ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

GS: Remove old SetupPrim code generators

This commit is contained in:
TellowKrinkle 2021-08-28 00:02:24 -05:00 committed by refractionpcsx2
parent fd0351ca8f
commit 805b647c73
10 changed files with 0 additions and 2202 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
pcsx2/CMakeLists.txt
View File

@ -650,12 +650,6 @@ set(pcsx2GSSources
GS/Renderers/SW/GSRendererSW.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.all.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.x64.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.x64.avx.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.x64.avx2.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.x86.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.x86.avx.cpp
GS/Renderers/SW/GSSetupPrimCodeGenerator.x86.avx2.cpp
GS/Renderers/SW/GSTextureCacheSW.cpp
GS/Renderers/SW/GSTextureSW.cpp
GS/Renderers/OpenGL/GLLoader.cpp

17
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.h
View File

@ -32,23 +32,6 @@ class GSSetupPrimCodeGenerator : public GSCodeGenerator
uint32 z : 1, f : 1, t : 1, c : 1;
} m_en;
#if _M_SSE < 0x501
void Generate_SSE();
void Depth_SSE();
void Texture_SSE();
void Color_SSE();
void Generate_AVX();
void Depth_AVX();
void Texture_AVX();
void Color_AVX();
#else
void Generate_AVX2();
void Depth_AVX2();
void Texture_AVX2();
void Color_AVX2();
#endif
public:
GSSetupPrimCodeGenerator(void* param, uint64 key, void* code, size_t maxsize);
};

365
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.x64.avx.cpp
View File

@ -1,365 +0,0 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2021 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "PrecompiledHeader.h"
#include "GSSetupPrimCodeGenerator.h"
#include "GSVertexSW.h"
#include "GS/GS_codegen.h"
#if _M_SSE < 0x501 && (defined(_M_AMD64) || defined(_WIN64))
#define _rip_local(field) (m_rip ? ptr[rip + &m_local.field] : ptr[t0 + offsetof(GSScanlineLocalData, field)])
#define _rip_local_v(field, offset) (m_rip ? ptr[rip + &m_local.field] : ptr[t0 + offset])
void GSSetupPrimCodeGenerator::Generate_AVX()
{
// Technically we just need the delta < 2GB
m_rip = (size_t)&m_local < 0x80000000 && (size_t)getCurr() < 0x80000000;
#ifdef _WIN64
sub(rsp, 8 + 2 * 16);
vmovdqa(ptr[rsp + 0], xmm6);
vmovdqa(ptr[rsp + 16], xmm7);
#endif
if (!m_rip)
mov(t0, (size_t)&m_local);
if ((m_en.z || m_en.f) && m_sel.prim != GS_SPRITE_CLASS || m_en.t || m_en.c && m_sel.iip)
{
mov(rax, (size_t)g_const->m_shift_128b);
for (int i = 0; i < (m_sel.notest ? 2 : 5); i++)
{
vmovaps(Xmm(3 + i), ptr[rax + i * 16]);
}
}
Depth_AVX();
Texture_AVX();
Color_AVX();
#ifdef _WIN64
vmovdqa(xmm6, ptr[rsp + 0]);
vmovdqa(xmm7, ptr[rsp + 16]);
add(rsp, 8 + 2 * 16);
#endif
ret();
}
void GSSetupPrimCodeGenerator::Depth_AVX()
{
if (!m_en.z && !m_en.f)
{
return;
}
if (m_sel.prim != GS_SPRITE_CLASS)
{
// GSVector4 p = dscan.p;
vmovaps(xmm0, ptr[a2 + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// GSVector4 df = p.wwww();
vshufps(xmm1, xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
// m_local.d4.f = GSVector4i(df * 4.0f).xxzzlh();
vmulps(xmm2, xmm1, xmm3);
vcvttps2dq(xmm2, xmm2);
vpshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vmovdqa(_rip_local(d4.f), xmm2);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].f = GSVector4i(df * m_shift[i]).xxzzlh();
vmulps(xmm2, xmm1, Xmm(4 + i));
vcvttps2dq(xmm2, xmm2);
vpshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].f) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].f, variableOffset), xmm2);
}
}
if (m_en.z)
{
// GSVector4 dz = p.zzzz();
vshufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
// m_local.d4.z = dz * 4.0f;
vmulps(xmm1, xmm0, xmm3);
vmovdqa(_rip_local(d4.z), xmm1);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].z = dz * m_shift[i];
vmulps(xmm1, xmm0, Xmm(4 + i));
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].z) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].z, variableOffset), xmm1);
}
}
}
else
{
// GSVector4 p = vertex[index[1]].p;
mov(eax, ptr[a1 + sizeof(uint32) * 1]);
shl(eax, 6); // * sizeof(GSVertexSW)
add(rax, a0);
if (m_en.f)
{
// m_local.p.f = GSVector4i(p).zzzzh().zzzz();
vmovaps(xmm0, ptr[rax + offsetof(GSVertexSW, p)]);
vcvttps2dq(xmm1, xmm0);
vpshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
vpshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
vmovdqa(_rip_local(p.f), xmm1);
}
if (m_en.z)
{
// uint32 z is bypassed in t.w
vmovdqa(xmm0, ptr[rax + offsetof(GSVertexSW, t)]);
