GS/HW: Use texel fetch/load instead of sampler for palettes

Saves a multiply in the shaders in some cases, and frees up one
descriptor slot.

bin/resources/shaders/dx11/tfx.fx

@@ -130,7 +130,6 @@ Texture2D<float4> Palette : register(t1);
 Texture2D<float4> RtTexture : register(t2);
 Texture2D<float> PrimMinTexture : register(t3);
 SamplerState TextureSampler : register(s0);
-SamplerState PaletteSampler : register(s1);
 
 #ifdef DX12
 cbuffer cb0 : register(b0)
@@ -207,9 +206,14 @@ float4 sample_c(float2 uv, float uv_w)
 #endif
 }
 
-float4 sample_p(float u)
+float4 sample_p(uint u)
 {
-	return Palette.Sample(PaletteSampler, u);
+	return Palette.Load(int3(int(u), 0, 0));
+}
+
+float4 sample_p_norm(float u)
+{
+	return sample_p(uint(u * 255.5f));
 }
 
 float4 clamp_wrap_uv(float4 uv)
@@ -278,7 +282,7 @@ float4x4 sample_4c(float4 uv, float uv_w)
 	return c;
 }
 
-float4 sample_4_index(float4 uv, float uv_w)
+uint4 sample_4_index(float4 uv, float uv_w)
 {
 	float4 c;
 
@@ -288,25 +292,26 @@ float4 sample_4_index(float4 uv, float uv_w)
 	c.w = sample_c(uv.zw, uv_w).a;
 
 	// Denormalize value
-	uint4 i = uint4(c * 255.0f + 0.5f);
+	uint4 i = uint4(c * 255.5f);
 
 	if (PS_PAL_FMT == 1)
 	{
 		// 4HL
-		c = float4(i & 0xFu) / 255.0f;
+		return i & 0xFu;
 	}
 	else if (PS_PAL_FMT == 2)
 	{
 		// 4HH
-		c = float4(i >> 4u) / 255.0f;
+		return i >> 4u;
+	}
+	else
+	{
+		// 8
+		return i;
+	}
 }
 
-	// Most of texture will hit this code so keep normalized float value
+float4x4 sample_4p(uint4 u)
-	// 8 bits
-	return c * 255./256 + 0.5/256;
-}
-
-float4x4 sample_4p(float4 u)
 {
 	float4x4 c;
 
@@ -468,7 +473,7 @@ float4 fetch_red(int2 xy)
 		rt = fetch_raw_color(xy);
 	}
 
-	return sample_p(rt.r) * 255.0f;
+	return sample_p_norm(rt.r) * 255.0f;
 }
 
 float4 fetch_green(int2 xy)
@@ -485,7 +490,7 @@ float4 fetch_green(int2 xy)
 		rt = fetch_raw_color(xy);
 	}
 
-	return sample_p(rt.g) * 255.0f;
+	return sample_p_norm(rt.g) * 255.0f;
 }
 
 float4 fetch_blue(int2 xy)
@@ -502,19 +507,19 @@ float4 fetch_blue(int2 xy)
 		rt = fetch_raw_color(xy);
 	}
 
-	return sample_p(rt.b) * 255.0f;
+	return sample_p_norm(rt.b) * 255.0f;
 }
 
 float4 fetch_alpha(int2 xy)
 {
 	float4 rt = fetch_raw_color(xy);
-	return sample_p(rt.a) * 255.0f;
+	return sample_p_norm(rt.a) * 255.0f;
 }
 
 float4 fetch_rgb(int2 xy)
 {
 	float4 rt = fetch_raw_color(xy);
-	float4 c = float4(sample_p(rt.r).r, sample_p(rt.g).g, sample_p(rt.b).b, 1.0);
+	float4 c = float4(sample_p_norm(rt.r).r, sample_p_norm(rt.g).g, sample_p_norm(rt.b).b, 1.0);
 	return c * 255.0f;
 }
 

bin/resources/shaders/opengl/tfx_fs.glsl

@@ -133,9 +133,14 @@ vec4 sample_c(vec2 uv)
 #endif
 }
 
-vec4 sample_p(float idx)
+vec4 sample_p(uint idx)
 {
-    return texture(PaletteSampler, vec2(idx, 0.0f));
+	return texelFetch(PaletteSampler, ivec2(int(idx), 0), 0);
+}
+
+vec4 sample_p_norm(float u)
+{
+	return sample_p(uint(u * 255.5f));
 }
 
 vec4 clamp_wrap_uv(vec4 uv)
@@ -202,7 +207,7 @@ mat4 sample_4c(vec4 uv)
     return c;
 }
 
