using the roundss SSE4.1 instruction for inlining floor and ceil syscalls when possible

fixed sharedTraps_t listing syscalls that were not actually at the same index for all 3 VMs

changelog.txt

@@ -89,6 +89,8 @@ chg: r_textureMode and r_measureOverdraw were removed
 
 chg: r_speeds 1 now reports more precise timings, V-Sync status, GPU time and the selected back-end
 
+chg: using the roundss SSE4.1 instruction for inlining floor and ceil syscalls when possible
+
 chg: improved face and grid processing performance with SSE2
 
 chg: r_lightmap is now latched again

code/qcommon/g_public.h

@@ -227,6 +227,21 @@ typedef enum {
 	// 1.32
 	G_FS_SEEK,
 
+	G_MEMSET = 100,
+	G_MEMCPY,
+	G_STRNCPY,
+	G_SIN,
+	G_COS,
+	G_ATAN2,
+	G_SQRT,
+	G_MATRIXMULTIPLY,
+	G_ANGLEVECTORS,
+	G_PERPENDICULARVECTOR,
+	G_FLOOR,
+	G_CEIL,
+	G_TESTPRINTINT,
+	G_TESTPRINTFLOAT,
+
 	BOTLIB_SETUP = 200,				// ( void );
 	BOTLIB_SHUTDOWN,				// ( void );
 	BOTLIB_LIBVAR_SET,

code/qcommon/qcommon.h

@@ -270,15 +270,8 @@ typedef enum {
 	TRAP_SIN,
 	TRAP_COS,
 	TRAP_ATAN2,
-	TRAP_SQRT,
+	TRAP_SQRT
-	TRAP_MATRIXMULTIPLY,
+	// note that ceil/floor etc have different numbers across VMs
-	TRAP_ANGLEVECTORS,
-	TRAP_PERPENDICULARVECTOR,
-	TRAP_FLOOR,
-	TRAP_CEIL,
-
-	TRAP_TESTPRINTINT,
-	TRAP_TESTPRINTFLOAT
 } sharedTraps_t;
 
 typedef enum {

code/qcommon/vm_x86.cpp

@@ -863,6 +863,24 @@ static void EmitDATAFunc(vm_t *vm)
 }
 
 
+static qbool IsFloorTrap(vm_t *vm, int trap)
+{
+	if ( vm->index == VM_CGAME || vm->index == VM_UI )
+		return trap == ~107;
+
+	return trap == ~110; // VM_GAME
+}
+
+
+static qbool IsCeilTrap(vm_t *vm, int trap)
+{
+	if ( vm->index == VM_CGAME || vm->index == VM_UI )
+		return trap == ~108;
+
+	return trap == ~111; // VM_GAME
+}
+
+
 /*
 =================
 ConstOptimize
@@ -1114,6 +1132,20 @@ static qboolean ConstOptimize(vm_t *vm)
 			EmitCommand( LAST_COMMAND_STORE_FLOAT_EDI );
 			ip += 1;
 			return qtrue;
+		} else if ( IsFloorTrap( vm, v ) && ( cpu_features & CPU_SSE41 ) != 0 ) {
+			EmitString( "f3 0f 10 45 08" );		// movss xmm0, dword ptr [ebp + 8]
+			EmitAddEDI4( vm );
+			EmitString( "66 0f 3a 0a c0 01" );	// roundss xmm0, xmm0, 1 (exceptions not masked)
+			EmitCommand( LAST_COMMAND_STORE_FLOAT_EDI );
+			ip += 1;
+			return qtrue;
+		} else if ( IsCeilTrap( vm, v ) && ( cpu_features & CPU_SSE41 ) != 0 ) {
+			EmitString( "f3 0f 10 45 08" );		// movss xmm0, dword ptr [ebp + 8]
+			EmitAddEDI4( vm );
+			EmitString( "66 0f 3a 0a c0 02" );	// roundss xmm0, xmm0, 2 (exceptions not masked)
+			EmitCommand( LAST_COMMAND_STORE_FLOAT_EDI );
+			ip += 1;
+			return qtrue;
 		}
 
 		if ( v < 0 ) // syscall

code/server/sv_game.cpp

@@ -764,46 +764,46 @@ static intptr_t SV_GameSystemCalls( intptr_t* args )
 	case BOTLIB_AI_GENETIC_PARENTS_AND_CHILD_SELECTION:
 		return botlib_export->ai.GeneticParentsAndChildSelection(args[1], VMA(2), VMA(3), VMA(4), VMA(5));
 
-	case TRAP_MEMSET:
+	case G_MEMSET:
 		Com_Memset( VMA(1), args[2], args[3] );
 		return 0;
 
-	case TRAP_MEMCPY:
+	case G_MEMCPY:
 		Com_Memcpy( VMA(1), VMA(2), args[3] );
 		return 0;
 
-	case TRAP_STRNCPY:
+	case G_STRNCPY:
 		strncpy( VMA(1), VMA(2), args[3] );
 		return args[1];
 
-	case TRAP_SIN:
+	case G_SIN:
 		return PASSFLOAT( sin( VMF(1) ) );
 
-	case TRAP_COS:
+	case G_COS:
 		return PASSFLOAT( cos( VMF(1) ) );
 
-	case TRAP_ATAN2:
+	case G_ATAN2:
 		return PASSFLOAT( atan2( VMF(1), VMF(2) ) );
 
-	case TRAP_SQRT:
+	case G_SQRT:
 		return PASSFLOAT( sqrt( VMF(1) ) );
 
-	case TRAP_MATRIXMULTIPLY:
+	case G_MATRIXMULTIPLY:
 		MatrixMultiply( VMA(1), VMA(2), VMA(3) );
 		return 0;
 
-	case TRAP_ANGLEVECTORS:
+	case G_ANGLEVECTORS:
 		AngleVectors( VMA(1), VMA(2), VMA(3), VMA(4) );
 		return 0;
 
-	case TRAP_PERPENDICULARVECTOR:
+	case G_PERPENDICULARVECTOR:
 		PerpendicularVector( VMA(1), VMA(2) );
 		return 0;
 
-	case TRAP_FLOOR:
+	case G_FLOOR:
 		return PASSFLOAT( floor( VMF(1) ) );
 
-	case TRAP_CEIL:
+	case G_CEIL:
 		return PASSFLOAT( ceil( VMF(1) ) );
 
 	// extensions