vpshufd(xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
vmovdqa(_rip_local(p.z), xmm0);
}
}
}
void GSSetupPrimCodeGenerator::Texture_AVX()
{
if (!m_en.t)
{
return;
}
// GSVector4 t = dscan.t;
vmovaps(xmm0, ptr[a2 + offsetof(GSVertexSW, t)]);
vmulps(xmm1, xmm0, xmm3);
if (m_sel.fst)
{
// m_local.d4.stq = GSVector4i(t * 4.0f);
vcvttps2dq(xmm1, xmm1);
vmovdqa(_rip_local(d4.stq), xmm1);
}
else
{
// m_local.d4.stq = t * 4.0f;
vmovaps(_rip_local(d4.stq), xmm1);
}
for (int j = 0, k = m_sel.fst ? 2 : 3; j < k; j++)
{
// GSVector4 ds = t.xxxx();
// GSVector4 dt = t.yyyy();
// GSVector4 dq = t.zzzz();
vshufps(xmm1, xmm0, xmm0, (uint8)_MM_SHUFFLE(j, j, j, j));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4 v = ds/dt * m_shift[i];
vmulps(xmm2, xmm1, Xmm(4 + i));
if (m_sel.fst)
{
// m_local.d[i].s/t = GSVector4i(v);
vcvttps2dq(xmm2, xmm2);
const size_t variableOffsetS = offsetof(GSScanlineLocalData, d[0].s) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetT = offsetof(GSScanlineLocalData, d[0].t) + (i * sizeof(GSScanlineLocalData::d[0]));
switch (j)
{
case 0: vmovdqa(_rip_local_v(d[i].s, variableOffsetS), xmm2); break;
case 1: vmovdqa(_rip_local_v(d[i].t, variableOffsetT), xmm2); break;
}
}
else
{
// m_local.d[i].s/t/q = v;
const size_t variableOffsetS = offsetof(GSScanlineLocalData, d[0].s) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetT = offsetof(GSScanlineLocalData, d[0].t) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetQ = offsetof(GSScanlineLocalData, d[0].q) + (i * sizeof(GSScanlineLocalData::d[0]));
switch (j)
{
case 0: vmovaps(_rip_local_v(d[i].s, variableOffsetS), xmm2); break;
case 1: vmovaps(_rip_local_v(d[i].t, variableOffsetT), xmm2); break;
case 2: vmovaps(_rip_local_v(d[i].q, variableOffsetQ), xmm2); break;
}
}
}
}
}
void GSSetupPrimCodeGenerator::Color_AVX()
{
if (!m_en.c)
{
return;
}
if (m_sel.iip)
{
// GSVector4 c = dscan.c;
vmovaps(xmm0, ptr[a2 + offsetof(GSVertexSW, c)]);
// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
vmulps(xmm1, xmm0, xmm3);
vcvttps2dq(xmm1, xmm1);
vpshufd(xmm1, xmm1, _MM_SHUFFLE(3, 1, 2, 0));
vpackssdw(xmm1, xmm1);
vmovdqa(_rip_local(d4.c), xmm1);
// xmm3 is not needed anymore
// GSVector4 dr = c.xxxx();
// GSVector4 db = c.zzzz();
vshufps(xmm2, xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vshufps(xmm3, xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i r = GSVector4i(dr * m_shift[i]).ps32();
vmulps(xmm0, xmm2, Xmm(4 + i));
vcvttps2dq(xmm0, xmm0);
vpackssdw(xmm0, xmm0);
// GSVector4i b = GSVector4i(db * m_shift[i]).ps32();
vmulps(xmm1, xmm3, Xmm(4 + i));
vcvttps2dq(xmm1, xmm1);
vpackssdw(xmm1, xmm1);
// m_local.d[i].rb = r.upl16(b);
vpunpcklwd(xmm0, xmm1);
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].rb) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].rb, variableOffset), xmm0);
}
// GSVector4 c = dscan.c;
vmovaps(xmm0, ptr[a2 + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
// GSVector4 dg = c.yyyy();
// GSVector4 da = c.wwww();
vshufps(xmm2, xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
vshufps(xmm3, xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i g = GSVector4i(dg * m_shift[i]).ps32();
vmulps(xmm0, xmm2, Xmm(4 + i));
vcvttps2dq(xmm0, xmm0);
vpackssdw(xmm0, xmm0);
// GSVector4i a = GSVector4i(da * m_shift[i]).ps32();
vmulps(xmm1, xmm3, Xmm(4 + i));
vcvttps2dq(xmm1, xmm1);
vpackssdw(xmm1, xmm1);
// m_local.d[i].ga = g.upl16(a);
vpunpcklwd(xmm0, xmm1);
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].ga) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].ga, variableOffset), xmm0);
}
}
else
{
// GSVector4i c = GSVector4i(vertex[index[last].c);
int last = 0;
switch (m_sel.prim)
{
case GS_POINT_CLASS: last = 0; break;
case GS_LINE_CLASS: last = 1; break;
case GS_TRIANGLE_CLASS: last = 2; break;
case GS_SPRITE_CLASS: last = 1; break;
}
if (!(m_sel.prim == GS_SPRITE_CLASS && (m_en.z || m_en.f))) // if this is a sprite, the last vertex was already loaded in Depth()
{
mov(eax, ptr[a1 + sizeof(uint32) * last]);
shl(eax, 6); // * sizeof(GSVertexSW)
add(rax, a0);
}
vcvttps2dq(xmm0, ptr[rax + offsetof(GSVertexSW, c)]);
// c = c.upl16(c.zwxy());
vpshufd(xmm1, xmm0, _MM_SHUFFLE(1, 0, 3, 2));
vpunpcklwd(xmm0, xmm1);
// if(!tme) c = c.srl16(7);
if (m_sel.tfx == TFX_NONE)
{
vpsrlw(xmm0, 7);
}
// m_local.c.rb = c.xxxx();
// m_local.c.ga = c.zzzz();
vpshufd(xmm1, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vpshufd(xmm2, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
vmovdqa(_rip_local(c.rb), xmm1);
vmovdqa(_rip_local(c.ga), xmm2);
}
}
#endif