-vec4 sample_4_index(vec4 uv)
+uvec4 sample_4_index(vec4 uv)
 {
     vec4 c;
 
@@ -218,26 +223,22 @@ vec4 sample_4_index(vec4 uv)
     c.z = sample_c(uv.xw).a;
     c.w = sample_c(uv.zw).a;
 
-    uvec4 i = uvec4(c * 255.0f + 0.5f); // Denormalize value
+    uvec4 i = uvec4(c * 255.5f); // Denormalize value
 
 #if PS_PAL_FMT == 1
     // 4HL
-    return vec4(i & 0xFu) / 255.0f;
+    return i & 0xFu
-
 #elif PS_PAL_FMT == 2
     // 4HH
-    return vec4(i >> 4u) / 255.0f;
+    return i >> 4u;
-
 #else
-    // Most of texture will hit this code so keep normalized float value
+    // 8
-
+    return i;
-    // 8 bits
-    return c;
 #endif
 
 }
 
-mat4 sample_4p(vec4 u)
+mat4 sample_4p(uvec4 u)
 {
     mat4 c;
 
@@ -398,7 +399,7 @@ vec4 fetch_red()
 #else
     vec4 rt = fetch_raw_color();
 #endif
-    return sample_p(rt.r) * 255.0f;
+    return sample_p_norm(rt.r) * 255.0f;
 }
 
 vec4 fetch_green()
@@ -409,7 +410,7 @@ vec4 fetch_green()
 #else
     vec4 rt = fetch_raw_color();
 #endif
-    return sample_p(rt.g) * 255.0f;
+    return sample_p_norm(rt.g) * 255.0f;
 }
 
 vec4 fetch_blue()
@@ -420,19 +421,19 @@ vec4 fetch_blue()
 #else
     vec4 rt = fetch_raw_color();
 #endif
-    return sample_p(rt.b) * 255.0f;
+    return sample_p_norm(rt.b) * 255.0f;
 }
 
 vec4 fetch_alpha()
 {
     vec4 rt = fetch_raw_color();
-    return sample_p(rt.a) * 255.0f;
+    return sample_p_norm(rt.a) * 255.0f;
 }
 
 vec4 fetch_rgb()
 {
     vec4 rt = fetch_raw_color();
-    vec4 c = vec4(sample_p(rt.r).r, sample_p(rt.g).g, sample_p(rt.b).b, 1.0f);
+    vec4 c = vec4(sample_p_norm(rt.r).r, sample_p_norm(rt.g).g, sample_p_norm(rt.b).b, 1.0f);
     return c * 255.0f;
 }
 

bin/resources/shaders/vulkan/tfx.glsl

@@ -390,7 +390,7 @@ layout(location = 0) out vec4 o_col0;
 #endif
 
 layout(set = 1, binding = 0) uniform sampler2D Texture;
-layout(set = 1, binding = 1) uniform sampler2D Palette;
+layout(set = 1, binding = 1) uniform texture2D Palette;
 
 #if PS_FEEDBACK_LOOP_IS_NEEDED
 	#ifndef DISABLE_TEXTURE_BARRIER
@@ -443,9 +443,14 @@ vec4 sample_c(vec2 uv)
 #endif
 }
 
-vec4 sample_p(float u)
+vec4 sample_p(uint idx)
 {
-	return texture(Palette, vec2(u, 0.0f));
+	return texelFetch(Palette, ivec2(int(idx), 0), 0);
+}
+
+vec4 sample_p_norm(float u)
+{
+	return sample_p(uint(u * 255.5f));
 }
 
 vec4 clamp_wrap_uv(vec4 uv)
@@ -519,7 +524,7 @@ mat4 sample_4c(vec4 uv)
 	return c;
 }
 
-vec4 sample_4_index(vec4 uv)
+uvec4 sample_4_index(vec4 uv)
 {
 	vec4 c;
 
@@ -533,18 +538,17 @@ vec4 sample_4_index(vec4 uv)
 
 	#if PS_PAL_FMT == 1
 		// 4HL
-		c = vec4(i & 0xFu) / 255.0f;
+		c = i & 0xFu;
 	#elif PS_PAL_FMT == 2
 		// 4HH
-		c = vec4(i >> 4u) / 255.0f;
+		c = i >> 4u;
+	#else
+		// 8
+		return i;
 	#endif
-
-	// Most of texture will hit this code so keep normalized float value
-	// 8 bits
-	return c * 255./256 + 0.5/256;
 }
 
