rpcsx/rpcs3/Emu/RSX/Program/program_state_cache2.hpp

#pragma once

#include "ProgramStateCache.h"

#include "emmintrin.h"
#include "util/asm.hpp"

template <typename Traits>
void program_state_cache<Traits>::fill_fragment_constants_buffer(std::span<f32> dst_buffer, const RSXFragmentProgram &fragment_program, bool sanitize) const
{
	const auto I = m_fragment_shader_cache.find(fragment_program);
	if (I == m_fragment_shader_cache.end())
		return;

	ensure((dst_buffer.size_bytes() >= ::narrow<int>(I->second.FragmentConstantOffsetCache.size()) * 16u));

	f32* dst = dst_buffer.data();
	alignas(16) f32 tmp[4];
	for (usz offset_in_fragment_program : I->second.FragmentConstantOffsetCache)
	{
		char* data = static_cast<char*>(fragment_program.get_data()) + offset_in_fragment_program;
		const __m128i vector = _mm_loadu_si128(reinterpret_cast<__m128i*>(data));
		const __m128i shuffled_vector = _mm_or_si128(_mm_slli_epi16(vector, 8), _mm_srli_epi16(vector, 8));

		if (!patch_table.is_empty())
		{
			_mm_store_ps(tmp, _mm_castsi128_ps(shuffled_vector));

			for (int i = 0; i < 4; ++i)
			{
				bool patched = false;
				for (auto& e : patch_table.db)
				{
					//TODO: Use fp comparison with fabsf without hurting performance
					patched = e.second.test_and_set(tmp[i], &dst[i]);
					if (patched)
					{
						break;
					}
				}

				if (!patched)
				{
					dst[i] = tmp[i];
				}
			}
		}
		else if (sanitize)
		{
			//Convert NaNs and Infs to 0
			const auto masked = _mm_and_si128(shuffled_vector, _mm_set1_epi32(0x7fffffff));
			const auto valid = _mm_cmplt_epi32(masked, _mm_set1_epi32(0x7f800000));
			const auto result = _mm_and_si128(shuffled_vector, valid);
			_mm_stream_si128(utils::bless<__m128i>(dst), result);
		}
		else
		{
			_mm_stream_si128(utils::bless<__m128i>(dst), shuffled_vector);
		}

		dst += 4;
	}
}