368
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.x64.avx2.cpp
View File

@ -1,368 +0,0 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2021 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "PrecompiledHeader.h"
#include "GSSetupPrimCodeGenerator.h"
#include "GSVertexSW.h"
#include "GS/GS_codegen.h"
#if _M_SSE >= 0x501 && (defined(_M_AMD64) || defined(_WIN64))
#define _rip_local(field) (m_rip ? ptr[rip + &m_local.field] : ptr[t0 + offsetof(GSScanlineLocalData, field)])
#define _rip_local_v(field, offset) (m_rip ? ptr[rip + &m_local.field] : ptr[t0 + offset])
#define _m_shift(i) (Ymm(7 + i))
// FIXME windows ?
#define _vertex rcx
void GSSetupPrimCodeGenerator::Generate_AVX2()
{
// Technically we just need the delta < 2GB
m_rip = (size_t)&m_local < 0x80000000 && (size_t)getCurr() < 0x80000000;
#ifdef _WIN64
sub(rsp, 8 + 2 * 16);
vmovdqa(ptr[rsp + 0], ymm6);
vmovdqa(ptr[rsp + 16], ymm7);
#endif
if (!m_rip)
mov(t0, (size_t)&m_local);
if ((m_en.z || m_en.f) && m_sel.prim != GS_SPRITE_CLASS || m_en.t || m_en.c && m_sel.iip)
{
mov(rax, (size_t)g_const->m_shift_256b);
for (int i = 0; i < (m_sel.notest ? 2 : 9); i++)
{
vmovaps(_m_shift(i), ptr[rax + i * 32]);
}
}
// ymm7 to ymm 15 = m_shift[i]
Depth_AVX2();
Texture_AVX2();
Color_AVX2();
#ifdef _WIN64
vmovdqa(ymm6, ptr[rsp + 0]);
vmovdqa(ymm7, ptr[rsp + 16]);
add(rsp, 8 + 2 * 16);
#endif
ret();
}
void GSSetupPrimCodeGenerator::Depth_AVX2()
{
if (!m_en.z && !m_en.f)
{
return;
}
if (m_sel.prim != GS_SPRITE_CLASS)
{
const Ymm& dscan_p = ymm6;
// GSVector4 dp8 = dscan.p * GSVector4::broadcast32(&shift[0]);
vbroadcastf128(dscan_p, ptr[a2 + offsetof(GSVertexSW, p)]);
vmulps(ymm1, dscan_p, _m_shift(0));
if (m_en.z)
{
// m_local.d8.p.z = dp8.extract32<2>();
vextractps(_rip_local(d8.p.z), xmm1, 2);
// GSVector8 dz = GSVector8(dscan.p).zzzz();
vshufps(ymm2, dscan_p, dscan_p, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// m_local.d[i].z = dz * shift[1 + i];
vmulps(ymm0, ymm2, _m_shift(1 + i));
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].z) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovaps(_rip_local_v(d[i].z, variableOffset), ymm0);
}
}
if (m_en.f)
{
// m_local.d8.p.f = GSVector4i(dp8).extract32<3>();
// FIXME no truncate ? why ? vcvttps2dq ?
//vcvtps2dq(ymm2, ymm1); // let's guess a typo
vcvttps2dq(ymm2, ymm1);
vpextrd(_rip_local(d8.p.f), xmm2, 3);
// GSVector8 df = GSVector8(dscan.p).wwww();
vshufps(ymm3, dscan_p, dscan_p, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// m_local.d[i].f = GSVector8i(df * m_shift[i]).xxzzlh();
vmulps(ymm0, ymm3, _m_shift(1 + i));
vcvttps2dq(ymm0, ymm0);
vpshuflw(ymm0, ymm0, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(ymm0, ymm0, _MM_SHUFFLE(2, 2, 0, 0));
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].f) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].f, variableOffset), ymm0);
}
}
}
else
{
// GSVector4 p = vertex[index[1]].p;
mov(_vertex.cvt32(), ptr[a1 + sizeof(uint32) * 1]);
shl(_vertex.cvt32(), 6); // * sizeof(GSVertexSW)
add(_vertex, a0);
if (m_en.f)
{
// m_local.p.f = GSVector4i(vertex[index[1]].p).extract32<3>();
vmovaps(xmm0, ptr[_vertex + offsetof(GSVertexSW, p)]);
vcvttps2dq(xmm0, xmm0);
vpextrd(_rip_local(p.f), xmm0, 3);
}
if (m_en.z)
{
// m_local.p.z = vertex[index[1]].t.u32[3]; // uint32 z is bypassed in t.w
mov(eax, ptr[ecx + offsetof(GSVertexSW, t.w)]);
mov(_rip_local(p.z), eax);
}
}
}
void GSSetupPrimCodeGenerator::Texture_AVX2()
{
if (!m_en.t)
{
return;
}
// GSVector8 dt(dscan.t);
vbroadcastf128(ymm0, ptr[a2 + offsetof(GSVertexSW, t)]);
// GSVector8 dt8 = dt * shift[0];
vmulps(ymm1, ymm0, _m_shift(0));
if (m_sel.fst)
{
// m_local.84.stq = GSVector4i(t * 4.0f);
vcvttps2dq(ymm1, ymm1);
vmovdqa(_rip_local(d8.stq), xmm1);
}
else
{
// m_local.d8.stq = t * 4.0f;
vmovaps(_rip_local(d8.stq), xmm1);
}
for (int j = 0, k = m_sel.fst ? 2 : 3; j < k; j++)
{
// GSVector8 dstq = dt.xxxx/yyyy/zzzz();
