X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fr300%2Fr300_draw.c;h=e9968f9ffeffb8d958ac9f1245e8cd41e196c4b8;hb=45e76d2665b38ba3787548310efc59e969124c01;hp=fb416a05c0872df85406d5d0ed903db110846455;hpb=0d0f01e2e0b37ed5152614ceeff34da8e46b5e37;p=mesa.git

diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
index fb416a05c08..e9968f9ffef 100644
--- a/src/mesa/drivers/dri/r300/r300_draw.c
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -31,16 +31,18 @@
 #include "main/state.h"
 #include "main/api_validate.h"
 #include "main/enums.h"
+#include "main/simple_list.h"
 
 #include "r300_reg.h"
 #include "r300_context.h"
 #include "r300_emit.h"
 #include "r300_render.h"
-#include "r300_queryobj.h"
 #include "r300_state.h"
 #include "r300_tex.h"
+#include "r300_cmdbuf.h"
 
 #include "radeon_buffer_objects.h"
+#include "radeon_common_context.h"
 
 #include "tnl/tnl.h"
 #include "tnl/t_vp_build.h"
@@ -89,6 +91,10 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 	}
 	src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr);
 
+	radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
+			"%s: Fixing index buffer format. type %d\n",
+			__func__, mesa_ind_buf->type);
+
 	if (mesa_ind_buf->type == GL_UNSIGNED_BYTE) {
 		GLuint size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
 		GLubyte *in = (GLubyte *)src_ptr;
@@ -109,11 +115,12 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 #if MESA_BIG_ENDIAN
 	} else { /* if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) */
 		GLushort *in = (GLushort *)src_ptr;
-		size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
+		GLuint size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
 
-		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offet, size, 4);
+		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo,
+				     &r300->ind_buf.bo_offset, size, 4);
 
-		assert(r300->ind_buf.bo->ptr != NULL)
+		assert(r300->ind_buf.bo->ptr != NULL);
 		out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 
 		for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
@@ -143,6 +150,7 @@ static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 		r300->ind_buf.bo = NULL;
 		return;
 	}
+	radeon_print(RADEON_RENDER, RADEON_TRACE, "%s\n", __func__);
 
 #if MESA_BIG_ENDIAN
 	if (mesa_ind_buf->type == GL_UNSIGNED_INT) {
@@ -236,10 +244,11 @@ static void r300ConvertAttrib(GLcontext *ctx, int count, const struct gl_client_
 	radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, sizeof(GLfloat) * input->Size * count, 32);
 	dst_ptr = (GLfloat *)ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
 
-	if (RADEON_DEBUG & DEBUG_FALLBACKS) {
-		fprintf(stderr, "%s: Converting vertex attributes, attribute data format %x,", __FUNCTION__, input->Type);
-		fprintf(stderr, "stride %d, components %d\n", stride, input->Size);
-	}
+	radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
+			"%s: Converting vertex attributes, attribute data format %x,"
+			"stride %d, components %d\n"
+			, __FUNCTION__, input->Type
+			, stride, input->Size);
 
 	assert(src_ptr != NULL);
 
@@ -290,6 +299,8 @@ static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *i
 		mapped_named_bo = GL_TRUE;
 	}
 
+	radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT, "%s. Vertex alignment doesn't match hw requirements.\n", __func__);
+
 	{
 		GLvoid *src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr);
 		GLvoid *dst_ptr = ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
@@ -317,6 +328,7 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st
 	GLenum type;
 	GLuint stride;
 
+	radeon_print(RADEON_RENDER, RADEON_TRACE, "%s\n", __func__);
 	stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB;
 
 	if (input->Type == GL_DOUBLE || input->Type == GL_UNSIGNED_INT || input->Type == GL_INT ||
@@ -327,7 +339,6 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st
 
 		type = GL_FLOAT;
 
-		r300ConvertAttrib(ctx, count, input, &r300_attr);
 		if (input->StrideB == 0) {
 			r300_attr.stride = 0;
 		} else {
@@ -338,41 +349,14 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st
 	} else {
 		type = input->Type;
 		r300_attr.dwords = (getTypeSize(type) * input->Size + 3)/ 4;
-		if (input->BufferObj->Name) {
-			if (stride % 4 != 0) {
-				assert(((int) input->Ptr) % input->StrideB == 0);
-				r300AlignDataToDword(ctx, input, count, &r300_attr);
-				r300_attr.is_named_bo = GL_FALSE;
-			} else {
-				r300_attr.stride = input->StrideB;
-				r300_attr.bo_offset = (GLuint) input->Ptr;
-				r300_attr.bo = get_radeon_buffer_object(input->BufferObj)->bo;
-				r300_attr.is_named_bo = GL_TRUE;
-			}
-		} else {
-			int size;
-			uint32_t *dst;
+		if (!input->BufferObj->Name) {
 
 			if (input->StrideB == 0) {
-				size = getTypeSize(input->Type) * input->Size;
-				count = 1;
 				r300_attr.stride = 0;
 			} else {
-				size = getTypeSize(input->Type) * input->Size * count;
 				r300_attr.stride = (getTypeSize(type) * input->Size + 3) & ~3;
 			}
 
