diff --git a/common/sse2simde.h b/common/sse2simde.h
new file mode 100644
index 0000000000..20a4ebebc5
--- /dev/null
+++ b/common/sse2simde.h
@@ -0,0 +1,22 @@
+//========= Copyright Valve Corporation, All rights reserved. ============//
+//
+// Purpose: Provide the SSE intrinsics API on non-x86 / non-ARM targets
+//          (e.g. RISC-V) via SIMDe (SIMD Everywhere, https://github.com/simd-everywhere/simde).
+//
+//          This is the counterpart of sse2neon.h for ARM: code written against
+//          <xmmintrin.h>/<emmintrin.h> keeps working, and SIMDe maps it to the
+//          native vector extension where it can (RVV on RISC-V) or to portable C.
+//
+//          Requires the SIMDe headers (Debian/Ubuntu: libsimde-dev, Fedora: simde-devel).
+//
+//===========================================================================//
+#ifndef SSE2SIMDE_H
+#define SSE2SIMDE_H
+
+#ifndef SIMDE_ENABLE_NATIVE_ALIASES
+#define SIMDE_ENABLE_NATIVE_ALIASES
+#endif
+
+#include <simde/x86/sse4.1.h>
+
+#endif // SSE2SIMDE_H
diff --git a/mathlib/sse.cpp b/mathlib/sse.cpp
index 6122b66487..7587408e84 100644
--- a/mathlib/sse.cpp
+++ b/mathlib/sse.cpp
@@ -13,6 +13,8 @@
 #include "mathlib/vector.h"
 #if defined(__arm__) || defined(__aarch64__)
 #include "sse2neon.h"
+#elif defined(__riscv)
+#include "sse2simde.h"
 #endif
 
 #include "sse.h"
@@ -180,7 +182,7 @@ float _SSE_RSqrtFast(float x)
 	Assert( s_bMathlibInitialized );
 
 	float rroot;
-#if defined(__arm__) || defined(__aarch64__)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv)
         rroot = _SSE_RSqrtAccurate(x);
 #elif _WIN32
 	_asm
@@ -217,7 +219,7 @@ float FASTCALL _SSE_VectorNormalize (Vector& vec)
 	// be much of a performance win, considering you will very likely miss 3 branch predicts in a row.
 	if ( v[0] || v[1] || v[2] )
 	{
-#if defined(__arm__) || defined(__aarch64__)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv)
 		float rsqrt = _SSE_RSqrtAccurate( v[0] * v[0] + v[1] * v[1] + v[2] * v[2] );
 		r[0] = v[0] * rsqrt;
 		r[1] = v[1] * rsqrt;
@@ -296,7 +298,7 @@ void FASTCALL _SSE_VectorNormalizeFast (Vector& vec)
 float _SSE_InvRSquared(const float* v)
 {
 	float	inv_r2 = 1.f;
-#if defined(__arm__) || defined(__aarch64__)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv)
 	return _SSE_RSqrtAccurate( FLT_EPSILON + v[0] * v[0] + v[1] * v[1] + v[2] * v[2] );
 #elif _WIN32
 	_asm { // Intel SSE only routine
@@ -391,7 +393,7 @@ typedef __m64 v2si;   // vector of 2 int (mmx)
 
 void _SSE_SinCos(float x, float* s, float* c)
 {
-#if defined(__arm__) || defined(__aarch64__)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv)
 #if defined( OSX )
     __sincosf(x, s, c);
 #elif defined( POSIX )
@@ -607,7 +609,7 @@ void _SSE_SinCos(float x, float* s, float* c)
 