vshufps(ymm1, ymm0, ymm0, (uint8)_MM_SHUFFLE(j, j, j, j));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// GSVector8 v = dstq * shift[1 + i];
vmulps(ymm2, ymm1, _m_shift(1 + i));
if (m_sel.fst)
{
// m_local.d[i].s/t = GSVector8::cast(GSVector8i(v));
vcvttps2dq(ymm2, ymm2);
const size_t variableOffsetS = offsetof(GSScanlineLocalData, d[0].s) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetT = offsetof(GSScanlineLocalData, d[0].t) + (i * sizeof(GSScanlineLocalData::d[0]));
switch (j)
{
case 0: vmovdqa(_rip_local_v(d[i].s, variableOffsetS), ymm2); break;
case 1: vmovdqa(_rip_local_v(d[i].t, variableOffsetT), ymm2); break;
}
}
else
{
// m_local.d[i].s/t/q = v;
const size_t variableOffsetS = offsetof(GSScanlineLocalData, d[0].s) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetT = offsetof(GSScanlineLocalData, d[0].t) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetQ = offsetof(GSScanlineLocalData, d[0].q) + (i * sizeof(GSScanlineLocalData::d[0]));
switch (j)
{
case 0: vmovaps(_rip_local_v(d[i].s, variableOffsetS), ymm2); break;
case 1: vmovaps(_rip_local_v(d[i].t, variableOffsetT), ymm2); break;
case 2: vmovaps(_rip_local_v(d[i].q, variableOffsetQ), ymm2); break;
}
}
}
}
}
void GSSetupPrimCodeGenerator::Color_AVX2()
{
if (!m_en.c)
{
return;
}
if (m_sel.iip)
{
const Ymm& dscan_c = ymm6;
// GSVector8 dc(dscan.c);
vbroadcastf128(dscan_c, ptr[a2 + offsetof(GSVertexSW, c)]);
// m_local.d8.c = GSVector4i(c * 4.0f).xzyw().ps32();
vmulps(ymm1, dscan_c, ymm3);
vcvttps2dq(ymm1, ymm1);
vpshufd(ymm1, ymm1, _MM_SHUFFLE(3, 1, 2, 0));
vpackssdw(ymm1, ymm1);
vmovq(_rip_local(d8.c), xmm1);
// GSVector8 dr = dc.xxxx();
// GSVector8 db = dc.zzzz();
vshufps(ymm2, dscan_c, dscan_c, _MM_SHUFFLE(0, 0, 0, 0));
vshufps(ymm3, dscan_c, dscan_c, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// GSVector8i r = GSVector8i(dr * shift[1 + i]).ps32();
vmulps(ymm0, ymm2, _m_shift(1 + i));
vcvttps2dq(ymm0, ymm0);
vpackssdw(ymm0, ymm0);
// GSVector4i b = GSVector8i(db * shift[1 + i]).ps32();
vmulps(ymm1, ymm3, _m_shift(1 + i));
vcvttps2dq(ymm1, ymm1);
vpackssdw(ymm1, ymm1);
// m_local.d[i].rb = r.upl16(b);
vpunpcklwd(ymm0, ymm1);
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].rb) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].rb, variableOffset), ymm0);
}
// GSVector8 dg = dc.yyyy();
// GSVector8 da = dc.wwww();
vshufps(ymm2, dscan_c, dscan_c, _MM_SHUFFLE(1, 1, 1, 1));
vshufps(ymm3, dscan_c, dscan_c, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// GSVector8i g = GSVector8i(dg * shift[1 + i]).ps32();
vmulps(ymm0, ymm2, _m_shift(1 + i));
vcvttps2dq(ymm0, ymm0);
vpackssdw(ymm0, ymm0);
// GSVector8i a = GSVector8i(da * shift[1 + i]).ps32();
vmulps(ymm1, ymm3, _m_shift(1 + i));
vcvttps2dq(ymm1, ymm1);
vpackssdw(ymm1, ymm1);
// m_local.d[i].ga = g.upl16(a);
vpunpcklwd(ymm0, ymm1);
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].ga) + (i * sizeof(GSScanlineLocalData::d[0]));
vmovdqa(_rip_local_v(d[i].ga, variableOffset), ymm0);
}
}
else
{
// GSVector4i c = GSVector4i(vertex[index[last].c);
int last = 0;
switch (m_sel.prim)
{
case GS_POINT_CLASS: last = 0; break;
case GS_LINE_CLASS: last = 1; break;
case GS_TRIANGLE_CLASS: last = 2; break;
case GS_SPRITE_CLASS: last = 1; break;
}
if (!(m_sel.prim == GS_SPRITE_CLASS && (m_en.z || m_en.f))) // if this is a sprite, the last vertex was already loaded in Depth()
{
mov(_vertex.cvt32(), ptr[a1 + sizeof(uint32) * last]);
shl(_vertex.cvt32(), 6); // * sizeof(GSVertexSW)
add(_vertex, a0);
}
vbroadcasti128(ymm0, ptr[_vertex + offsetof(GSVertexSW, c)]);
vcvttps2dq(ymm0, ymm0);
// c = c.upl16(c.zwxy());
vpshufd(ymm1, ymm0, _MM_SHUFFLE(1, 0, 3, 2));
vpunpcklwd(ymm0, ymm1);
// if(!tme) c = c.srl16(7);
if (m_sel.tfx == TFX_NONE)
{
vpsrlw(ymm0, 7);
}
// m_local.c.rb = c.xxxx();
// m_local.c.ga = c.zzzz();
vpshufd(ymm1, ymm0, _MM_SHUFFLE(0, 0, 0, 0));
vpshufd(ymm2, ymm0, _MM_SHUFFLE(2, 2, 2, 2));
vmovdqa(_rip_local(c.rb), ymm1);
vmovdqa(_rip_local(c.ga), ymm2);
}
}
#endif

