mirror of
https://github.com/asmjit/asmjit.git
synced 2025-12-18 04:54:36 +03:00
3fcd65cf80
- Compiler no longer works on its own, it requires Assembler. - Labels created by Assembler and Compiler now share their IDs, so they can be used nearly interchangeably without weird side-effects and hacks. - Renamed getError() and setError() to getLastError() and setLastError(). - Renamed compiler nodes to "HL" nodes (preparation for HLStream). - Renamed FuncConv to CallConv. - Function calling convention is now part of FuncPrototype. - Added a possibility to align by inserting zeros (kAlignZero) - Fixed assertion in X86Compiler that didn't like unhandled function argument(s). - Added Compiler::embedConstPool() helper, which can be handy if you use your own ConstPool. - Code refactorization and other minor changes. - CpuTicks::now() renamed to Utils::getTickCount(). - error.h merged with globals.h - Documentation updates related to recent API changes.
181 lines
3.2 KiB
C++
181 lines
3.2 KiB
C++
// [AsmJit]
|
|
// Complete x86/x64 JIT and Remote Assembler for C++.
|
|
//
|
|
// [License]
|
|
// Zlib - See LICENSE.md file in the package.
|
|
|
|
// [Guard]
|
|
#ifndef _TEST_GENBLEND_H
|
|
#define _TEST_GENBLEND_H
|
|
|
|
// [Dependencies]
|
|
#include "../asmjit/asmjit.h"
|
|
|
|
namespace asmgen {
|
|
|
|
enum { kGenBlendInstCount = 65 };
|
|
|
|
// Generate a typical alpha blend function using SSE2 instruction set. Used
|
|
// for benchmarking and also in test86. The generated code should be stable
|
|
// and fully functional.
|
|
static void blend(asmjit::X86Compiler& c) {
|
|
using namespace asmjit;
|
|
using namespace asmjit::x86;
|
|
|
|
X86GpVar dst = c.newIntPtr("dst");
|
|
X86GpVar src = c.newIntPtr("src");
|
|
|
|
X86GpVar i = c.newIntPtr("i");
|
|
X86GpVar j = c.newIntPtr("j");
|
|
X86GpVar t = c.newIntPtr("t");
|
|
|
|
X86XmmVar cZero = c.newXmm("cZero");
|
|
X86XmmVar cMul255A = c.newXmm("cMul255A");
|
|
X86XmmVar cMul255M = c.newXmm("cMul255M");
|
|
|
|
X86XmmVar x0 = c.newXmm("x0");
|
|
X86XmmVar x1 = c.newXmm("x1");
|
|
X86XmmVar y0 = c.newXmm("y0");
|
|
X86XmmVar a0 = c.newXmm("a0");
|
|
X86XmmVar a1 = c.newXmm("a1");
|
|
|
|
Label L_SmallLoop(c);
|
|
Label L_SmallEnd(c);
|
|
|
|
Label L_LargeLoop(c);
|
|
Label L_LargeEnd(c);
|
|
|
|
Label L_Data(c);
|
|
|
|
c.addFunc(FuncBuilder3<Void, void*, const void*, size_t>(kCallConvHost));
|
|
|
|
c.setArg(0, dst);
|
|
c.setArg(1, src);
|
|
c.setArg(2, i);
|
|
|
|
c.alloc(dst);
|
|
c.alloc(src);
|
|
c.alloc(i);
|
|
|
|
// How many pixels have to be processed to make the loop aligned.
|
|
c.lea(t, ptr(L_Data));
|
|
c.xor_(j, j);
|
|
c.xorps(cZero, cZero);
|
|
|
|
c.sub(j, dst);
|
|
c.movaps(cMul255A, ptr(t, 0));
|
|
|
|
c.and_(j, 15);
|
|
c.movaps(cMul255M, ptr(t, 16));
|
|
|
|
c.shr(j, 2);
|
|
c.jz(L_SmallEnd);
|
|
|
|
// j = min(i, j).
|
|
c.cmp(j, i);
|
|
c.cmovg(j, i);
|
|
|
|
// i -= j.
|
|
c.sub(i, j);
|
|
|
|
// Small loop.
|
|
c.bind(L_SmallLoop);
|
|
|
|
c.pcmpeqb(a0, a0);
|
|
c.movd(y0, ptr(src));
|
|
|
|
c.pxor(a0, y0);
|
|
c.movd(x0, ptr(dst));
|
|
|
|
c.psrlw(a0, 8);
|
|
c.punpcklbw(x0, cZero);
|
|
|
|
c.pshuflw(a0, a0, X86Util::shuffle(1, 1, 1, 1));
|
|
c.punpcklbw(y0, cZero);
|
|
|
|
c.pmullw(x0, a0);
|
|
c.paddsw(x0, cMul255A);
|
|
c.pmulhuw(x0, cMul255M);
|
|
|
|
c.paddw(x0, y0);
|
|
c.packuswb(x0, x0);
|
|
|
|
c.movd(ptr(dst), x0);
|
|
|
|
c.add(dst, 4);
|
|
c.add(src, 4);
|
|
|
|
c.dec(j);
|
|
c.jnz(L_SmallLoop);
|
|
|
|
// Second section, prepare for an aligned loop.
|
|
c.bind(L_SmallEnd);
|
|
|
|
c.test(i, i);
|
|
c.mov(j, i);
|
|
c.jz(c.getFunc()->getExitLabel());
|
|
|
|
c.and_(j, 3);
|
|
c.shr(i, 2);
|
|
c.jz(L_LargeEnd);
|
|
|
|
// Aligned loop.
|
|
c.bind(L_LargeLoop);
|
|
|
|
c.movups(y0, ptr(src));
|
|
c.pcmpeqb(a0, a0);
|
|
c.movaps(x0, ptr(dst));
|
|
|
|
c.xorps(a0, y0);
|
|
c.movaps(x1, x0);
|
|
|
|
c.psrlw(a0, 8);
|
|
c.punpcklbw(x0, cZero);
|
|
|
|
c.movaps(a1, a0);
|
|
c.punpcklwd(a0, a0);
|
|
|
|
c.punpckhbw(x1, cZero);
|
|
c.punpckhwd(a1, a1);
|
|
|
|
c.pshufd(a0, a0, X86Util::shuffle(3, 3, 1, 1));
|
|
c.pshufd(a1, a1, X86Util::shuffle(3, 3, 1, 1));
|
|
|
|
c.pmullw(x0, a0);
|
|
c.pmullw(x1, a1);
|
|
|
|
c.paddsw(x0, cMul255A);
|
|
c.paddsw(x1, cMul255A);
|
|
|
|
c.pmulhuw(x0, cMul255M);
|
|
c.pmulhuw(x1, cMul255M);
|
|
|
|
c.add(src, 16);
|
|
c.packuswb(x0, x1);
|
|
|
|
c.paddw(x0, y0);
|
|
c.movaps(ptr(dst), x0);
|
|
|
|
c.add(dst, 16);
|
|
|
|
c.dec(i);
|
|
c.jnz(L_LargeLoop);
|
|
|
|
c.bind(L_LargeEnd);
|
|
c.test(j, j);
|
|
c.jnz(L_SmallLoop);
|
|
|
|
c.endFunc();
|
|
|
|
// Data.
|
|
c.align(kAlignData, 16);
|
|
c.bind(L_Data);
|
|
c.dxmm(Vec128::fromSw(0x0080));
|
|
c.dxmm(Vec128::fromSw(0x0101));
|
|
}
|
|
|
|
} // asmgen namespace
|
|
|
|
// [Guard]
|
|
#endif // _TEST_GENBLEND_H
|