-mat4 sample_4p(vec4 u)
+mat4 sample_4p(uvec4 u)
 {
 	mat4 c;
 
@@ -709,7 +713,7 @@ vec4 fetch_red(ivec2 xy)
 		rt = fetch_raw_color(xy);
 	#endif
 
-	return sample_p(rt.r) * 255.0f;
+	return sample_p_norm(rt.r) * 255.0f;
 }
 
 vec4 fetch_green(ivec2 xy)
@@ -723,7 +727,7 @@ vec4 fetch_green(ivec2 xy)
 		rt = fetch_raw_color(xy);
 	#endif
 
-	return sample_p(rt.g) * 255.0f;
+	return sample_p_norm(rt.g) * 255.0f;
 }
 
 vec4 fetch_blue(ivec2 xy)
@@ -737,19 +741,19 @@ vec4 fetch_blue(ivec2 xy)
 		rt = fetch_raw_color(xy);
 	#endif
 
-	return sample_p(rt.b) * 255.0f;
+	return sample_p_norm(rt.b) * 255.0f;
 }
 
 vec4 fetch_alpha(ivec2 xy)
 {
 	vec4 rt = fetch_raw_color(xy);
-	return sample_p(rt.a) * 255.0f;
+	return sample_p_norm(rt.a) * 255.0f;
 }
 
 vec4 fetch_rgb(ivec2 xy)
 {
 	vec4 rt = fetch_raw_color(xy);
-	vec4 c = vec4(sample_p(rt.r).r, sample_p(rt.g).g, sample_p(rt.b).b, 1.0);
+	vec4 c = vec4(sample_p_norm(rt.r).r, sample_p_norm(rt.g).g, sample_p_norm(rt.b).b, 1.0);
 	return c * 255.0f;
 }
 

pcsx2/GS/Renderers/DX11/GSDevice11.cpp

@@ -682,7 +682,7 @@ void GSDevice11::StretchRect(GSTexture* sTex, const GSVector4& sRect, GSTexture*
 	// ps
 
 	PSSetShaderResources(sTex, nullptr);
-	PSSetSamplerState(linear ? m_convert.ln.get() : m_convert.pt.get(), nullptr);
+	PSSetSamplerState(linear ? m_convert.ln.get() : m_convert.pt.get());
 	PSSetShader(ps, ps_cb);
 
 	//
@@ -759,7 +759,7 @@ void GSDevice11::PresentRect(GSTexture* sTex, const GSVector4& sRect, GSTexture*
 	// ps
 
 	PSSetShaderResources(sTex, nullptr);
-	PSSetSamplerState(linear ? m_convert.ln.get() : m_convert.pt.get(), nullptr);
+	PSSetSamplerState(linear ? m_convert.ln.get() : m_convert.pt.get());
 	PSSetShader(m_present.ps[static_cast<u32>(shader)].get(), m_present.ps_cb.get());
 
 	//
@@ -958,7 +958,7 @@ void GSDevice11::SetupDATE(GSTexture* rt, GSTexture* ds, const GSVertexPT1* vert
 
 	// ps
 	PSSetShaderResources(rt, nullptr);
-	PSSetSamplerState(m_convert.pt.get(), nullptr);
+	PSSetSamplerState(m_convert.pt.get());
 	PSSetShader(m_convert.ps[static_cast<int>(datm ? ShaderConvert::DATM_1 : ShaderConvert::DATM_0)].get(), nullptr);
 
 	//
@@ -1184,10 +1184,9 @@ void GSDevice11::PSSetShaderResource(int i, GSTexture* sr)
 	m_state.ps_sr_views[i] = sr ? static_cast<ID3D11ShaderResourceView*>(*static_cast<GSTexture11*>(sr)) : nullptr;
 }
 
-void GSDevice11::PSSetSamplerState(ID3D11SamplerState* ss0, ID3D11SamplerState* ss1)
+void GSDevice11::PSSetSamplerState(ID3D11SamplerState* ss0)
 {
 	m_state.ps_ss[0] = ss0;
-	m_state.ps_ss[1] = ss1;
 }
 
 void GSDevice11::PSSetShader(ID3D11PixelShader* ps, ID3D11Buffer* ps_cb)

pcsx2/GS/Renderers/DX11/GSDevice11.h

@@ -110,7 +110,7 @@ public:
 
 private:
 	static constexpr u32 MAX_TEXTURES = 4;
-	static constexpr u32 MAX_SAMPLERS = 2;
+	static constexpr u32 MAX_SAMPLERS = 1;
 
 	int m_d3d_texsize;
 
