diff --git a/rpcs3/Emu/RSX/VK/VKCompute.h b/rpcs3/Emu/RSX/VK/VKCompute.h
index 2801519668..7f6046996c 100644
--- a/rpcs3/Emu/RSX/VK/VKCompute.h
+++ b/rpcs3/Emu/RSX/VK/VKCompute.h
@@ -591,7 +591,7 @@ namespace vk
 		}
 	};
 
-	template<typename To, typename From, bool _SwapSrc = false, bool _SwapDst = false>
+	template<typename From, typename To, bool _SwapSrc = false, bool _SwapDst = false>
 	struct cs_fconvert_task : cs_shuffle_base
 	{
 		u32 m_ssbo_length = 0;
@@ -629,7 +629,7 @@ namespace vk
 				"	uint out_offset = params[0].z >> 2;\n"
 				"	uvec4 tmp;\n";
 
-			work_kernel +=
+			work_kernel =
 				"		if (index >= block_length)\n"
 				"			return;\n";
 
@@ -651,7 +651,7 @@ namespace vk
 				}
 
 				// Convert
-				work_kernel += "		tmp.z = pack_e4m12_pack16(tmp);\n";
+				work_kernel += "		tmp.z = pack_e4m12_pack16(tmp.xy);\n";
 
 				if constexpr (_SwapDst)
 				{
@@ -681,7 +681,7 @@ namespace vk
 
 				if constexpr (_SwapDst)
 				{
-					work_kernel += "		tmp.yz = bswap_u16(tmp.yz);\n";
+					work_kernel += "		tmp.yz = bswap_u32(tmp.yz);\n";
 				}
 
 				work_kernel +=
diff --git a/rpcs3/Emu/RSX/VK/VKTexture.cpp b/rpcs3/Emu/RSX/VK/VKTexture.cpp
index d68b89c584..cb8628854d 100644
--- a/rpcs3/Emu/RSX/VK/VKTexture.cpp
+++ b/rpcs3/Emu/RSX/VK/VKTexture.cpp
@@ -108,12 +108,12 @@ namespace vk
 			// 3. Do conversion with byteswap [D32->D16F]
 			if (!swap_bytes) [[likely]]
 			{
-				auto job = vk::get_compute_task<vk::cs_fconvert_task<u16, u32>>();
+				auto job = vk::get_compute_task<vk::cs_fconvert_task<f32, f16>>();
 				job->run(cmd, dst, z32_offset, packed32_length, data_offset);
 			}
 			else
 			{
-				auto job = vk::get_compute_task<vk::cs_fconvert_task<u16, u32, false, true>>();
+				auto job = vk::get_compute_task<vk::cs_fconvert_task<f32, f16, false, true>>();
 				job->run(cmd, dst, z32_offset, packed32_length, data_offset);
 			}
 
@@ -237,7 +237,7 @@ namespace vk
 				VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT);
 
 			// 2. Do conversion with byteswap [D16F->D32F]
-			auto job = vk::get_compute_task<vk::cs_fconvert_task<u32, u16>>();
+			auto job = vk::get_compute_task<vk::cs_fconvert_task<f16, f32>>();
 			job->run(cmd, src, data_offset, packed16_length, z32_offset);
 
 			// 4. Post-compute barrier