From: Alex Deucher Date: Wed, 4 Nov 2009 21:59:13 +0000 (-0500) Subject: r600: rework draw functions X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=9fce12b894c3af33d7a0732332446893682a48d5;p=mesa.git r600: rework draw functions Seems INDX_OFFSET doesn't work properly on some cards, so change back to immediate mode indices. Seems to only affect DRI1. Needs more investigation. Rework and clean up the draw functions. Signed-off-by: Alex Deucher --- diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c index b20b129d170..c345b9d8ac9 100644 --- a/src/mesa/drivers/dri/r600/r700_render.c +++ b/src/mesa/drivers/dri/r600/r700_render.c @@ -259,16 +259,6 @@ static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim uint32_t vgt_index_type = 0; uint32_t vgt_primitive_type = 0; uint32_t vgt_num_indices = 0; - GLboolean bUseDrawIndex; - - if(NULL != context->ind_buf.bo) - { - bUseDrawIndex = GL_TRUE; - } - else - { - bUseDrawIndex = GL_FALSE; - } type = r700PrimitiveType(prim); num_indices = r700NumVerts(end - start, prim); @@ -280,85 +270,154 @@ static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim if (type < 0 || num_indices <= 0) return; - if(GL_TRUE == bUseDrawIndex) - { - total_emit = 3 /* VGT_PRIMITIVE_TYPE */ - + 2 /* VGT_INDEX_TYPE */ - + 2 /* NUM_INSTANCES */ - + 3 /* VGT_INDEX_OFFSET */ - + 5 + 2; /* DRAW_INDEX */ - } - else - { - total_emit = 3 /* VGT_PRIMITIVE_TYPE */ - + 2 /* VGT_INDEX_TYPE */ - + 2 /* NUM_INSTANCES */ - + 3 /* VGT_INDEX_OFFSET */ - + 3; /* DRAW_INDEX_AUTO */ - } - - BEGIN_BATCH_NO_AUTOSTATE(total_emit); - // prim SETfield(vgt_primitive_type, type, VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift, VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask); - R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1)); - R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX); - R600_OUT_BATCH(vgt_primitive_type); - // index type SETfield(vgt_index_type, DI_INDEX_SIZE_32_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask); - if(GL_TRUE == bUseDrawIndex) + if(GL_TRUE != context->ind_buf.is_32bit) { - if(GL_TRUE != context->ind_buf.is_32bit) - { SETfield(vgt_index_type, DI_INDEX_SIZE_16_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask); - } } + vgt_num_indices = num_indices; + SETfield(vgt_draw_initiator, DI_SRC_SEL_DMA, SOURCE_SELECT_shift, SOURCE_SELECT_mask); + SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift, MAJOR_MODE_mask); + + total_emit = 3 /* VGT_PRIMITIVE_TYPE */ + + 2 /* VGT_INDEX_TYPE */ + + 2 /* NUM_INSTANCES */ + + 5 + 2; /* DRAW_INDEX */ + + BEGIN_BATCH_NO_AUTOSTATE(total_emit); + // prim + R600_OUT_BATCH_REGSEQ(VGT_PRIMITIVE_TYPE, 1); + R600_OUT_BATCH(vgt_primitive_type); + // index type R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0)); R600_OUT_BATCH(vgt_index_type); - // num instances R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0)); R600_OUT_BATCH(1); - // draw packet - vgt_num_indices = num_indices; + R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX, 3)); + R600_OUT_BATCH(context->ind_buf.bo_offset); + R600_OUT_BATCH(0); + R600_OUT_BATCH(vgt_num_indices); + R600_OUT_BATCH(vgt_draw_initiator); + R600_OUT_BATCH_RELOC(context->ind_buf.bo_offset, + context->ind_buf.bo, + context->ind_buf.bo_offset, + RADEON_GEM_DOMAIN_GTT, 0, 0); + END_BATCH(); + COMMIT_BATCH(); +} + +static void r700RunRenderPrimitiveImmediate(GLcontext * ctx, int start, int end, int prim) +{ + context_t *context = R700_CONTEXT(ctx); + BATCH_LOCALS(&context->radeon); + int type, i; + uint32_t num_indices, total_emit = 0; + uint32_t vgt_draw_initiator = 0; + uint32_t vgt_index_type = 0; + uint32_t vgt_primitive_type = 0; + uint32_t vgt_num_indices = 0; + + type = r700PrimitiveType(prim); + num_indices = r700NumVerts(end - start, prim); - if(GL_TRUE == bUseDrawIndex) + radeon_print(RADEON_RENDER, RADEON_TRACE, + "%s type %x num_indices %d\n", + __func__, type, num_indices); + + if (type < 0 || num_indices <= 0) + return; + + SETfield(vgt_primitive_type, type, + VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift, VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask); + + if (num_indices > 0xffff) { - SETfield(vgt_draw_initiator, DI_SRC_SEL_DMA, SOURCE_SELECT_shift, SOURCE_SELECT_mask); + SETfield(vgt_index_type, DI_INDEX_SIZE_32_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask); } else { - SETfield(vgt_draw_initiator, DI_SRC_SEL_AUTO_INDEX, SOURCE_SELECT_shift, SOURCE_SELECT_mask); + SETfield(vgt_index_type, DI_INDEX_SIZE_16_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask); } + vgt_num_indices = num_indices; SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift, MAJOR_MODE_mask); - if(GL_TRUE == bUseDrawIndex) + if (start == 0) { - R600_OUT_BATCH_REGSEQ(VGT_INDX_OFFSET, 1); - R600_OUT_BATCH(0); - R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX, 3)); - R600_OUT_BATCH(context->ind_buf.bo_offset); - R600_OUT_BATCH(0); - R600_OUT_BATCH(vgt_num_indices); - R600_OUT_BATCH(vgt_draw_initiator); - R600_OUT_BATCH_RELOC(context->ind_buf.bo_offset, - context->ind_buf.bo, - context->ind_buf.bo_offset, - RADEON_GEM_DOMAIN_GTT, 0, 0); + SETfield(vgt_draw_initiator, DI_SRC_SEL_AUTO_INDEX, SOURCE_SELECT_shift, SOURCE_SELECT_mask); } else { - R600_OUT_BATCH_REGSEQ(VGT_INDX_OFFSET, 1); - R600_OUT_BATCH(start); - R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_AUTO,1)); + if (num_indices > 0xffff) + { + total_emit += num_indices; + } + else + { + total_emit += (num_indices + 1) / 2; + } + SETfield(vgt_draw_initiator, DI_SRC_SEL_IMMEDIATE, SOURCE_SELECT_shift, SOURCE_SELECT_mask); + } + + total_emit += 3 /* VGT_PRIMITIVE_TYPE */ + + 2 /* VGT_INDEX_TYPE */ + + 2 /* NUM_INSTANCES */ + + 3; /* DRAW */ + + BEGIN_BATCH_NO_AUTOSTATE(total_emit); + // prim + R600_OUT_BATCH_REGSEQ(VGT_PRIMITIVE_TYPE, 1); + R600_OUT_BATCH(vgt_primitive_type); + // index type + R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0)); + R600_OUT_BATCH(vgt_index_type); + // num instances + R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0)); + R600_OUT_BATCH(1); + // draw packet + if(start == 0) + { + R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_AUTO, 1)); R600_OUT_BATCH(vgt_num_indices); R600_OUT_BATCH(vgt_draw_initiator); } + else + { + if (num_indices > 0xffff) + { + R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (num_indices + 1))); + R600_OUT_BATCH(vgt_num_indices); + R600_OUT_BATCH(vgt_draw_initiator); + for (i = start; i < (start + num_indices); i++) + { + R600_OUT_BATCH(i); + } + } + else + { + R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (((num_indices + 1) / 2) + 1))); + R600_OUT_BATCH(vgt_num_indices); + R600_OUT_BATCH(vgt_draw_initiator); + for (i = start; i < (start + num_indices); i += 2) + { + if ((i + 1) == (start + num_indices)) + { + R600_OUT_BATCH(i); + } + else + { + R600_OUT_BATCH(((i + 1) << 16) | (i)); + } + } + } + } END_BATCH(); COMMIT_BATCH(); @@ -379,11 +438,16 @@ static GLuint r700PredictRenderSize(GLcontext* ctx, dwords = PRE_EMIT_STATE_BUFSZ; if (ib) - dwords += nr_prims * 17; + dwords += nr_prims * 14; else { for (i = 0; i < nr_prims; ++i) { - dwords += 13; + if (prim[i].start == 0) + dwords += 10; + else if (prim[i].count > 0xffff) + dwords += prim[i].count + 10; + else + dwords += ((prim[i].count + 1) / 2) + 10; } } @@ -840,10 +904,16 @@ static GLboolean r700TryDrawPrims(GLcontext *ctx, radeon_debug_add_indent(); for (i = 0; i < nr_prims; ++i) { - r700RunRenderPrimitive(ctx, - prim[i].start, - prim[i].start + prim[i].count, - prim[i].mode); + if (context->ind_buf.bo) + r700RunRenderPrimitive(ctx, + prim[i].start, + prim[i].start + prim[i].count, + prim[i].mode); + else + r700RunRenderPrimitiveImmediate(ctx, + prim[i].start, + prim[i].start + prim[i].count, + prim[i].mode); } radeon_debug_remove_indent();