@@ -221,7 +221,6 @@ private:
 	std::unordered_map<PSSelector, wil::com_ptr_nothrow<ID3D11PixelShader>, GSHWDrawConfig::PSSelectorHash> m_ps;
 	wil::com_ptr_nothrow<ID3D11Buffer> m_ps_cb;
 	std::unordered_map<u32, wil::com_ptr_nothrow<ID3D11SamplerState>> m_ps_ss;
-	wil::com_ptr_nothrow<ID3D11SamplerState> m_palette_ss;
 	std::unordered_map<u32, wil::com_ptr_nothrow<ID3D11DepthStencilState>> m_om_dss;
 	std::unordered_map<u32, wil::com_ptr_nothrow<ID3D11BlendState>> m_om_bs;
 	wil::com_ptr_nothrow<ID3D11RasterizerState> m_rs;
@@ -287,7 +286,7 @@ public:
 	void PSSetShaderResource(int i, GSTexture* sr);
 	void PSSetShader(ID3D11PixelShader* ps, ID3D11Buffer* ps_cb);
 	void PSUpdateShaderState();
-	void PSSetSamplerState(ID3D11SamplerState* ss0, ID3D11SamplerState* ss1);
+	void PSSetSamplerState(ID3D11SamplerState* ss0);
 
 	void OMSetDepthStencilState(ID3D11DepthStencilState* dss, u8 sref);
 	void OMSetBlendState(ID3D11BlendState* bs, float bf);

pcsx2/GS/Renderers/DX11/GSTextureFX11.cpp

@@ -43,24 +43,6 @@ bool GSDevice11::CreateTextureFX()
 
 	hr = m_dev->CreateBuffer(&bd, nullptr, m_ps_cb.put());
 
-	if (FAILED(hr))
-		return false;
-
-	D3D11_SAMPLER_DESC sd;
-
-	memset(&sd, 0, sizeof(sd));
-
-	sd.Filter = D3D11_FILTER_MIN_MAG_MIP_POINT;
-	sd.AddressU = D3D11_TEXTURE_ADDRESS_CLAMP;
-	sd.AddressV = D3D11_TEXTURE_ADDRESS_CLAMP;
-	sd.AddressW = D3D11_TEXTURE_ADDRESS_CLAMP;
-	sd.MinLOD = -FLT_MAX;
-	sd.MaxLOD = FLT_MAX;
-	sd.MaxAnisotropy = 1;
-	sd.ComparisonFunc = D3D11_COMPARISON_NEVER;
-
-	hr = m_dev->CreateSamplerState(&sd, m_palette_ss.put());
-
 	if (FAILED(hr))
 		return false;
 
@@ -212,7 +194,7 @@ void GSDevice11::SetupPS(const PSSelector& sel, const GSHWDrawConfig::PSConstant
 		m_ctx->UpdateSubresource(m_ps_cb.get(), 0, NULL, cb, 0, 0);
 	}
 
-	wil::com_ptr_nothrow<ID3D11SamplerState> ss0, ss1;
+	wil::com_ptr_nothrow<ID3D11SamplerState> ss0;
 
 	if (sel.tfx != 4)
 	{
@@ -267,14 +249,9 @@ void GSDevice11::SetupPS(const PSSelector& sel, const GSHWDrawConfig::PSConstant
 
 			m_ps_ss[ssel.key] = ss0;
 		}
-
-		if (sel.pal_fmt)
-		{
-			ss1 = m_palette_ss;
-		}
 	}
 
-	PSSetSamplerState(ss0.get(), ss1.get());
+	PSSetSamplerState(ss0.get());
 