374
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.x64.cpp
View File

@ -1,374 +0,0 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2021 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "PrecompiledHeader.h"
#include "GSSetupPrimCodeGenerator.h"
#include "GSVertexSW.h"
#include "GS/GS_codegen.h"
#if _M_SSE < 0x501 && (defined(_M_AMD64) || defined(_WIN64))
void GSSetupPrimCodeGenerator::Generate_SSE()
{
#ifdef _WIN64
sub(rsp, 8 + 2 * 16);
vmovdqa(ptr[rsp + 0], xmm6);
vmovdqa(ptr[rsp + 16], xmm7);
#endif
mov(t0, (size_t)&m_local);
if ((m_en.z || m_en.f) && m_sel.prim != GS_SPRITE_CLASS || m_en.t || m_en.c && m_sel.iip)
{
mov(rax, (size_t)g_const->m_shift_128b[0]);
for (int i = 0; i < (m_sel.notest ? 2 : 5); i++)
{
movaps(Xmm(3 + i), ptr[rax + i * 16]);
}
}
Depth_SSE();
Texture_SSE();
Color_SSE();
#ifdef _WIN64
vmovdqa(xmm6, ptr[rsp + 0]);
vmovdqa(xmm7, ptr[rsp + 16]);
add(rsp, 8 + 2 * 16);
#endif
ret();
}
void GSSetupPrimCodeGenerator::Depth_SSE()
{
if (!m_en.z && !m_en.f)
{
return;
}
if (m_sel.prim != GS_SPRITE_CLASS)
{
// GSVector4 p = dscan.p;
movaps(xmm0, ptr[a2 + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// GSVector4 df = p.wwww();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
// m_local.d4.f = GSVector4i(df * 4.0f).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, d4.f)], xmm2);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].f = GSVector4i(df * m_shift[i]).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].f) + (i * sizeof(GSScanlineLocalData::d[0]));
movdqa(ptr[t0 + variableOffset], xmm2);
}
}
if (m_en.z)
{
// GSVector4 dz = p.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
// m_local.d4.z = dz * 4.0f;
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, d4.z)], xmm1);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].z = dz * m_shift[i];
movaps(xmm1, xmm0);
mulps(xmm1, Xmm(4 + i));
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].z) + (i * sizeof(GSScanlineLocalData::d[0]));
movdqa(ptr[t0 + variableOffset], xmm1);
}
}
}
else
{
// GSVector4 p = vertex[index[1]].p;
mov(eax, ptr[a1 + sizeof(uint32) * 1]);
shl(eax, 6); // * sizeof(GSVertexSW)
add(rax, a0);
movaps(xmm0, ptr[rax + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// m_local.p.f = GSVector4i(p).zzzzh().zzzz();
cvttps2dq(xmm1, xmm0);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, p.f)], xmm1);
}
if (m_en.z)
{
// uint32 z is bypassed in t.w
vmovdqa(xmm0, ptr[rax + offsetof(GSVertexSW, t)]);
vpshufd(xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
vmovdqa(ptr[t0 + offsetof(GSScanlineLocalData, p.z)], xmm0);
}
}
}
void GSSetupPrimCodeGenerator::Texture_SSE()
{
if (!m_en.t)
{
return;
}
// GSVector4 t = dscan.t;
movaps(xmm0, ptr[a2 + offsetof(GSVertexSW, t)]);
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
if (m_sel.fst)
{
// m_local.d4.stq = GSVector4i(t * 4.0f);
cvttps2dq(xmm1, xmm1);
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, d4.stq)], xmm1);
}
else
{
// m_local.d4.stq = t * 4.0f;
movaps(ptr[t0 + offsetof(GSScanlineLocalData, d4.stq)], xmm1);
}
for (int j = 0, k = m_sel.fst ? 2 : 3; j < k; j++)
{
// GSVector4 ds = t.xxxx();
// GSVector4 dt = t.yyyy();
// GSVector4 dq = t.zzzz();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, (uint8)_MM_SHUFFLE(j, j, j, j));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4 v = ds/dt * m_shift[i];
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
if (m_sel.fst)
{
// m_local.d[i].s/t = GSVector4i(v);
cvttps2dq(xmm2, xmm2);
const size_t variableOffsetS = offsetof(GSScanlineLocalData, d[0].s) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetT = offsetof(GSScanlineLocalData, d[0].t) + (i * sizeof(GSScanlineLocalData::d[0]));
switch (j)
{
case 0: movdqa(ptr[t0 + variableOffsetS], xmm2); break;
case 1: movdqa(ptr[t0 + variableOffsetT], xmm2); break;
}
}
else
{
// m_local.d[i].s/t/q = v;
const size_t variableOffsetS = offsetof(GSScanlineLocalData, d[0].s) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetT = offsetof(GSScanlineLocalData, d[0].t) + (i * sizeof(GSScanlineLocalData::d[0]));
const size_t variableOffsetQ = offsetof(GSScanlineLocalData, d[0].q) + (i * sizeof(GSScanlineLocalData::d[0]));
switch (j)
{
case 0: movaps(ptr[t0 + variableOffsetS], xmm2); break;
case 1: movaps(ptr[t0 + variableOffsetT], xmm2); break;
case 2: movaps(ptr[t0 + variableOffsetQ], xmm2); break;
}
}
}
}
}
void GSSetupPrimCodeGenerator::Color_SSE()
{
if (!m_en.c)
{
return;
}
if (m_sel.iip)
{
// GSVector4 c = dscan.c;
movaps(xmm0, ptr[a2 + offsetof(GSVertexSW, c)]);
movaps(xmm1, xmm0);
// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
movaps(xmm2, xmm0);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshufd(xmm2, xmm2, _MM_SHUFFLE(3, 1, 2, 0));
packssdw(xmm2, xmm2);
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, d4.c)], xmm2);
// xmm3 is not needed anymore
// GSVector4 dr = c.xxxx();
// GSVector4 db = c.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i r = GSVector4i(dr * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i b = GSVector4i(db * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_local.d[i].rb = r.upl16(b);
punpcklwd(xmm2, xmm3);
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].rb) + (i * sizeof(GSScanlineLocalData::d[0]));
movdqa(ptr[t0 + variableOffset], xmm2);
}
// GSVector4 c = dscan.c;
movaps(xmm0, ptr[a2 + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
movaps(xmm1, xmm0);
// GSVector4 dg = c.yyyy();
// GSVector4 da = c.wwww();
shufps(xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i g = GSVector4i(dg * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i a = GSVector4i(da * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_local.d[i].ga = g.upl16(a);
punpcklwd(xmm2, xmm3);
const size_t variableOffset = offsetof(GSScanlineLocalData, d[0].ga) + (i * sizeof(GSScanlineLocalData::d[0]));
movdqa(ptr[t0 + variableOffset], xmm2);
}
}
else
{
// GSVector4i c = GSVector4i(vertex[index[last].c);
int last = 0;
switch (m_sel.prim)
{
case GS_POINT_CLASS: last = 0; break;
case GS_LINE_CLASS: last = 1; break;
case GS_TRIANGLE_CLASS: last = 2; break;
case GS_SPRITE_CLASS: last = 1; break;
}
if (!(m_sel.prim == GS_SPRITE_CLASS && (m_en.z || m_en.f))) // if this is a sprite, the last vertex was already loaded in Depth()
{
mov(eax, ptr[a1 + sizeof(uint32) * last]);
shl(eax, 6); // * sizeof(GSVertexSW)
add(rax, a0);
}
cvttps2dq(xmm0, ptr[rax + offsetof(GSVertexSW, c)]);
// c = c.upl16(c.zwxy());
pshufd(xmm1, xmm0, _MM_SHUFFLE(1, 0, 3, 2));
punpcklwd(xmm0, xmm1);
// if(!tme) c = c.srl16(7);
if (m_sel.tfx == TFX_NONE)
{
psrlw(xmm0, 7);
}
// m_local.c.rb = c.xxxx();
// m_local.c.ga = c.zzzz();
pshufd(xmm1, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm2, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, c.rb)], xmm1);
movdqa(ptr[t0 + offsetof(GSScanlineLocalData, c.ga)], xmm2);
}
}
#endif

