X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fr300%2Fr300_draw.c;h=3dcd986e2267f3ae09abc6342e0efaa30dbe4456;hb=323d1fb3910d7e53cb5200ee90849b2231fd96fb;hp=1d6e6db773db1ce246933bb9b4434f8773f08a8a;hpb=9e018d822523e559fa8d92c3b5a83dd5554a0676;p=mesa.git diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c index 1d6e6db773d..3dcd986e226 100644 --- a/src/mesa/drivers/dri/r300/r300_draw.c +++ b/src/mesa/drivers/dri/r300/r300_draw.c @@ -31,6 +31,7 @@ #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" @@ -38,8 +39,10 @@ #include "r300_render.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" @@ -79,19 +82,25 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer GLvoid *src_ptr; GLuint *out; int i; + GLboolean mapped_named_bo = GL_FALSE; if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) { ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj); + mapped_named_bo = GL_TRUE; assert(mesa_ind_buf->obj->Pointer != NULL); } 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; radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4); - + radeon_bo_map(r300->ind_buf.bo, 1); assert(r300->ind_buf.bo->ptr != NULL); out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset); @@ -102,15 +111,17 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer if (i < mesa_ind_buf->count) { *out++ = in[i]; } - + radeon_bo_unmap(r300->ind_buf.bo); #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) + radeon_bo_map(r300->ind_buf.bo, 1); + 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) { @@ -120,23 +131,28 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer if (i < mesa_ind_buf->count) { *out++ = in[i]; } + radeon_bo_unmap(r300->ind_buf.bo); #endif } r300->ind_buf.is_32bit = GL_FALSE; r300->ind_buf.count = mesa_ind_buf->count; + + if (mapped_named_bo) { + ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, mesa_ind_buf->obj); + } } static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf) { r300ContextPtr r300 = R300_CONTEXT(ctx); - GLboolean mapped_named_bo = GL_FALSE; if (!mesa_ind_buf) { 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) { @@ -145,6 +161,7 @@ static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer #endif const GLvoid *src_ptr; GLvoid *dst_ptr; + GLboolean mapped_named_bo = GL_FALSE; if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) { ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj); @@ -158,12 +175,18 @@ static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4); + radeon_bo_map(r300->ind_buf.bo, 1); assert(r300->ind_buf.bo->ptr != NULL); dst_ptr = ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset); _mesa_memcpy(dst_ptr, src_ptr, size); + radeon_bo_unmap(r300->ind_buf.bo); r300->ind_buf.is_32bit = (mesa_ind_buf->type == GL_UNSIGNED_INT); r300->ind_buf.count = mesa_ind_buf->count; + + if (mapped_named_bo) { + ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, mesa_ind_buf->obj); + } } else { r300FixupIndexBuffer(ctx, mesa_ind_buf); } @@ -223,12 +246,14 @@ 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); + radeon_bo_map(attr->bo, 1); 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); @@ -260,6 +285,7 @@ static void r300ConvertAttrib(GLcontext *ctx, int count, const struct gl_client_ break; } + radeon_bo_unmap(attr->bo); if (mapped_named_bo) { ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj); } @@ -274,11 +300,15 @@ static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *i radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, size, 32); + radeon_bo_map(attr->bo, 1); + if (!input->BufferObj->Pointer) { ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj); 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); @@ -295,6 +325,7 @@ static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *i ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj); } + radeon_bo_unmap(attr->bo); attr->stride = dst_stride; } @@ -306,6 +337,7 @@ 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 || @@ -316,7 +348,6 @@ 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 { @@ -327,41 +358,14 @@ 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; } } @@ -456,7 +460,7 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_vertex_buffer *vbuf = &r300->vbuf; - + radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s\n", __func__); { int i, tmp; @@ -480,33 +484,88 @@ 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; - 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; + 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 { + + 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); + radeon_bo_map(vbuf->attribs[index].bo, 1); + 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; + } + radeon_bo_unmap(vbuf->attribs[index].bo); - if (vbuf->attribs[i].is_named_bo) { - radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, - aos->bo, - RADEON_GEM_DOMAIN_GTT, 0); } } - r300->radeon.tcl.aos_count = vbuf->num_attribs; - if (r300->ind_buf.bo) { - radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, - r300->ind_buf.bo, - RADEON_GEM_DOMAIN_GTT, 0); + 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) @@ -515,6 +574,7 @@ 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; @@ -534,6 +594,33 @@ 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, @@ -545,6 +632,9 @@ 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 ); @@ -555,14 +645,6 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx)); - r300SetupIndexBuffer(ctx, ib); - - /* 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 + (50*sizeof(int)), - __FUNCTION__); - r300SetVertexFormat(ctx, arrays, max_index + 1); if (r300->fallback) @@ -572,6 +654,18 @@ 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); @@ -583,6 +677,13 @@ static GLboolean r300TryDrawPrims(GLcontext *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; } @@ -591,28 +692,26 @@ static void r300DrawPrims(GLcontext *ctx, const struct _mesa_prim *prim, GLuint nr_prims, const struct _mesa_index_buffer *ib, + GLboolean index_bounds_valid, GLuint min_index, GLuint max_index) { - struct split_limits limits; GLboolean retval; - if (ib) - limits.max_verts = 0xffffffff; - else - limits.max_verts = 65535; - - limits.max_indices = 65535; - limits.max_vb_size = 1024*1024; + /* This check should get folded into just the places that + * min/max index are really needed. + */ + if (!index_bounds_valid) { + vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index); + } 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; } - if ((ib && ib->count > 65535)) { - vbo_split_prims (ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims, &limits); - return; - } /* Make an attempt at drawing */ retval = r300TryDrawPrims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);