 	PSSetShader(i->second.get(), m_ps_cb.get());
 }

pcsx2/GS/Renderers/DX12/GSDevice12.cpp

@@ -1143,7 +1143,7 @@ bool GSDevice12::CreateRootSignatures()
 	rsb.AddCBVParameter(0, D3D12_SHADER_VISIBILITY_ALL);
 	rsb.AddCBVParameter(1, D3D12_SHADER_VISIBILITY_PIXEL);
 	rsb.AddDescriptorTable(D3D12_DESCRIPTOR_RANGE_TYPE_SRV, 0, 2, D3D12_SHADER_VISIBILITY_PIXEL);
-	rsb.AddDescriptorTable(D3D12_DESCRIPTOR_RANGE_TYPE_SAMPLER, 0, 2, D3D12_SHADER_VISIBILITY_PIXEL);
+	rsb.AddDescriptorTable(D3D12_DESCRIPTOR_RANGE_TYPE_SAMPLER, 0, NUM_TFX_SAMPLERS, D3D12_SHADER_VISIBILITY_PIXEL);
 	rsb.AddDescriptorTable(D3D12_DESCRIPTOR_RANGE_TYPE_SRV, 2, 2, D3D12_SHADER_VISIBILITY_PIXEL);
 	if (!(m_tfx_root_signature = rsb.Create()))
 		return false;
@@ -1806,8 +1806,7 @@ void GSDevice12::InitializeState()
 {
 	for (u32 i = 0; i < NUM_TOTAL_TFX_TEXTURES; i++)
 		m_tfx_textures[i] = m_null_texture.GetSRVDescriptor();
-	for (u32 i = 0; i < NUM_TFX_SAMPLERS; i++)
+	m_tfx_sampler_sel = GSHWDrawConfig::SamplerSelector::Point().key;
-		m_tfx_sampler_sel[i] = GSHWDrawConfig::SamplerSelector::Point().key;
 
 	InvalidateCachedState();
 }
@@ -1816,9 +1815,7 @@ void GSDevice12::InitializeSamplers()
 {
 	bool result = GetSampler(&m_point_sampler_cpu, GSHWDrawConfig::SamplerSelector::Point());
 	result = result && GetSampler(&m_linear_sampler_cpu, GSHWDrawConfig::SamplerSelector::Linear());
-
+	result = result && GetSampler(&m_tfx_sampler, m_tfx_sampler_sel);
-	for (u32 i = 0; i < NUM_TFX_SAMPLERS; i++)
-		result = result && GetSampler(&m_tfx_samplers[i], m_tfx_sampler_sel[i]);
 
 	if (!result)
 		pxFailRel("Failed to initialize samplers");
@@ -1970,13 +1967,13 @@ void GSDevice12::PSSetShaderResource(int i, GSTexture* sr, bool check_state)
 	m_dirty_flags |= (i < 2) ? DIRTY_FLAG_TFX_TEXTURES : DIRTY_FLAG_TFX_RT_TEXTURES;
 }
 
-void GSDevice12::PSSetSampler(u32 index, GSHWDrawConfig::SamplerSelector sel)
+void GSDevice12::PSSetSampler(GSHWDrawConfig::SamplerSelector sel)
 {
-	if (m_tfx_sampler_sel[index] == sel.key)
+	if (m_tfx_sampler_sel == sel.key)
 		return;
 
-	GetSampler(&m_tfx_samplers[index], sel);
+	GetSampler(&m_tfx_sampler, sel);
-	m_tfx_sampler_sel[index] = sel.key;
+	m_tfx_sampler_sel = sel.key;
 	m_dirty_flags |= DIRTY_FLAG_TFX_SAMPLERS;
 }
 
@@ -2330,7 +2327,7 @@ bool GSDevice12::ApplyTFXState(bool already_execed)
 
 	if (flags & DIRTY_FLAG_TFX_SAMPLERS)
 	{
-		if (!g_d3d12_context->GetSamplerAllocator().LookupGroup(&m_tfx_samplers_handle_gpu, m_tfx_samplers.data()))
+		if (!g_d3d12_context->GetSamplerAllocator().LookupSingle(&m_tfx_samplers_handle_gpu, m_tfx_sampler))
 		{
 			ExecuteCommandListAndRestartRenderPass(false, "Ran out of sampler groups");
 			return ApplyTFXState(true);
@@ -2555,7 +2552,7 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 	if (config.tex)
 	{
 		PSSetShaderResource(0, config.tex, config.tex != config.rt);
-		PSSetSampler(0, config.sampler);
+		PSSetSampler(config.sampler);
 	}
 	if (config.pal)
 		PSSetShaderResource(1, config.pal, true);

pcsx2/GS/Renderers/DX12/GSDevice12.h

@@ -112,7 +112,7 @@ public:
 		NUM_TFX_TEXTURES = 2,
 		NUM_TFX_RT_TEXTURES = 2,
 		NUM_TOTAL_TFX_TEXTURES = NUM_TFX_TEXTURES + NUM_TFX_RT_TEXTURES,
-		NUM_TFX_SAMPLERS = 2,
+		NUM_TFX_SAMPLERS = 1,
 		NUM_UTILITY_TEXTURES = 1,
 		NUM_UTILITY_SAMPLERS = 1,
 		CONVERT_PUSH_CONSTANTS_SIZE = 96,
@@ -279,7 +279,7 @@ public:
 	void IASetIndexBuffer(const void* index, size_t count);
 