 float _SSE_cos( float x )
 {
-#if defined(__arm__) || defined(__aarch64__)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv)
 	return cos(x);
 #elif _WIN32
 	float temp;
diff --git a/public/materialsystem/imesh.h b/public/materialsystem/imesh.h
index 6a952e0099..e853a79032 100644
--- a/public/materialsystem/imesh.h
+++ b/public/materialsystem/imesh.h
@@ -1152,7 +1152,7 @@ inline void CVertexBuilder::FastAdvanceNVertices( int n )
 //-----------------------------------------------------------------------------
 inline void CVertexBuilder::FastVertex( const ModelVertexDX7_t &vertex )
 {
-#if defined(__arm__) || defined(__aarch64__) || defined(PLATFORM_WINDOWS_PC64)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv) || defined(PLATFORM_WINDOWS_PC64)
 	FastVertexSSE( vertex );
 #else
 	Assert( m_CompressionType == VERTEX_COMPRESSION_NONE ); // FIXME: support compressed verts if needed
@@ -1309,7 +1309,7 @@ inline void CVertexBuilder::Fast4VerticesSSE(
 			movntps [edi + 80+96], xmm5
 
 	}
-#elif defined(__arm__) || defined(PLATFORM_WINDOWS_PC64)
+#elif defined(__arm__) || defined(__riscv) || defined(PLATFORM_WINDOWS_PC64)
 	const void *pReadA = &vtx_a;
 	const void *pReadB = &vtx_b;
 	const void *pReadC = &vtx_c;
@@ -1354,7 +1354,7 @@ inline void CVertexBuilder::Fast4VerticesSSE(
 
 inline void CVertexBuilder::FastVertex( const ModelVertexDX8_t &vertex )
 {
-#if defined(__arm__) || defined(__aarch64__) || defined(PLATFORM_WINDOWS_PC64)
+#if defined(__arm__) || defined(__aarch64__) || defined(__riscv) || defined(PLATFORM_WINDOWS_PC64)
 	FastVertexSSE( vertex );
 #else
 	Assert( m_CompressionType == VERTEX_COMPRESSION_NONE ); // FIXME: support compressed verts if needed
diff --git a/public/mathlib/mathlib.h b/public/mathlib/mathlib.h
index 1e8e0266ea..da8e4d5262 100644
--- a/public/mathlib/mathlib.h
+++ b/public/mathlib/mathlib.h
@@ -1203,6 +1203,8 @@ FORCEINLINE int RoundFloatToInt(float f)
 	return pResult[1];
 #elif defined (__arm__) ||  defined (__aarch64__)
         return (int)(f + 0.5f);
+#elif defined( PLATFORM_RISCV64 )
+	return (int)lrintf( f );	// round to nearest (even), like cvtss2si on x86
 #else
 #error Unknown architecture
 #endif
@@ -1235,6 +1237,8 @@ FORCEINLINE unsigned long RoundFloatToUnsignedLong(float f)
 #else  // !X360
 #if defined(__arm__) || defined(__aarch64__)
         return (unsigned long)(f + 0.5f);
+#elif defined( PLATFORM_RISCV64 )
+	return (unsigned long)llrintf( f );
 #elif defined( PLATFORM_WINDOWS_PC64 )
 	uint nRet = ( uint ) f;
 	if ( nRet & 1 )
diff --git a/public/mathlib/ssemath.h b/public/mathlib/ssemath.h
index 541419950f..98566a86e1 100644
--- a/public/mathlib/ssemath.h
+++ b/public/mathlib/ssemath.h
@@ -10,6 +10,8 @@
 #include <xboxmath.h>
 #elif defined(__arm__) || defined(__aarch64__)
 #include "sse2neon.h"
+#elif defined(__riscv)
+#include "sse2simde.h"
 #else
 #include <xmmintrin.h>
 #endif
diff --git a/public/mathlib/vector4d.h b/public/mathlib/vector4d.h
index 99e0995ff3..f3c68c7a8a 100644
--- a/public/mathlib/vector4d.h
+++ b/public/mathlib/vector4d.h
@@ -25,6 +25,8 @@
 
 #if defined (__arm__) || defined(__aarch64__)
 #include "sse2neon.h"
+#elif defined(__riscv)
+#include "sse2simde.h"
 #endif
 
 // forward declarations
diff --git a/public/saverestoretypes.h b/public/saverestoretypes.h
index e470f662c9..fec300e254 100644
--- a/public/saverestoretypes.h
+++ b/public/saverestoretypes.h
@@ -512,7 +512,7 @@ inline const char *CSaveRestoreSegment::StringFromSymbol( int token )
 ///             compilers. Either way, there's no portable intrinsic.
 