335
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.x86.avx.cpp
View File

@ -1,335 +0,0 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2021 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "PrecompiledHeader.h"
#include "GSSetupPrimCodeGenerator.h"
#include "GSVertexSW.h"
#include "GS/GS_codegen.h"
#if _M_SSE < 0x501 && !(defined(_M_AMD64) || defined(_WIN64))
static const int _args = 0;
static const int _vertex = _args + 4;
static const int _index = _args + 8;
static const int _dscan = _args + 12;
void GSSetupPrimCodeGenerator::Generate_AVX()
{
if ((m_en.z || m_en.f) && m_sel.prim != GS_SPRITE_CLASS || m_en.t || m_en.c && m_sel.iip)
{
mov(edx, dword[esp + _dscan]);
for (int i = 0; i < (m_sel.notest ? 2 : 5); i++)
{
vmovaps(Xmm(3 + i), ptr[g_const->m_shift_128b[i]]);
}
}
Depth_AVX();
Texture_AVX();
Color_AVX();
ret();
}
void GSSetupPrimCodeGenerator::Depth_AVX()
{
if (!m_en.z && !m_en.f)
{
return;
}
if (m_sel.prim != GS_SPRITE_CLASS)
{
// GSVector4 p = dscan.p;
vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// GSVector4 df = p.wwww();
vshufps(xmm1, xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
// m_local.d4.f = GSVector4i(df * 4.0f).xxzzlh();
vmulps(xmm2, xmm1, xmm3);
vcvttps2dq(xmm2, xmm2);
vpshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vmovdqa(ptr[&m_local.d4.f], xmm2);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].f = GSVector4i(df * m_shift[i]).xxzzlh();
vmulps(xmm2, xmm1, Xmm(4 + i));
vcvttps2dq(xmm2, xmm2);
vpshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
vmovdqa(ptr[&m_local.d[i].f], xmm2);
}
}
if (m_en.z)
{
// GSVector4 dz = p.zzzz();
vshufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
// m_local.d4.z = dz * 4.0f;
vmulps(xmm1, xmm0, xmm3);
vmovdqa(ptr[&m_local.d4.z], xmm1);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].z = dz * m_shift[i];
vmulps(xmm1, xmm0, Xmm(4 + i));
vmovdqa(ptr[&m_local.d[i].z], xmm1);
}
}
}
else
{
// GSVector4 p = vertex[index[1]].p;
mov(ecx, ptr[esp + _index]);
mov(ecx, ptr[ecx + sizeof(uint32) * 1]);
shl(ecx, 6); // * sizeof(GSVertexSW)
add(ecx, ptr[esp + _vertex]);
vmovaps(xmm0, ptr[ecx + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// m_local.p.f = GSVector4i(p).zzzzh().zzzz();
vcvttps2dq(xmm1, xmm0);
vpshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
vpshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
vmovdqa(ptr[&m_local.p.f], xmm1);
}
if (m_en.z)
{
// uint32 z is bypassed in t.w
vmovdqa(xmm0, ptr[ecx + offsetof(GSVertexSW, t)]);
vpshufd(xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
vmovdqa(ptr[&m_local.p.z], xmm0);
}
}
}
void GSSetupPrimCodeGenerator::Texture_AVX()
{
if (!m_en.t)
{
return;
}
// GSVector4 t = dscan.t;
vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, t)]);
vmulps(xmm1, xmm0, xmm3);
if (m_sel.fst)
{
// m_local.d4.stq = GSVector4i(t * 4.0f);
vcvttps2dq(xmm1, xmm1);
vmovdqa(ptr[&m_local.d4.stq], xmm1);
}
else
{
// m_local.d4.stq = t * 4.0f;
vmovaps(ptr[&m_local.d4.stq], xmm1);
}
for (int j = 0, k = m_sel.fst ? 2 : 3; j < k; j++)
{
// GSVector4 ds = t.xxxx();
// GSVector4 dt = t.yyyy();
// GSVector4 dq = t.zzzz();
vshufps(xmm1, xmm0, xmm0, (uint8)_MM_SHUFFLE(j, j, j, j));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4 v = ds/dt * m_shift[i];
vmulps(xmm2, xmm1, Xmm(4 + i));
if (m_sel.fst)
{
// m_local.d[i].s/t = GSVector4i(v);
vcvttps2dq(xmm2, xmm2);
switch (j)
{
case 0: vmovdqa(ptr[&m_local.d[i].s], xmm2); break;
case 1: vmovdqa(ptr[&m_local.d[i].t], xmm2); break;
}
}
else
{
// m_local.d[i].s/t/q = v;
switch (j)
{
case 0: vmovaps(ptr[&m_local.d[i].s], xmm2); break;
case 1: vmovaps(ptr[&m_local.d[i].t], xmm2); break;
case 2: vmovaps(ptr[&m_local.d[i].q], xmm2); break;
}
}
}
}
}
void GSSetupPrimCodeGenerator::Color_AVX()
{
if (!m_en.c)
{
return;
}
if (m_sel.iip)
{
// GSVector4 c = dscan.c;
vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]);
// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
vmulps(xmm1, xmm0, xmm3);
vcvttps2dq(xmm1, xmm1);
vpshufd(xmm1, xmm1, _MM_SHUFFLE(3, 1, 2, 0));
vpackssdw(xmm1, xmm1);
vmovdqa(ptr[&m_local.d4.c], xmm1);
// xmm3 is not needed anymore
// GSVector4 dr = c.xxxx();
// GSVector4 db = c.zzzz();
vshufps(xmm2, xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vshufps(xmm3, xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i r = GSVector4i(dr * m_shift[i]).ps32();
vmulps(xmm0, xmm2, Xmm(4 + i));
vcvttps2dq(xmm0, xmm0);
vpackssdw(xmm0, xmm0);
// GSVector4i b = GSVector4i(db * m_shift[i]).ps32();
vmulps(xmm1, xmm3, Xmm(4 + i));
vcvttps2dq(xmm1, xmm1);
vpackssdw(xmm1, xmm1);
// m_local.d[i].rb = r.upl16(b);
vpunpcklwd(xmm0, xmm1);
vmovdqa(ptr[&m_local.d[i].rb], xmm0);
}
// GSVector4 c = dscan.c;
vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
// GSVector4 dg = c.yyyy();
// GSVector4 da = c.wwww();
vshufps(xmm2, xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
vshufps(xmm3, xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i g = GSVector4i(dg * m_shift[i]).ps32();
vmulps(xmm0, xmm2, Xmm(4 + i));
vcvttps2dq(xmm0, xmm0);
vpackssdw(xmm0, xmm0);
// GSVector4i a = GSVector4i(da * m_shift[i]).ps32();
vmulps(xmm1, xmm3, Xmm(4 + i));
vcvttps2dq(xmm1, xmm1);
vpackssdw(xmm1, xmm1);
// m_local.d[i].ga = g.upl16(a);
vpunpcklwd(xmm0, xmm1);
vmovdqa(ptr[&m_local.d[i].ga], xmm0);
}
}
else
{
// GSVector4i c = GSVector4i(vertex[index[last].c);
int last = 0;
switch (m_sel.prim)
{
case GS_POINT_CLASS: last = 0; break;
case GS_LINE_CLASS: last = 1; break;
case GS_TRIANGLE_CLASS: last = 2; break;
case GS_SPRITE_CLASS: last = 1; break;
}
if (!(m_sel.prim == GS_SPRITE_CLASS && (m_en.z || m_en.f))) // if this is a sprite, the last vertex was already loaded in Depth()
{
mov(ecx, ptr[esp + _index]);
mov(ecx, ptr[ecx + sizeof(uint32) * last]);
shl(ecx, 6); // * sizeof(GSVertexSW)
add(ecx, ptr[esp + _vertex]);
}
vcvttps2dq(xmm0, ptr[ecx + offsetof(GSVertexSW, c)]);
// c = c.upl16(c.zwxy());
vpshufd(xmm1, xmm0, _MM_SHUFFLE(1, 0, 3, 2));
vpunpcklwd(xmm0, xmm1);
// if(!tme) c = c.srl16(7);
if (m_sel.tfx == TFX_NONE)
{
vpsrlw(xmm0, 7);
}
// m_local.c.rb = c.xxxx();
// m_local.c.ga = c.zzzz();
vpshufd(xmm1, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
vpshufd(xmm2, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
vmovdqa(ptr[&m_local.c.rb], xmm1);
vmovdqa(ptr[&m_local.c.ga], xmm2);
}
}
#endif