 	void PSSetShaderResource(int i, GSTexture* sr, bool check_state);
-	void PSSetSampler(u32 index, GSHWDrawConfig::SamplerSelector sel);
+	void PSSetSampler(GSHWDrawConfig::SamplerSelector sel);
 
 	void OMSetRenderTargets(GSTexture* rt, GSTexture* ds, const GSVector4i& scissor);
 
@@ -404,8 +404,8 @@ private:
 
 	std::array<D3D12_GPU_VIRTUAL_ADDRESS, NUM_TFX_CONSTANT_BUFFERS> m_tfx_constant_buffers{};
 	std::array<D3D12::DescriptorHandle, NUM_TOTAL_TFX_TEXTURES> m_tfx_textures{};
-	std::array<D3D12::DescriptorHandle, NUM_TFX_SAMPLERS> m_tfx_samplers{};
+	D3D12::DescriptorHandle m_tfx_sampler;
-	std::array<u32, NUM_TFX_SAMPLERS> m_tfx_sampler_sel{};
+	u32 m_tfx_sampler_sel = 0;
 	D3D12::DescriptorHandle m_tfx_textures_handle_gpu;
 	D3D12::DescriptorHandle m_tfx_samplers_handle_gpu;
 	D3D12::DescriptorHandle m_tfx_rt_textures_handle_gpu;

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -3244,7 +3244,7 @@ void GSTextureCache::Palette::InitializeTexture()
 		// sampling such texture are always normalized by 255.
 		// This is because indexes are stored as normalized values of an RGBA texture (e.g. index 15 will be read as (15/255),
 		// and therefore will read texel 15/255 * texture size).
-		m_tex_palette = g_gs_device->CreateTexture(256, 1, 1, GSTexture::Format::Color);
+		m_tex_palette = g_gs_device->CreateTexture(m_pal, 1, 1, GSTexture::Format::Color);
 		m_tex_palette->Update(GSVector4i(0, 0, m_pal, 1), m_clut, m_pal * sizeof(m_clut[0]));
 	}
 }

pcsx2/GS/Renderers/Metal/tfx.metal

@@ -283,8 +283,6 @@ vertex MainVSOut vs_main_expand(
 
 // MARK: - Fragment functions
 
-constexpr sampler palette_sampler(filter::nearest, address::clamp_to_edge);
-
 struct PSMain
 {
 	texture2d<float> tex;
@@ -349,9 +347,14 @@ struct PSMain
 		}
 	}
 
-	float4 sample_p(float idx)
+	float4 sample_p(uint idx)
 	{
-		return palette.sample(palette_sampler, float2(idx, 0));
+		return palette.read(uint2(idx, 0));
+	}
+
+	float4 sample_p_norm(float u)
+	{
+		return sample_p(uint(u * 255.5f));
 	}
 
 	float4 clamp_wrap_uv(float4 uv)
@@ -415,7 +418,7 @@ struct PSMain
 		};
 	}
 
-	float4 sample_4_index(float4 uv)
+	uint4 sample_4_index(float4 uv)
 	{
 		float4 c;
 
@@ -432,15 +435,14 @@ struct PSMain
 		uint4 i = uint4(c * 255.5f); // Denormalize value
 
 		if (PS_PAL_FMT == 1)
-			return float4(i & 0xF) / 255.f;
+			return i & 0xF;
 		if (PS_PAL_FMT == 2)
-			return float4(i >> 4) / 255.f;
+			return i >> 4;
 
-		// Most textures will hit this code so keep normalized float value
+		return i;
-		return c;
 	}
 
