r600: support vertex_array_bgra

Use vertex program key mechanism and swizzle during vertex fetch - is there
a better way?

5 files changed, 25 insertions, 6 deletions

diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h
index 394fd757d4b..94662ab5479 100644
--- a/src/mesa/drivers/dri/r600/r600_context.h
+++ b/src/mesa/drivers/dri/r600/r600_context.h
@@ -108,6 +108,7 @@ typedef struct StreamDesc
 	GLint   size;   //number of data element
 	GLenum  type;  //data element type
 	GLsizei stride;
+	GLenum  format; // GL_RGBA,GLBGRA 
 
 	struct radeon_bo *bo;
 	GLint  bo_offset;
diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c
index 67e0ee77463..b0b38927e67 100644
--- a/src/mesa/drivers/dri/r600/r700_assembler.c
+++ b/src/mesa/drivers/dri/r600/r700_assembler.c
@@ -798,6 +798,7 @@ GLboolean assemble_vfetch_instruction2(r700_AssemblerBase* pAsm,
                                        GLubyte             element,
                                        GLuint              _signed,
                                        GLboolean           normalize,
+                                       GLenum              format,
                                        VTX_FETCH_METHOD  * pFetchMethod)
 {
     GLuint client_size_inbyte;
@@ -846,10 +847,21 @@ GLboolean assemble_vfetch_instruction2(r700_AssemblerBase* pAsm,
 	vfetch_instruction_ptr->m_Word0.f.src_sel_x        = SQ_SEL_X;
 	vfetch_instruction_ptr->m_Word0.f.mega_fetch_count = mega_fetch_count;
 
-	vfetch_instruction_ptr->m_Word1.f.dst_sel_x        = (size < 1) ? SQ_SEL_0 : SQ_SEL_X;
-	vfetch_instruction_ptr->m_Word1.f.dst_sel_y        = (size < 2) ? SQ_SEL_0 : SQ_SEL_Y;
-	vfetch_instruction_ptr->m_Word1.f.dst_sel_z        = (size < 3) ? SQ_SEL_0 : SQ_SEL_Z;
-	vfetch_instruction_ptr->m_Word1.f.dst_sel_w        = (size < 4) ? SQ_SEL_1 : SQ_SEL_W;
+	if(format == GL_BGRA)
+	{
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_x        = (size < 1) ? SQ_SEL_0 : SQ_SEL_Z;
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_y        = (size < 2) ? SQ_SEL_0 : SQ_SEL_Y;
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_z        = (size < 3) ? SQ_SEL_0 : SQ_SEL_X;
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_w        = (size < 4) ? SQ_SEL_1 : SQ_SEL_W;
+	}
+	else
+	{
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_x        = (size < 1) ? SQ_SEL_0 : SQ_SEL_X;
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_y        = (size < 2) ? SQ_SEL_0 : SQ_SEL_Y;
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_z        = (size < 3) ? SQ_SEL_0 : SQ_SEL_Z;
+		vfetch_instruction_ptr->m_Word1.f.dst_sel_w        = (size < 4) ? SQ_SEL_1 : SQ_SEL_W;
+
+	}
 
 	vfetch_instruction_ptr->m_Word1.f.use_const_fields = 1;
     vfetch_instruction_ptr->m_Word1.f.data_format      = data_format;
diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h
index c66db502a17..ee125826612 100644
--- a/src/mesa/drivers/dri/r600/r700_assembler.h
+++ b/src/mesa/drivers/dri/r600/r700_assembler.h
@@ -422,6 +422,7 @@ GLboolean assemble_vfetch_instruction2(r700_AssemblerBase* pAsm,
                                        GLubyte             element,
                                        GLuint              _signed,
                                        GLboolean           normalize,
+                                       GLenum              format,
                                        VTX_FETCH_METHOD  * pFetchMethod);
 GLboolean cleanup_vfetch_instructions(r700_AssemblerBase* pAsm);
 GLuint gethelpr(r700_AssemblerBase* pAsm);
diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c
index ffc6068bd88..4ca5ce693a8 100644
--- a/src/mesa/drivers/dri/r600/r700_vertprog.c
+++ b/src/mesa/drivers/dri/r600/r700_vertprog.c
@@ -179,7 +179,8 @@ GLboolean Process_Vertex_Program_Vfetch_Instructions2(
                                       context->stream_desc[i].size,
                                       context->stream_desc[i].element,
                                       context->stream_desc[i]._signed,
-                                      context->stream_desc[i].normalize,						            
+                                      context->stream_desc[i].normalize,
+                                      context->stream_desc[i].format,
                                      &vtxFetchMethod);
     }
 
@@ -308,6 +309,7 @@ struct r700_vertex_program* r700TranslateVertexShader(GLcontext *ctx,
 		vp->aos_desc[i].size   = context->stream_desc[i].size;
 		vp->aos_desc[i].stride = context->stream_desc[i].stride;
 		vp->aos_desc[i].type   = context->stream_desc[i].type;
+		vp->aos_desc[i].format = context->stream_desc[i].format;
 	}
 
 	if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)
@@ -368,7 +370,8 @@ void r700SelectVertexShader(GLcontext *ctx)
 	match = GL_TRUE;
 	for(i=0; i<context->nNumActiveAos; i++)
 	{
-		if (vp->aos_desc[i].size != context->stream_desc[i].size)
+		if (vp->aos_desc[i].size != context->stream_desc[i].size ||
+		    vp->aos_desc[i].format != context->stream_desc[i].format)
 		{
 			match = GL_FALSE;
 			break;
@@ -471,6 +474,7 @@ static void r700TranslateAttrib(GLcontext *ctx, GLuint unLoc, int count, const s
 	pStreamDesc->size = input->Size;
 	pStreamDesc->dst_loc = context->nNumActiveAos;
 	pStreamDesc->element = unLoc;
+	pStreamDesc->format = input->Format;
 
 	switch (pStreamDesc->type) 
 	{ //GetSurfaceFormat
diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.h b/src/mesa/drivers/dri/r600/r700_vertprog.h
index 00824c29d32..645c9ac84aa 100644
--- a/src/mesa/drivers/dri/r600/r700_vertprog.h
+++ b/src/mesa/drivers/dri/r600/r700_vertprog.h
@@ -39,6 +39,7 @@ typedef struct ArrayDesc //TEMP
 	GLint size;   //number of data element
 	GLenum type;  //data element type
 	GLsizei stride;
+	GLenum format; //GL_RGBA or GL_BGRA
 } ArrayDesc;
 
 struct r700_vertex_program