B_4S4H_2D2S([Values(0u)] uint rd, [Values(1u, 0u)] uint rn, diff --git a/Ryujinx.Tests/Cpu/CpuTestSimdReg32.cs b/Ryujinx.Tests/Cpu/CpuTestSimdReg32.cs index e8298521..ed1c0f74 100644 --- a/Ryujinx.Tests/Cpu/CpuTestSimdReg32.cs +++ b/Ryujinx.Tests/Cpu/CpuTestSimdReg32.cs @@ -100,6 +100,13 @@ namespace Ryujinx.Tests.Cpu #endregion #region "ValueSource (Types)" + private static ulong[] _8B1D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, + 0x8080808080808080ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + private static ulong[] _8B4H2S1D_() { return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, @@ -530,6 +537,36 @@ namespace Ryujinx.Tests.Cpu CompareAgainstUnicorn(); } + [Test, Pairwise, Description("VMULL.<P8, P64> <Qd>, <Dn>, <Dm>")] + public void Vmull_I_P8_P64([Values(0u, 1u)] uint rd, + [Values(0u, 1u)] uint rn, + [Values(0u, 1u)] uint rm, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong d0, + [ValueSource(nameof(_8B1D_))] [Random(RndCnt)] ulong d1, + [Values(0u/*, 2u*/)] uint size) // <P8, P64> + { + /*if (size == 2u) + { + Assert.Ignore("Ryujinx.Tests.Unicorn.UnicornException : Invalid instruction (UC_ERR_INSN_INVALID)"); + }*/ + + uint opcode = 0xf2800e00u; // VMULL.P8 Q0, D0, D0 + + rd >>= 1; rd <<= 1; + + opcode |= (((rd & 0x10) << 18) | (rd & 0xf) << 12); + opcode |= (((rn & 0x10) << 3) | (rn & 0xf) << 16); + opcode |= (((rm & 0x10) << 1) | (rm & 0xf) << 0); + + opcode |= (size & 0x3) << 20; + + V128 v0 = MakeVectorE0E1(d0, d1); + + SingleOpcode(opcode, v0: v0); + + CompareAgainstUnicorn(); + } + [Test, Pairwise, Description("VSHL.<size> {<Vd>}, <Vm>, <Vn>")] public void Vshl([Values(0u)] uint rd, [Values(1u, 0u)] uint rn, |