-	float4x4 sample_4p(float4 u)
+	float4x4 sample_4p(uint4 u)
 	{
 		return {
 			sample_p(u.x),
@@ -559,30 +561,30 @@ struct PSMain
 	float4 fetch_red()
 	{
 		float rt = PS_TEX_IS_DEPTH ? float(fetch_raw_depth() & 0xFF) / 255.f : fetch_raw_color().r;
-		return sample_p(rt) * 255.f;
+		return sample_p_norm(rt) * 255.f;
 	}
 
 	float4 fetch_green()
 	{
 		float rt = PS_TEX_IS_DEPTH ? float((fetch_raw_depth() >> 8) & 0xFF) / 255.f : fetch_raw_color().g;
-		return sample_p(rt) * 255.f;
+		return sample_p_norm(rt) * 255.f;
 	}
 
 	float4 fetch_blue()
 	{
 		float rt = PS_TEX_IS_DEPTH ? float((fetch_raw_depth() >> 16) & 0xFF) / 255.f : fetch_raw_color().b;
-		return sample_p(rt) * 255.f;
+		return sample_p_norm(rt) * 255.f;
 	}
 
 	float4 fetch_alpha()
 	{
-		return sample_p(fetch_raw_color().a) * 255.f;
+		return sample_p_norm(fetch_raw_color().a) * 255.f;
 	}
 
 	float4 fetch_rgb()
 	{
 		float4 rt = fetch_raw_color();
-		return float4(sample_p(rt.r).r, sample_p(rt.g).g, sample_p(rt.b).b, 1) * 255.f;
+		return float4(sample_p_norm(rt.r).r, sample_p_norm(rt.g).g, sample_p_norm(rt.b).b, 1) * 255.f;
 	}
 
 	float4 fetch_gXbY()

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp

@@ -1084,9 +1084,7 @@ void GSDeviceVK::ClearSamplerCache()
 	m_point_sampler = GetSampler(GSHWDrawConfig::SamplerSelector::Point());
 	m_linear_sampler = GetSampler(GSHWDrawConfig::SamplerSelector::Linear());
 	m_utility_sampler = m_point_sampler;
-
+	m_tfx_sampler = m_point_sampler;
-	for (u32 i = 0; i < std::size(m_tfx_samplers); i++)
-		m_tfx_samplers[i] = GetSampler(m_tfx_sampler_sel[i]);
 }
 
 static void AddMacro(std::stringstream& ss, const char* name, const char* value)
@@ -1236,8 +1234,8 @@ bool GSDeviceVK::CreatePipelineLayouts()
 	if ((m_tfx_ubo_ds_layout = dslb.Create(dev)) == VK_NULL_HANDLE)
 		return false;
 	Vulkan::Util::SetObjectName(dev, m_tfx_ubo_ds_layout, "TFX UBO descriptor layout");
-	for (u32 i = 0; i < NUM_TFX_SAMPLERS; i++)
+	dslb.AddBinding(0, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1, VK_SHADER_STAGE_FRAGMENT_BIT);
-		dslb.AddBinding(i, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1, VK_SHADER_STAGE_FRAGMENT_BIT);
+	dslb.AddBinding(1, VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1, VK_SHADER_STAGE_FRAGMENT_BIT);
 	if ((m_tfx_sampler_ds_layout = dslb.Create(dev)) == VK_NULL_HANDLE)
 		return false;
 	Vulkan::Util::SetObjectName(dev, m_tfx_sampler_ds_layout, "TFX sampler descriptor layout");
@@ -2308,11 +2306,8 @@ void GSDeviceVK::InitializeState()
 	if (m_linear_sampler)
 		Vulkan::Util::SetObjectName(g_vulkan_context->GetDevice(), m_point_sampler, "Linear sampler");
 
-	for (u32 i = 0; i < NUM_TFX_SAMPLERS; i++)
+	m_tfx_sampler_sel = GSHWDrawConfig::SamplerSelector::Point().key;
-	{
+	m_tfx_sampler = m_point_sampler;
-		m_tfx_sampler_sel[i] = GSHWDrawConfig::SamplerSelector::Point().key;
-		m_tfx_samplers[i] = m_point_sampler;
-	}
 
 	InvalidateCachedState();
 }
@@ -2463,13 +2458,13 @@ void GSDeviceVK::PSSetShaderResource(int i, GSTexture* sr, bool check_state)
 	m_dirty_flags |= (i < 2) ? DIRTY_FLAG_TFX_SAMPLERS_DS : DIRTY_FLAG_TFX_RT_TEXTURE_DS;
 }
 
-void GSDeviceVK::PSSetSampler(u32 index, GSHWDrawConfig::SamplerSelector sel)
+void GSDeviceVK::PSSetSampler(GSHWDrawConfig::SamplerSelector sel)
 {
-	if (m_tfx_sampler_sel[index] == sel.key)
+	if (m_tfx_sampler_sel == sel.key)
 		return;
 
-	m_tfx_sampler_sel[index] = sel.key;
+	m_tfx_sampler_sel = sel.key;
-	m_tfx_samplers[index] = GetSampler(sel);
+	m_tfx_sampler = GetSampler(sel);
 	m_dirty_flags |= DIRTY_FLAG_TFX_SAMPLERS_DS;
 }
 