-			radeonAllocDmaRegion(&r300->radeon, &r300_attr.bo, &r300_attr.bo_offset, size, 32);
-			assert(r300_attr.bo->ptr != NULL);
-			dst = (uint32_t *)ADD_POINTERS(r300_attr.bo->ptr, r300_attr.bo_offset);
-			switch (r300_attr.dwords) {
-				case 1: radeonEmitVec4(dst, input->Ptr, input->StrideB, count); break;
-				case 2: radeonEmitVec8(dst, input->Ptr, input->StrideB, count); break;
-				case 3: radeonEmitVec12(dst, input->Ptr, input->StrideB, count); break;
-				case 4: radeonEmitVec16(dst, input->Ptr, input->StrideB, count); break;
-				default: assert(0); break;
-			}
-
 			r300_attr.is_named_bo = GL_FALSE;
 		}
 	}
@@ -467,7 +451,7 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	struct r300_vertex_buffer *vbuf = &r300->vbuf;
-	int ret;
+	radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s\n", __func__);
 	{
 		int i, tmp;
 
@@ -491,29 +475,86 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar
 	r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS);
 	if (r300->fallback)
 		return;
+}
 
-	{
-		int i;
+static void r300AllocDmaRegions(GLcontext *ctx, const struct gl_client_array *input[], int count)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_vertex_buffer *vbuf = &r300->vbuf;
+	GLuint stride;
+	int ret;
+	int i, index;
+	radeon_print(RADEON_RENDER, RADEON_VERBOSE,
+			"%s: count %d num_attribs %d\n",
+			__func__, count, vbuf->num_attribs);
+
+	for (index = 0; index < vbuf->num_attribs; index++) {
+		struct radeon_aos *aos = &r300->radeon.tcl.aos[index];
+		i = vbuf->attribs[index].element;
 
-		for (i = 0; i < vbuf->num_attribs; i++) {
-			struct radeon_aos *aos = &r300->radeon.tcl.aos[i];
+		stride = (input[i]->StrideB == 0) ? getTypeSize(input[i]->Type) * input[i]->Size : input[i]->StrideB;
+
+		if (input[i]->Type == GL_DOUBLE || input[i]->Type == GL_UNSIGNED_INT || input[i]->Type == GL_INT ||
+#if MESA_BIG_ENDIAN
+				getTypeSize(input[i]->Type) != 4 ||
+#endif
+				stride < 4) {
+
+			r300ConvertAttrib(ctx, count, input[i], &vbuf->attribs[index]);
+		} else {
+			if (input[i]->BufferObj->Name) {
+				if (stride % 4 != 0) {
+					assert(((intptr_t) input[i]->Ptr) % input[i]->StrideB == 0);
+					r300AlignDataToDword(ctx, input[i], count, &vbuf->attribs[index]);
+					vbuf->attribs[index].is_named_bo = GL_FALSE;
+				} else {
+					vbuf->attribs[index].stride = input[i]->StrideB;
+					vbuf->attribs[index].bo_offset = (intptr_t) input[i]->Ptr;
+					vbuf->attribs[index].bo = get_radeon_buffer_object(input[i]->BufferObj)->bo;
+					vbuf->attribs[index].is_named_bo = GL_TRUE;
+				}
+			} else {
 
-			aos->count = vbuf->attribs[i].stride == 0 ? 1 : count;
-			aos->stride = vbuf->attribs[i].stride / sizeof(float);
-			aos->offset = vbuf->attribs[i].bo_offset;
-			aos->components = vbuf->attribs[i].dwords;
-			aos->bo = vbuf->attribs[i].bo;
+				int size;
+				int local_count = count;
+				uint32_t *dst;
+
+				if (input[i]->StrideB == 0) {
+					size = getTypeSize(input[i]->Type) * input[i]->Size;
+					local_count = 1;
+				} else {
+					size = getTypeSize(input[i]->Type) * input[i]->Size * local_count;
+				}
+
+				radeonAllocDmaRegion(&r300->radeon, &vbuf->attribs[index].bo, &vbuf->attribs[index].bo_offset, size, 32);
+				assert(vbuf->attribs[index].bo->ptr != NULL);
+				dst = (uint32_t *)ADD_POINTERS(vbuf->attribs[index].bo->ptr, vbuf->attribs[index].bo_offset);
+				switch (vbuf->attribs[index].dwords) {
+					case 1: radeonEmitVec4(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
+					case 2: radeonEmitVec8(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
+					case 3: radeonEmitVec12(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
+					case 4: radeonEmitVec16(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
+					default: assert(0); break;
+				}
 
-			if (vbuf->attribs[i].is_named_bo) {
-				radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, r300->vbuf.attribs[i].bo, RADEON_GEM_DOMAIN_GTT, 0);
 			}
 		}
 