360
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.x86.avx2.cpp
View File

@ -1,360 +0,0 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2021 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "PrecompiledHeader.h"
#include "GSSetupPrimCodeGenerator.h"
#include "GSVertexSW.h"
#include "GS/GS_codegen.h"
#if _M_SSE >= 0x501 && !(defined(_M_AMD64) || defined(_WIN64))
static const int _args = 0;
static const int _vertex = _args + 4;
static const int _index = _args + 8;
static const int _dscan = _args + 12;
void GSSetupPrimCodeGenerator::Generate_AVX2()
{
if ((m_en.z || m_en.f) && m_sel.prim != GS_SPRITE_CLASS || m_en.t || m_en.c && m_sel.iip)
{
mov(edx, dword[esp + _dscan]);
for (int i = 0; i < (m_sel.notest ? 2 : 5); i++)
{
vmovaps(Ymm(3 + i), ptr[g_const->m_shift_256b[i]]);
}
}
Depth_AVX2();
Texture_AVX2();
Color_AVX2();
ret();
}
void GSSetupPrimCodeGenerator::Depth_AVX2()
{
if (!m_en.z && !m_en.f)
{
return;
}
if (m_sel.prim != GS_SPRITE_CLASS)
{
// GSVector4 dp8 = dscan.p * GSVector4::broadcast32(&shift[0]);
vbroadcastf128(ymm0, ptr[edx + offsetof(GSVertexSW, p)]);
vmulps(ymm1, ymm0, ymm3);
if (m_en.z)
{
// m_local.d8.p.z = dp8.extract32<2>();
vextractps(ptr[&m_local.d8.p.z], xmm1, 2);
}
if (m_en.f)
{
// m_local.d8.p.f = GSVector4i(dp8).extract32<3>();
vcvtps2dq(ymm2, ymm1);
vpextrd(ptr[&m_local.d8.p.f], xmm2, 3);
}
if (m_en.z)
{
// GSVector8 dz = GSVector8(dscan.p).zzzz();
vshufps(ymm2, ymm0, ymm0, _MM_SHUFFLE(2, 2, 2, 2));
}
if (m_en.f)
{
// GSVector8 df = GSVector8(dscan.p).wwww();
vshufps(ymm1, ymm0, ymm0, _MM_SHUFFLE(3, 3, 3, 3));
}
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
if (m_en.z)
{
// m_local.d[i].z = dz * shift[1 + i];
if (i < 4)
vmulps(ymm0, ymm2, Ymm(4 + i));
else
vmulps(ymm0, ymm2, ptr[g_const->m_shift_256b[i + 1]]);
vmovaps(ptr[&m_local.d[i].z], ymm0);
}
if (m_en.f)
{
// m_local.d[i].f = GSVector8i(df * m_shift[i]).xxzzlh();
if (i < 4)
vmulps(ymm0, ymm1, Ymm(4 + i));
else
vmulps(ymm0, ymm1, ptr[g_const->m_shift_256b[i + 1]]);
vcvttps2dq(ymm0, ymm0);
vpshuflw(ymm0, ymm0, _MM_SHUFFLE(2, 2, 0, 0));
vpshufhw(ymm0, ymm0, _MM_SHUFFLE(2, 2, 0, 0));
vmovdqa(ptr[&m_local.d[i].f], ymm0);
}
}
}
else
{
// GSVector4 p = vertex[index[1]].p;
mov(ecx, ptr[esp + _index]);
mov(ecx, ptr[ecx + sizeof(uint32) * 1]);
shl(ecx, 6); // * sizeof(GSVertexSW)
add(ecx, ptr[esp + _vertex]);
if (m_en.f)
{
// m_local.p.f = GSVector4i(vertex[index[1]].p).extract32<3>();
vmovaps(xmm0, ptr[ecx + offsetof(GSVertexSW, p)]);
vcvttps2dq(xmm0, xmm0);
vpextrd(ptr[&m_local.p.f], xmm0, 3);
}
if (m_en.z)
{
// m_local.p.z = vertex[index[1]].t.u32[3]; // uint32 z is bypassed in t.w
mov(eax, ptr[ecx + offsetof(GSVertexSW, t.w)]);
mov(ptr[&m_local.p.z], eax);
}
}
}
void GSSetupPrimCodeGenerator::Texture_AVX2()
{
if (!m_en.t)
{
return;
}
// GSVector8 dt(dscan.t);
vbroadcastf128(ymm0, ptr[edx + offsetof(GSVertexSW, t)]);
// GSVector8 dt8 = dt * shift[0];
vmulps(ymm1, ymm0, ymm3);
if (m_sel.fst)
{
// m_local.d8.stq = GSVector8::cast(GSVector8i(dt8));
vcvttps2dq(ymm1, ymm1);
vmovdqa(ptr[&m_local.d8.stq], xmm1);
}
else
{
// m_local.d8.stq = dt8;
vmovaps(ptr[&m_local.d8.stq], xmm1);
}
for (int j = 0, k = m_sel.fst ? 2 : 3; j < k; j++)
{
// GSVector8 dstq = dt.xxxx/yyyy/zzzz();
vshufps(ymm1, ymm0, ymm0, (uint8)_MM_SHUFFLE(j, j, j, j));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// GSVector8 v = dstq * shift[1 + i];
if (i < 4)
vmulps(ymm2, ymm1, Ymm(4 + i));
else
vmulps(ymm2, ymm1, ptr[g_const->m_shift_256b[i + 1]]);
if (m_sel.fst)
{
// m_local.d[i].s/t = GSVector8::cast(GSVector8i(v));
vcvttps2dq(ymm2, ymm2);
switch (j)
{
case 0: vmovdqa(ptr[&m_local.d[i].s], ymm2); break;
case 1: vmovdqa(ptr[&m_local.d[i].t], ymm2); break;
}
}
else
{
// m_local.d[i].s/t/q = v;
switch (j)
{
case 0: vmovaps(ptr[&m_local.d[i].s], ymm2); break;
case 1: vmovaps(ptr[&m_local.d[i].t], ymm2); break;
case 2: vmovaps(ptr[&m_local.d[i].q], ymm2); break;
}
}
}
}
}
void GSSetupPrimCodeGenerator::Color_AVX2()
{
if (!m_en.c)
{
return;
}
if (m_sel.iip)
{
// GSVector8 dc(dscan.c);
vbroadcastf128(ymm0, ptr[edx + offsetof(GSVertexSW, c)]);
// m_local.d8.c = GSVector8i(dc * shift[0]).xzyw().ps32();
vmulps(ymm1, ymm0, ymm3);
vcvttps2dq(ymm1, ymm1);
vpshufd(ymm1, ymm1, _MM_SHUFFLE(3, 1, 2, 0));
vpackssdw(ymm1, ymm1);
vmovq(ptr[&m_local.d8.c], xmm1);
// ymm3 is not needed anymore
// GSVector8 dr = dc.xxxx();
// GSVector8 db = dc.zzzz();
vshufps(ymm2, ymm0, ymm0, _MM_SHUFFLE(0, 0, 0, 0));
vshufps(ymm3, ymm0, ymm0, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// GSVector8i r = GSVector8i(dr * shift[1 + i]).ps32();
if (i < 4)
vmulps(ymm0, ymm2, Ymm(4 + i));
else
vmulps(ymm0, ymm2, ptr[g_const->m_shift_256b[i + 1]]);
vcvttps2dq(ymm0, ymm0);
vpackssdw(ymm0, ymm0);
// GSVector4i b = GSVector8i(db * shift[1 + i]).ps32();
if (i < 4)
vmulps(ymm1, ymm3, Ymm(4 + i));
else
vmulps(ymm1, ymm3, ptr[g_const->m_shift_256b[i + 1]]);
vcvttps2dq(ymm1, ymm1);
vpackssdw(ymm1, ymm1);
// m_local.d[i].rb = r.upl16(b);
vpunpcklwd(ymm0, ymm1);
vmovdqa(ptr[&m_local.d[i].rb], ymm0);
}
// GSVector8 dc(dscan.c);
vbroadcastf128(ymm0, ptr[edx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
// GSVector8 dg = dc.yyyy();
// GSVector8 da = dc.wwww();
vshufps(ymm2, ymm0, ymm0, _MM_SHUFFLE(1, 1, 1, 1));
vshufps(ymm3, ymm0, ymm0, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 8); i++)
{
// GSVector8i g = GSVector8i(dg * shift[1 + i]).ps32();
if (i < 4)
vmulps(ymm0, ymm2, Ymm(4 + i));
else
vmulps(ymm0, ymm2, ptr[g_const->m_shift_256b[i + 1]]);
vcvttps2dq(ymm0, ymm0);
vpackssdw(ymm0, ymm0);
// GSVector8i a = GSVector8i(da * shift[1 + i]).ps32();
if (i < 4)
vmulps(ymm1, ymm3, Ymm(4 + i));
else
vmulps(ymm1, ymm3, ptr[g_const->m_shift_256b[i + 1]]);
vcvttps2dq(ymm1, ymm1);
vpackssdw(ymm1, ymm1);
// m_local.d[i].ga = g.upl16(a);
vpunpcklwd(ymm0, ymm1);
vmovdqa(ptr[&m_local.d[i].ga], ymm0);
}
}
else
{
// GSVector8i c = GSVector8i(GSVector8(vertex[index[last]].c));
int last = 0;
switch (m_sel.prim)
{
case GS_POINT_CLASS: last = 0; break;
case GS_LINE_CLASS: last = 1; break;
case GS_TRIANGLE_CLASS: last = 2; break;
case GS_SPRITE_CLASS: last = 1; break;
}
if (!(m_sel.prim == GS_SPRITE_CLASS && (m_en.z || m_en.f))) // if this is a sprite, the last vertex was already loaded in Depth()
{
mov(ecx, ptr[esp + _index]);
mov(ecx, ptr[ecx + sizeof(uint32) * last]);
shl(ecx, 6); // * sizeof(GSVertexSW)
add(ecx, ptr[esp + _vertex]);
}
vbroadcasti128(ymm0, ptr[ecx + offsetof(GSVertexSW, c)]);
vcvttps2dq(ymm0, ymm0);
// c = c.upl16(c.zwxy());
vpshufd(ymm1, ymm0, _MM_SHUFFLE(1, 0, 3, 2));
vpunpcklwd(ymm0, ymm1);
// if(!tme) c = c.srl16(7);
if (m_sel.tfx == TFX_NONE)
{
vpsrlw(ymm0, 7);
}
// m_local.c.rb = c.xxxx();
// m_local.c.ga = c.zzzz();
vpshufd(ymm1, ymm0, _MM_SHUFFLE(0, 0, 0, 0));
vpshufd(ymm2, ymm0, _MM_SHUFFLE(2, 2, 2, 2));
vmovdqa(ptr[&m_local.c.rb], ymm1);
vmovdqa(ptr[&m_local.c.ga], ymm2);
}
}
#endif