 // Newer GCC versions provide this in this header, older did by default.
-#if !defined( _rotr ) && defined( COMPILER_GCC ) && !defined( __arm__ ) && !defined( __aarch64__ )
+#if !defined( _rotr ) && defined( COMPILER_GCC ) && !defined( __arm__ ) && !defined( __aarch64__ ) && !defined( __riscv )
 #include <x86intrin.h>
 #endif
 
diff --git a/public/steam/steamtypes.h b/public/steam/steamtypes.h
index c32150e2b1..a443031166 100644
--- a/public/steam/steamtypes.h
+++ b/public/steam/steamtypes.h
@@ -24,7 +24,7 @@ typedef unsigned char uint8;
 	#define POSIX 1
 #endif
 
-#if defined(__x86_64__) || defined(_WIN64) || defined(__aarch64__)
+#if defined(__x86_64__) || defined(_WIN64) || defined(__aarch64__) || (defined(__riscv) && __riscv_xlen == 64)
 #define X64BITS
 #endif
 
diff --git a/public/tier0/platform.h b/public/tier0/platform.h
index 73047e1bfb..ea14c1a57a 100644
--- a/public/tier0/platform.h
+++ b/public/tier0/platform.h
@@ -9,7 +9,11 @@
 #ifndef PLATFORM_H
 #define PLATFORM_H
 
-#if defined(__x86_64__) || defined(_WIN64) || defined(__aarch64__)
+#if defined(__riscv) && (__riscv_xlen == 64)
+#define PLATFORM_RISCV64 1
+#endif
+
+#if defined(__x86_64__) || defined(_WIN64) || defined(__aarch64__) || defined(PLATFORM_RISCV64)
 #define PLATFORM_64BITS 1
 #endif
 
@@ -455,7 +459,7 @@ typedef void * HINSTANCE;
 	// On OSX, SIGTRAP doesn't really stop the thread cold when debugging.
 	// So if being debugged, use INT3 which is precise.
 #if defined(OSX) || defined(PLATFORM_BSD)
-# if defined(__arm__) || defined(__aarch64__)
+# if defined(__arm__) || defined(__aarch64__) || defined(PLATFORM_RISCV64)
 #  ifdef __clang__
 #   define DebuggerBreak()  do { if ( Plat_IsInDebugSession() ) { __builtin_debugtrap(); } else { raise(SIGTRAP); } } while(0)
 #  elif defined __GNUC__
@@ -898,7 +902,7 @@ static FORCEINLINE double fsel(double fComparand, double fValGE, double fLT)
 
 		#endif
 	#endif
-#elif defined (__arm__) || defined (__aarch64__)
+#elif defined (__arm__) || defined (__aarch64__) || defined(PLATFORM_RISCV64)
 	inline void SetupFPUControlWord() {}
 #else
 	inline void SetupFPUControlWord()
@@ -1069,7 +1073,7 @@ inline T QWordSwapC( T dw )
 // The typically used methods.
 //-------------------------------------
 
-#if (defined(__i386__) || defined(__amd64__) || defined(__arm__) || defined(__aarch64__)) && !defined(VALVE_LITTLE_ENDIAN)
+#if (defined(__i386__) || defined(__amd64__) || defined(__arm__) || defined(__aarch64__) || defined(PLATFORM_RISCV64)) && !defined(VALVE_LITTLE_ENDIAN)
 #define VALVE_LITTLE_ENDIAN 1
 #endif
 