-		r300->radeon.tcl.aos_count = vbuf->num_attribs;
-		ret = radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, r300->radeon.dma.current, RADEON_GEM_DOMAIN_GTT, 0);
-		if (ret)
-			r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, GL_TRUE);
+		aos->count = vbuf->attribs[index].stride == 0 ? 1 : count;
+		aos->stride = vbuf->attribs[index].stride / sizeof(float);
+		aos->components = vbuf->attribs[index].dwords;
+		aos->bo = vbuf->attribs[index].bo;
+		aos->offset = vbuf->attribs[index].bo_offset;
+
+		if (vbuf->attribs[index].is_named_bo) {
+			radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, r300->vbuf.attribs[index].bo, RADEON_GEM_DOMAIN_GTT, 0);
+		}
 	}
+
+	r300->radeon.tcl.aos_count = vbuf->num_attribs;
+	ret = radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, first_elem(&r300->radeon.dma.reserved)->bo, RADEON_GEM_DOMAIN_GTT, 0);
+	r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, ret);
+
 }
 
 static void r300FreeData(GLcontext *ctx)
@@ -522,6 +563,7 @@ static void r300FreeData(GLcontext *ctx)
 	 * to prevent double unref in radeonReleaseArrays
 	 * called during context destroy
 	 */
+	radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s\n", __func__);
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	{
 		int i;
@@ -541,6 +583,33 @@ static void r300FreeData(GLcontext *ctx)
 	}
 }
 
+static GLuint r300PredictTryDrawPrimsSize(GLcontext *ctx, GLuint nr_prims)
+{
+	struct r300_context *r300 = R300_CONTEXT(ctx);
+	struct r300_vertex_buffer *vbuf = &r300->vbuf;
+	GLboolean flushed;
+	GLuint dwords;
+	GLuint state_size;
+
+	dwords = 2*CACHE_FLUSH_BUFSZ;
+	dwords += PRE_EMIT_STATE_BUFSZ;
+	dwords += (AOS_BUFSZ(vbuf->num_attribs)
+		+ SCISSORS_BUFSZ*2
+		+ FIREAOS_BUFSZ )*nr_prims;
+
+	state_size = radeonCountStateEmitSize(&r300->radeon);
+	flushed = rcommonEnsureCmdBufSpace(&r300->radeon,
+			dwords + state_size,
+			__FUNCTION__);
+	if (flushed)
+		dwords += radeonCountStateEmitSize(&r300->radeon);
+	else
+		dwords += state_size;
+
+	radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s: total prediction size is %d.\n", __FUNCTION__, dwords);
+	return dwords;
+}
+
 static GLboolean r300TryDrawPrims(GLcontext *ctx,
 					 const struct gl_client_array *arrays[],
 					 const struct _mesa_prim *prim,
@@ -552,6 +621,9 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 	struct r300_context *r300 = R300_CONTEXT(ctx);
 	GLuint i;
 
+	radeon_print(RADEON_RENDER, RADEON_NORMAL, "%s: %u (%d-%d) cs begin at %d\n",
+				__FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw );
+
 	if (ctx->NewState)
 		_mesa_update_state( ctx );
 
@@ -562,14 +634,6 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 
 	r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx));
 
-	/* ensure we have the cmd buf space in advance to cover
- 	 * the state + DMA AOS pointers */
-	rcommonEnsureCmdBufSpace(&r300->radeon,
-                           r300->radeon.hw.max_state_size + (60*sizeof(int)),
-                          __FUNCTION__);
-
-	r300SetupIndexBuffer(ctx, ib);
-
 	r300SetVertexFormat(ctx, arrays, max_index + 1);
 
 	if (r300->fallback)
@@ -579,21 +643,36 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 
 	r300UpdateShaderStates(r300);
 
+	/* ensure we have the cmd buf space in advance to cover
+	 * the state + DMA AOS pointers */
+	GLuint emit_end = r300PredictTryDrawPrimsSize(ctx, nr_prims)
+		+ r300->radeon.cmdbuf.cs->cdw;
+
+	r300SetupIndexBuffer(ctx, ib);
+
+	r300AllocDmaRegions(ctx, arrays, max_index + 1);
+
+	if (r300->fallback)
+		return GL_FALSE;
+
 	r300EmitCacheFlush(r300);
 	radeonEmitState(&r300->radeon);
 
-	r300EmitQueryBegin(ctx);
-
 	for (i = 0; i < nr_prims; ++i) {
 		r300RunRenderPrimitive(ctx, prim[i].start, prim[i].start + prim[i].count, prim[i].mode);
 	}
 
 	r300EmitCacheFlush(r300);
 
-	r300EmitQueryEnd(ctx);
-
 	r300FreeData(ctx);
 
+	radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s: %u (%d-%d) cs ending at %d\n",
+			__FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw );
+
+	if (emit_end < r300->radeon.cmdbuf.cs->cdw)
+		WARN_ONCE("Rendering was %d commands larger than predicted size."
+				" We might overflow  command buffer.\n", r300->radeon.cmdbuf.cs->cdw - emit_end);
+
 	return GL_TRUE;
 }
 
@@ -616,6 +695,9 @@ static void r300DrawPrims(GLcontext *ctx,
 	}
 
 	if (min_index) {
+		radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
+				"%s: Rebasing primitives. %p nr_prims %d min_index %u max_index %u\n",
+				__func__, prim, nr_prims, min_index, max_index);
 		vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims );
 		return;
 	}