Unlike radeon and r200 drivers r300AllocCmdBuf allocs dwords not bytes.

6 files changed, 70 insertions, 23 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index 3fe69c3b2e..df825ba3b4 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -601,7 +601,7 @@ void r300EmitBlit(r300ContextPtr rmesa,
 	assert(h < (1 << 16));
 
 	cmd =
-	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 8 * sizeof(int),
+	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 8,
 							__FUNCTION__);
 
 	cmd[0].header.cmd_type = R300_CMD_PACKET3;
@@ -631,7 +631,7 @@ void r300EmitWait(r300ContextPtr rmesa, GLuint flags)
 
 		cmd =
 		    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa,
-								1 * sizeof(int),
+								1,
 								__FUNCTION__);
 		cmd[0].i = 0;
 		cmd[0].wait.cmd_type = R300_CMD_WAIT;
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index cd7b536592..da8497c4ee 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -267,12 +267,12 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 		_mesa_enable_extension( ctx, "GL_NV_vertex_program");
 #endif
 	/* currently bogus data */
-	ctx->Const.MaxVertexProgramInstructions=128;
-	ctx->Const.MaxVertexProgramAttribs=64;
-	ctx->Const.MaxVertexProgramTemps=64;
-	ctx->Const.MaxVertexProgramLocalParams=64;
-	ctx->Const.MaxVertexProgramEnvParams=64;
-	ctx->Const.MaxVertexProgramAddressRegs=8;
+	ctx->Const.MaxVertexProgramInstructions=VSF_MAX_FRAGMENT_LENGTH;
+	ctx->Const.MaxVertexProgramAttribs=16; // r420
+	ctx->Const.MaxVertexProgramTemps=VSF_MAX_FRAGMENT_TEMPS;
+	ctx->Const.MaxVertexProgramLocalParams=256; // r420
+	ctx->Const.MaxVertexProgramEnvParams=256; // r420
+	ctx->Const.MaxVertexProgramAddressRegs=1;
 		
 	driInitExtensions(ctx, card_extensions, GL_TRUE);
 	
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index 7879de1b88..d956848dd1 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -513,8 +513,11 @@ struct r300_vap_reg_state {
 
 /* Vertex shader state */
 
-/* 64 appears to be the maximum */
-#define VSF_MAX_FRAGMENT_LENGTH (64*4)
+/* Tested with rv350 and verified from misc web pages. */
+#define VSF_MAX_FRAGMENT_LENGTH (256*4)
+	
+/* Tested with rv350 and verified from misc web pages. */
+#define VSF_MAX_FRAGMENT_TEMPS (32)
 
 
 struct r300_vertex_shader_fragment {
diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h
index eb2a913f1c..6085ccf000 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.h
+++ b/src/mesa/drivers/dri/r300/r300_emit.h
@@ -200,13 +200,14 @@ static __inline__ uint32_t cmdpacify(void)
 	}
 
 	/* must be sent to switch to 2d commands */
+/*
 void static inline end_3d(PREFIX_VOID)
 {
 LOCAL_VARS
 (void)cmd_reserved; (void)cmd_written;
 
 cmd=(drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, \
-					0, \
+					0, \ // ??!
 					__FUNCTION__); \
 
 cmd[0].header.cmd_type=R300_CMD_END3D;
@@ -218,11 +219,12 @@ LOCAL_VARS
 (void)cmd_reserved; (void)cmd_written;
 
 cmd=(drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, \
-					0, \
+					0, \ // ??!
 					__FUNCTION__); \
 
 cmd[0].i=cmdcpdelay(count);
 }
+*/
 
 /* fire vertex buffer */
 static void inline fire_AOS(PREFIX int vertex_count, int type)
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index ad2cdf585a..874bd856dd 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1607,18 +1607,30 @@ void r300GenerateSimpleVertexShader(r300ContextPtr r300)
 void r300SetupVertexShader(r300ContextPtr rmesa)
 {
 	GLcontext* ctx = rmesa->radeon.glCtx;
+	struct r300_vertex_shader_fragment unk4={
+			length: 4,
+			body: { f: {
+				/*0.0*/(rand()%100)/10.0,
+				/*0.0*/(rand()%100)/10.0,
+				/*1.0*/(rand()%100)/10.0,
+				/*0.0*/(rand()%100)/10.0
+				} }
+			};
 	LOCAL_VARS
 
-	if(rmesa->current_vp != NULL){
-		r300SetupVertexProgram(rmesa);
-		return ;
-	}
-
 	/* Reset state, in case we don't use something */
 	((drm_r300_cmd_header_t*)rmesa->hw.vpp.cmd)->vpu.count = 0;
 	((drm_r300_cmd_header_t*)rmesa->hw.vpi.cmd)->vpu.count = 0;
 	((drm_r300_cmd_header_t*)rmesa->hw.vps.cmd)->vpu.count = 0;
 
+	/* Not sure why this doesnt work...
+	   0x400 area might have something to do with pixel shaders as it appears right after pfs programming.
+	   0x406 is set to { 0.0, 0.0, 1.0, 0.0 } most of the time but should change with smooth points and in other rare cases. */
+	//setup_vertex_shader_fragment(rmesa, 0x406, &unk4);
+	if(rmesa->current_vp != NULL){
+		r300SetupVertexProgram(rmesa);
+		return ;
+	}
 