@@ -1235,7 +1239,7 @@ PLATFORM_INTERFACE struct tm *		Plat_localtime( const time_t *timep, struct tm *
 
 inline uint64 Plat_Rdtsc()
 {
-#if (defined( __arm__ ) || defined( __aarch64__ )) && defined (POSIX)
+#if (defined( __arm__ ) || defined( __aarch64__ ) || defined( PLATFORM_RISCV64 )) && defined (POSIX)
 	struct timespec t;
 	clock_gettime( CLOCK_REALTIME, &t);
 	return t.tv_sec * 1000000000ULL + t.tv_nsec;
diff --git a/public/tier0/tslist.h b/public/tier0/tslist.h
index f5661258a9..f6bf3a8b88 100644
--- a/public/tier0/tslist.h
+++ b/public/tier0/tslist.h
@@ -53,7 +53,13 @@ inline bool ThreadInterlockedAssignIf128( int128 volatile * pDest, const int128
     // We do not want the original comparand modified by the swap
     // so operate on a local copy.
     int128 local_comparand = comparand;
+#if defined( __riscv )
+	// RISC-V (without Zacas) has no 128-bit CAS instruction, so there is no
+	// __sync_bool_compare_and_swap_16. libatomic provides a (lock-based) 16-byte CAS.
+	return __atomic_compare_exchange_n( pDest, &local_comparand, value, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST );
+#else
 	return __sync_bool_compare_and_swap( pDest, local_comparand, value );
+#endif
 }
 #endif
 
diff --git a/scripts/build-debian-riscv64.sh b/scripts/build-debian-riscv64.sh
new file mode 100755
index 0000000000..298d89a2b5
--- /dev/null
+++ b/scripts/build-debian-riscv64.sh
@@ -0,0 +1,9 @@
+#!/bin/sh
+# Native build on Linux riscv64 (RV64GC), e.g. Debian 13 (trixie).
+
+git submodule init && git submodule update
+sudo apt-get update
+sudo apt-get install -y g++ python-is-python3 pkg-config libsimde-dev libopenal-dev libpng-dev libjpeg-dev libfreetype-dev libfontconfig-dev libcurl4-gnutls-dev libsdl2-dev zlib1g-dev libbz2-dev libedit-dev libgl-dev
+
+./waf configure -T debug --disable-warns $* &&
+./waf build
diff --git a/tier0/cpu.cpp b/tier0/cpu.cpp
index 0a1d84be0c..16ea7791d0 100644
--- a/tier0/cpu.cpp
+++ b/tier0/cpu.cpp
@@ -22,7 +22,7 @@ const tchar* GetProcessorVendorId();
 
 static bool cpuid(uint32 function, uint32& out_eax, uint32& out_ebx, uint32& out_ecx, uint32& out_edx)
 {
-#if defined (__arm__) || defined (__aarch64__) || defined( _X360 )
+#if defined (__arm__) || defined (__aarch64__) || defined( __riscv ) || defined( _X360 )
 	return false;
 #elif defined(GNUC)
 
@@ -337,6 +337,8 @@ const tchar* GetProcessorVendorId()
 	return "PPC";
 #elif defined ( __arm__ )
 	return "ARM";
+#elif defined ( __riscv )
+	return "RISC-V";
 #else
 	uint32 unused, VendorIDRegisters[3];
 
@@ -376,6 +378,8 @@ const tchar* GetProcessorArchName()
         return "aarch64";
 #elif defined __arm__ || defined _M_ARM
         return "arm";
+#elif defined(__riscv) && (__riscv_xlen == 64)
+	return "riscv64";
 #else
 #error "Unknown architecture"
 #endif
diff --git a/tier0/cpu_posix.cpp b/tier0/cpu_posix.cpp
index a86f40a77b..359cc7f05c 100644
--- a/tier0/cpu_posix.cpp
+++ b/tier0/cpu_posix.cpp
@@ -124,7 +124,7 @@ uint64 CalculateCPUFreq()
 		}
 	}
 