@@ -2739,8 +2734,8 @@ bool GSDeviceVK::ApplyTFXState(bool already_execed)
 			return ApplyTFXState(true);
 		}
 
-		dsub.AddCombinedImageSamplerDescriptorWrites(
+		dsub.AddCombinedImageSamplerDescriptorWrite(ds, 0, m_tfx_textures[0], m_tfx_sampler);
-			ds, 0, m_tfx_textures.data(), m_tfx_samplers.data(), NUM_TFX_SAMPLERS);
+		dsub.AddImageDescriptorWrite(ds, 1, m_tfx_textures[1]);
 		dsub.Update(dev);
 
 		m_tfx_descriptor_sets[1] = ds;
@@ -2764,10 +2759,10 @@ bool GSDeviceVK::ApplyTFXState(bool already_execed)
 		}
 
 		if (m_features.texture_barrier)
-			dsub.AddInputAttachmentDescriptorWrite(ds, 0, m_tfx_textures[NUM_TFX_SAMPLERS]);
+			dsub.AddInputAttachmentDescriptorWrite(ds, 0, m_tfx_textures[NUM_TFX_DRAW_TEXTURES]);
 		else
-			dsub.AddImageDescriptorWrite(ds, 0, m_tfx_textures[NUM_TFX_SAMPLERS]);
+			dsub.AddImageDescriptorWrite(ds, 0, m_tfx_textures[NUM_TFX_DRAW_TEXTURES]);
-		dsub.AddImageDescriptorWrite(ds, 1, m_tfx_textures[NUM_TFX_SAMPLERS + 1]);
+		dsub.AddImageDescriptorWrite(ds, 1, m_tfx_textures[NUM_TFX_DRAW_TEXTURES + 1]);
 		dsub.Update(dev);
 
 		m_tfx_descriptor_sets[2] = ds;
@@ -3028,7 +3023,7 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 	if (config.tex)
 	{
 		PSSetShaderResource(0, config.tex, config.tex != config.rt);
-		PSSetSampler(0, config.sampler);
+		PSSetSampler(config.sampler);
 	}
 	if (config.pal)
 		PSSetShaderResource(1, config.pal, true);

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.h

@@ -72,9 +72,9 @@ public:
 	{
 		NUM_TFX_DESCRIPTOR_SETS = 3,
 		NUM_TFX_DYNAMIC_OFFSETS = 2,
-		NUM_TFX_SAMPLERS = 2,
+		NUM_TFX_DRAW_TEXTURES = 2,
 		NUM_TFX_RT_TEXTURES = 2,
-		NUM_TFX_TEXTURES = NUM_TFX_SAMPLERS + NUM_TFX_RT_TEXTURES,
+		NUM_TFX_TEXTURES = NUM_TFX_DRAW_TEXTURES + NUM_TFX_RT_TEXTURES,
 		NUM_CONVERT_TEXTURES = 1,
 		NUM_CONVERT_SAMPLERS = 1,
 		CONVERT_PUSH_CONSTANTS_SIZE = 96,
@@ -260,7 +260,7 @@ public:
 	void IASetIndexBuffer(const void* index, size_t count);
 
 	void PSSetShaderResource(int i, GSTexture* sr, bool check_state);
-	void PSSetSampler(u32 index, GSHWDrawConfig::SamplerSelector sel);
+	void PSSetSampler(GSHWDrawConfig::SamplerSelector sel);
 
 	void OMSetRenderTargets(GSTexture* rt, GSTexture* ds, const GSVector4i& scissor, bool feedback_loop);
 
@@ -372,8 +372,8 @@ private:
 	u8 m_blend_constant_color = 0;
 
 	std::array<VkImageView, NUM_TFX_TEXTURES> m_tfx_textures{};
-	std::array<VkSampler, NUM_TFX_SAMPLERS> m_tfx_samplers{};
+	VkSampler m_tfx_sampler = VK_NULL_HANDLE;
-	std::array<u32, NUM_TFX_SAMPLERS> m_tfx_sampler_sel{};
+	u32 m_tfx_sampler_sel = 0;
 	std::array<VkDescriptorSet, NUM_TFX_DESCRIPTOR_SETS> m_tfx_descriptor_sets{};
 	std::array<u32, NUM_TFX_DYNAMIC_OFFSETS> m_tfx_dynamic_offsets{};