 /* This needs to be replaced by vertex shader generation code */
 
diff --git a/src/mesa/drivers/dri/r300/r300_vertexprog.c b/src/mesa/drivers/dri/r300/r300_vertexprog.c
index 02e0b5b739..811f9ccb55 100644
--- a/src/mesa/drivers/dri/r300/r300_vertexprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertexprog.c
@@ -422,7 +422,6 @@ static unsigned long t_opcode(enum vp_opcode opcode)
 		case VP_OPCODE_EXP: return R300_VPI_OUT_OP_EXP;
 		case VP_OPCODE_FRC: return R300_VPI_OUT_OP_FRC;
 		case VP_OPCODE_LG2: return R300_VPI_OUT_OP_LG2;
-		case VP_OPCODE_LIT: return R300_VPI_OUT_OP_LIT;
 		case VP_OPCODE_LOG: return R300_VPI_OUT_OP_LOG;
 		case VP_OPCODE_MAD: return R300_VPI_OUT_OP_MAD;
 		case VP_OPCODE_MAX: return R300_VPI_OUT_OP_MAX;
@@ -472,7 +471,10 @@ static void translate_program(struct r300_vertex_program *vp)
 	VERTEX_SHADER_INSTRUCTION t2rs[1024];
 	VERTEX_SHADER_INSTRUCTION *o_inst;
 	unsigned long operands;
-	int u_temp_i=63; /* Initial value should be last tmp reg that hw supports */
+	/* Initial value should be last tmp reg that hw supports.
+	   Strangely enough r300 doesnt mind even though these would be out of range.
+	   Smart enough to realize that it doesnt need it? */
+	int u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1;
 #ifdef SRCS_WRITABLE
 	struct vp_src_register src[3];
 #else	
@@ -501,12 +503,12 @@ static void translate_program(struct r300_vertex_program *vp)
 				o_inst->src1=MAKE_VSF_SOURCE(t_src_index(vp, &src[2]),
 						SWIZZLE_X, SWIZZLE_Y,
 						SWIZZLE_Z, SWIZZLE_W,
-						t_src_class(src[0].File), VSF_FLAG_NONE);
+						t_src_class(src[2].File), VSF_FLAG_NONE);
 
 				o_inst->src2=MAKE_VSF_SOURCE(t_src_index(vp, &src[2]),
 						SWIZZLE_ZERO, SWIZZLE_ZERO,
 						SWIZZLE_ZERO, SWIZZLE_ZERO,
-						t_src_class(src[0].File), VSF_FLAG_NONE);
+						t_src_class(src[2].File), VSF_FLAG_NONE);
 				o_inst->src3=0;
 				o_inst++;
 						
@@ -601,7 +603,8 @@ static void translate_program(struct r300_vertex_program *vp)
 #ifdef SRCS_WRITABLE
 			vpi->Opcode=VP_OPCODE_MAX;
 			src[1]=src[0];
-			src[1].Negate=GL_TRUE;
+			src[1].Negate=!src[0].Negate;
+			operands=op_operands(vpi->Opcode);
 			break;
 #else
 			o_inst->op=MAKE_VSF_OP(R300_VPI_OUT_OP_MAX, vpi->DstReg.Index,
@@ -647,6 +650,33 @@ static void translate_program(struct r300_vertex_program *vp)
 			u_temp_i--;
 			goto next;
 			
+		case VP_OPCODE_LIT://LIT TMP 1.Y Z TMP 1{} {X W Z Y} TMP 1{} {Y W Z X} TMP 1{} {Y X Z W} 
+			o_inst->op=MAKE_VSF_OP(R300_VPI_OUT_OP_LIT, vpi->DstReg.Index,
+					t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File));
+			/* NOTE: Users swizzling might not work. */
+			o_inst->src1=MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
+					t_swizzle(src[0].Swizzle[0]), // x
+					t_swizzle(src[0].Swizzle[3]), // w
+					t_swizzle(src[0].Swizzle[2]), // z
+					t_swizzle(src[0].Swizzle[1]), // y
+					t_src_class(src[0].File),
+					src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE);
+			o_inst->src2=MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
+					t_swizzle(src[0].Swizzle[1]), // y
+					t_swizzle(src[0].Swizzle[3]), // w
+					t_swizzle(src[0].Swizzle[2]), // z
+					t_swizzle(src[0].Swizzle[0]), // x
+					t_src_class(src[0].File),
+					src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE);
+			o_inst->src3=MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
+					t_swizzle(src[0].Swizzle[1]), // y
+					t_swizzle(src[0].Swizzle[0]), // x
+					t_swizzle(src[0].Swizzle[2]), // z
+					t_swizzle(src[0].Swizzle[3]), // w
+					t_src_class(src[0].File),
+					src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE);
+			goto next;
+			
 		case VP_OPCODE_DPH://DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ONE} PARAM 0{} {X Y Z W} 
 			o_inst->op=MAKE_VSF_OP(R300_VPI_OUT_OP_DOT, vpi->DstReg.Index,
 					t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File));
@@ -662,7 +692,7 @@ static void translate_program(struct r300_vertex_program *vp)
 			o_inst->src3=0;
 			goto next;
 			
-		case VP_OPCODE_XPD:
+		case VP_OPCODE_XPD: /* Broken due to swizzling */
 		/* ADD TMP 0.X Y Z PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} 
 		   MUL TMP 1.X Y Z W TMP 0{} {Z X Y ZERO} PARAM 1{} {Y Z X ZERO} 
 		   MAD RESULT 1.X Y Z W TMP 0{} {Y Z X ONE} PARAM 1{} {Z X Y ONE} TMP 1{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W*/