-#if !defined(__arm__) && !defined(__aarch64__)
+#if !defined(__arm__) && !defined(__aarch64__) && !defined(__riscv)
 	// fallback mechanism to calculate when failed
 	// Compute the period. Loop until we get 3 consecutive periods that
 	// are the same to within a small error. The error is chosen
diff --git a/tier1/pathmatch.cpp b/tier1/pathmatch.cpp
index 8e439dfe7d..f545ed3772 100644
--- a/tier1/pathmatch.cpp
+++ b/tier1/pathmatch.cpp
@@ -71,7 +71,7 @@ static bool s_bShowDiag;
 #ifdef POSIX
 #include <signal.h>
 #define DEBUG_BREAK() raise(SIGINT)
-#elif !defined (__arm__)
+#elif !defined (__arm__) && !defined(__aarch64__) && !defined(__riscv)
 #define DEBUG_BREAK() __asm__ __volatile__ ( "int $3" )
 #else
 #define DEBUG_BREAK() 
diff --git a/tier1/processor_detect_linux.cpp b/tier1/processor_detect_linux.cpp
index 64d771a248..472f1ab96a 100644
--- a/tier1/processor_detect_linux.cpp
+++ b/tier1/processor_detect_linux.cpp
@@ -13,7 +13,7 @@ bool CheckMMXTechnology(void) { return false; }
 bool CheckSSETechnology(void) { return false; }
 bool CheckSSE2Technology(void) { return false; }
 bool Check3DNowTechnology(void) { return false; }
-#elif defined (__arm__) || defined (__aarch64__)
+#elif defined (__arm__) || defined (__aarch64__) || defined(__riscv)
 bool CheckMMXTechnology(void) { return false; }
 bool CheckSSETechnology(void) { return false; }
 bool CheckSSE2Technology(void) { return false; }
diff --git a/tier1/reliabletimer.cpp b/tier1/reliabletimer.cpp
index f575a67982..552c8ea49b 100644
--- a/tier1/reliabletimer.cpp
+++ b/tier1/reliabletimer.cpp
@@ -87,7 +87,7 @@ int64 CReliableTimer::GetPerformanceCountNow()
 	uint64 ulNow;
 	SYS_TIMEBASE_GET( ulNow );
 	return ulNow;
-#elif (defined( __arm__ ) || defined( __aarch64__ )) && defined (POSIX)
+#elif (defined( __arm__ ) || defined( __aarch64__ ) || defined( __riscv )) && defined (POSIX)
 	struct timespec ts;
 	clock_gettime(CLOCK_REALTIME, &ts);
 	return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
diff --git a/wscript b/wscript
index 16f2263bd2..5636db97fd 100644
--- a/wscript
+++ b/wscript
@@ -523,7 +523,8 @@ def configure(conf):
 
 	if conf.env.DEST_CPU in ['x86', 'x86_64']:
 		flags += ['-mfpmath=sse']
-	elif conf.env.DEST_CPU in ['arm', 'aarch64']:
+	elif conf.env.DEST_CPU in ['arm', 'aarch64', 'riscv']:
+		# char is unsigned by default on ARM and RISC-V, the code assumes x86 (signed)
 		flags += ['-fsigned-char']
 
 	if conf.env.DEST_CPU == 'arm':
@@ -532,6 +533,11 @@ def configure(conf):
 	if conf.env.DEST_OS == 'freebsd':
 		linkflags += ['-lexecinfo']
 
+	if conf.env.DEST_CPU == 'riscv':
+		# 16-byte CAS (tier0/tslist.h) comes from libatomic on RISC-V.
+		# Added as LIB (not a linker flag) so it comes after the objects (--as-needed).
+		conf.env.append_unique('LIB', ['atomic'])
+
 	if conf.env.DEST_OS != 'win32':
 		cflags += flags
 		linkflags += flags