350
pcsx2/GS/Renderers/SW/GSSetupPrimCodeGenerator.x86.cpp
View File

@ -1,350 +0,0 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2021 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "PrecompiledHeader.h"
#include "GSSetupPrimCodeGenerator.h"
#include "GSVertexSW.h"
#include "GS/GS_codegen.h"
#if _M_SSE < 0x501 && !(defined(_M_AMD64) || defined(_WIN64))
static const int _args = 0;
static const int _vertex = _args + 4;
static const int _index = _args + 8;
static const int _dscan = _args + 12;
void GSSetupPrimCodeGenerator::Generate_SSE()
{
if ((m_en.z || m_en.f) && m_sel.prim != GS_SPRITE_CLASS || m_en.t || m_en.c && m_sel.iip)
{
mov(edx, dword[esp + _dscan]);
for (int i = 0; i < (m_sel.notest ? 2 : 5); i++)
{
movaps(Xmm(3 + i), ptr[g_const->m_shift_128b[i]]);
}
}
Depth_SSE();
Texture_SSE();
Color_SSE();
ret();
}
void GSSetupPrimCodeGenerator::Depth_SSE()
{
if (!m_en.z && !m_en.f)
{
return;
}
if (m_sel.prim != GS_SPRITE_CLASS)
{
// GSVector4 p = dscan.p;
movaps(xmm0, ptr[edx + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// GSVector4 df = p.wwww();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
// m_local.d4.f = GSVector4i(df * 4.0f).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(ptr[&m_local.d4.f], xmm2);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].f = GSVector4i(df * m_shift[i]).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(ptr[&m_local.d[i].f], xmm2);
}
}
if (m_en.z)
{
// GSVector4 dz = p.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
// m_local.d4.z = dz * 4.0f;
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
movdqa(ptr[&m_local.d4.z], xmm1);
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// m_local.d[i].z = dz * m_shift[i];
movaps(xmm1, xmm0);
mulps(xmm1, Xmm(4 + i));
movdqa(ptr[&m_local.d[i].z], xmm1);
}
}
}
else
{
// GSVector4 p = vertex[index[1]].p;
mov(ecx, ptr[esp + _index]);
mov(ecx, ptr[ecx + sizeof(uint32) * 1]);
shl(ecx, 6); // * sizeof(GSVertexSW)
add(ecx, ptr[esp + _vertex]);
movaps(xmm0, ptr[ecx + offsetof(GSVertexSW, p)]);
if (m_en.f)
{
// m_local.p.f = GSVector4i(p).zzzzh().zzzz();
cvttps2dq(xmm1, xmm0);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(ptr[&m_local.p.f], xmm1);
}
if (m_en.z)
{
// uint32 z is bypassed in t.w
movdqa(xmm0, ptr[ecx + offsetof(GSVertexSW, t)]);
pshufd(xmm0, xmm0, _MM_SHUFFLE(3, 3, 3, 3));
movdqa(ptr[&m_local.p.z], xmm0);
}
}
}
void GSSetupPrimCodeGenerator::Texture_SSE()
{
if (!m_en.t)
{
return;
}
// GSVector4 t = dscan.t;
movaps(xmm0, ptr[edx + offsetof(GSVertexSW, t)]);
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
if (m_sel.fst)
{
// m_local.d4.stq = GSVector4i(t * 4.0f);
cvttps2dq(xmm1, xmm1);
movdqa(ptr[&m_local.d4.stq], xmm1);
}
else
{
// m_local.d4.stq = t * 4.0f;
movaps(ptr[&m_local.d4.stq], xmm1);
}
for (int j = 0, k = m_sel.fst ? 2 : 3; j < k; j++)
{
// GSVector4 ds = t.xxxx();
// GSVector4 dt = t.yyyy();
// GSVector4 dq = t.zzzz();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, (uint8)_MM_SHUFFLE(j, j, j, j));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4 v = ds/dt * m_shift[i];
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
if (m_sel.fst)
{
// m_local.d[i].s/t = GSVector4i(v);
cvttps2dq(xmm2, xmm2);
switch (j)
{
case 0: movdqa(ptr[&m_local.d[i].s], xmm2); break;
case 1: movdqa(ptr[&m_local.d[i].t], xmm2); break;
}
}
else
{
// m_local.d[i].s/t/q = v;
switch (j)
{
case 0: movaps(ptr[&m_local.d[i].s], xmm2); break;
case 1: movaps(ptr[&m_local.d[i].t], xmm2); break;
case 2: movaps(ptr[&m_local.d[i].q], xmm2); break;
}
}
}
}
}
void GSSetupPrimCodeGenerator::Color_SSE()
{
if (!m_en.c)
{
return;
}
if (m_sel.iip)
{
// GSVector4 c = dscan.c;
movaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]);
movaps(xmm1, xmm0);
// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
movaps(xmm2, xmm0);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshufd(xmm2, xmm2, _MM_SHUFFLE(3, 1, 2, 0));
packssdw(xmm2, xmm2);
movdqa(ptr[&m_local.d4.c], xmm2);
// xmm3 is not needed anymore
// GSVector4 dr = c.xxxx();
// GSVector4 db = c.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i r = GSVector4i(dr * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i b = GSVector4i(db * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_local.d[i].rb = r.upl16(b);
punpcklwd(xmm2, xmm3);
movdqa(ptr[&m_local.d[i].rb], xmm2);
}
// GSVector4 c = dscan.c;
movaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
movaps(xmm1, xmm0);
// GSVector4 dg = c.yyyy();
// GSVector4 da = c.wwww();
shufps(xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
for (int i = 0; i < (m_sel.notest ? 1 : 4); i++)
{
// GSVector4i g = GSVector4i(dg * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i a = GSVector4i(da * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_local.d[i].ga = g.upl16(a);
punpcklwd(xmm2, xmm3);
movdqa(ptr[&m_local.d[i].ga], xmm2);
}
}
else
{
// GSVector4i c = GSVector4i(vertex[index[last].c);
int last = 0;
switch (m_sel.prim)
{
case GS_POINT_CLASS: last = 0; break;
case GS_LINE_CLASS: last = 1; break;
case GS_TRIANGLE_CLASS: last = 2; break;
case GS_SPRITE_CLASS: last = 1; break;
}
if (!(m_sel.prim == GS_SPRITE_CLASS && (m_en.z || m_en.f))) // if this is a sprite, the last vertex was already loaded in Depth()
{
mov(ecx, ptr[esp + _index]);
mov(ecx, ptr[ecx + sizeof(uint32) * last]);
shl(ecx, 6); // * sizeof(GSVertexSW)
add(ecx, ptr[esp + _vertex]);
}
cvttps2dq(xmm0, ptr[ecx + offsetof(GSVertexSW, c)]);
// c = c.upl16(c.zwxy());
pshufd(xmm1, xmm0, _MM_SHUFFLE(1, 0, 3, 2));
punpcklwd(xmm0, xmm1);
// if(!tme) c = c.srl16(7);
if (m_sel.tfx == TFX_NONE)
{
psrlw(xmm0, 7);
}
// m_local.c.rb = c.xxxx();
// m_local.c.ga = c.zzzz();
pshufd(xmm1, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm2, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(ptr[&m_local.c.rb], xmm1);
movdqa(ptr[&m_local.c.ga], xmm2);
}
}
#endif

6
pcsx2/pcsx2.vcxproj
View File

@ -492,12 +492,6 @@
<ClCompile Include="GS\Window\GSSettingsDlg.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.all.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x64.avx.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x64.avx2.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x64.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x86.avx.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x86.avx2.cpp" />
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x86.cpp" />
<ClCompile Include="GS\Renderers\OpenGL\GSShaderOGL.cpp" />
<ClCompile Include="GS\GSState.cpp" />
<ClCompile Include="GS\GSTables.cpp" />

21
pcsx2/pcsx2.vcxproj.filters
View File

@ -1544,27 +1544,6 @@
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.all.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x64.avx.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x64.avx2.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x64.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x86.avx.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x86.avx2.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSSetupPrimCodeGenerator.x86.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>
<ClCompile Include="GS\Renderers\SW\GSTextureCacheSW.cpp">
<Filter>System\Ps2\GS\Renderers\Software</Filter>
</ClCompile>