diff options
Diffstat (limited to 'src/gallium/auxiliary')
-rw-r--r-- | src/gallium/auxiliary/cso_cache/cso_context.c | 68 | ||||
-rw-r--r-- | src/gallium/auxiliary/cso_cache/cso_context.h | 8 | ||||
-rw-r--r-- | src/gallium/auxiliary/gallivm/lp_bld_depth.c | 467 | ||||
-rw-r--r-- | src/gallium/auxiliary/gallivm/lp_bld_depth.h | 5 | ||||
-rw-r--r-- | src/gallium/auxiliary/util/u_format.csv | 1 | ||||
-rw-r--r-- | src/gallium/auxiliary/util/u_format_tests.c | 7 | ||||
-rw-r--r-- | src/gallium/auxiliary/util/u_gen_mipmap.c | 1 | ||||
-rw-r--r-- | src/gallium/auxiliary/util/u_pack_color.h | 19 | ||||
-rw-r--r-- | src/gallium/auxiliary/util/u_tile.c | 55 |
9 files changed, 437 insertions, 194 deletions
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c index 4ed9e09c52..d6f8dd34bf 100644 --- a/src/gallium/auxiliary/cso_cache/cso_context.c +++ b/src/gallium/auxiliary/cso_cache/cso_context.c @@ -618,74 +618,6 @@ cso_restore_vertex_samplers(struct cso_context *ctx) } -enum pipe_error cso_set_sampler_textures( struct cso_context *ctx, - uint count, - struct pipe_texture **textures ) -{ - uint i; - - ctx->nr_fragment_sampler_views = count; - - for (i = 0; i < count; i++) { - struct pipe_sampler_view templ, *view; - - u_sampler_view_default_template(&templ, - textures[i], - textures[i]->format); - - view = ctx->pipe->create_sampler_view(ctx->pipe, - textures[i], - &templ); - - pipe_sampler_view_reference(&ctx->fragment_sampler_views[i], view); - } - for ( ; i < PIPE_MAX_SAMPLERS; i++) { - pipe_sampler_view_reference(&ctx->fragment_sampler_views[i], NULL); - } - - ctx->pipe->set_fragment_sampler_views(ctx->pipe, - count, - ctx->fragment_sampler_views); - - return PIPE_OK; -} - -void cso_save_sampler_textures( struct cso_context *ctx ) -{ - uint i; - - ctx->nr_fragment_sampler_views_saved = ctx->nr_fragment_sampler_views; - for (i = 0; i < ctx->nr_fragment_sampler_views; i++) { - assert(!ctx->fragment_sampler_views_saved[i]); - - pipe_sampler_view_reference(&ctx->fragment_sampler_views_saved[i], - ctx->fragment_sampler_views[i]); - } -} - -void cso_restore_sampler_textures( struct cso_context *ctx ) -{ - uint i; - - ctx->nr_fragment_sampler_views = ctx->nr_fragment_sampler_views_saved; - - for (i = 0; i < ctx->nr_fragment_sampler_views; i++) { - pipe_sampler_view_reference(&ctx->fragment_sampler_views[i], NULL); - ctx->fragment_sampler_views[i] = ctx->fragment_sampler_views_saved[i]; - ctx->fragment_sampler_views_saved[i] = NULL; - } - for ( ; i < PIPE_MAX_SAMPLERS; i++) { - pipe_sampler_view_reference(&ctx->fragment_sampler_views[i], NULL); - } - - ctx->pipe->set_fragment_sampler_views(ctx->pipe, - ctx->nr_fragment_sampler_views, - ctx->fragment_sampler_views); - - ctx->nr_fragment_sampler_views_saved = 0; -} - - enum pipe_error cso_set_depth_stencil_alpha(struct cso_context *ctx, const struct pipe_depth_stencil_alpha_state *templ) { diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h index a24077e009..d6bcb1fe8f 100644 --- a/src/gallium/auxiliary/cso_cache/cso_context.h +++ b/src/gallium/auxiliary/cso_cache/cso_context.h @@ -103,14 +103,6 @@ void cso_single_vertex_sampler_done(struct cso_context *cso); - -enum pipe_error cso_set_sampler_textures( struct cso_context *cso, - uint count, - struct pipe_texture **textures ); -void cso_save_sampler_textures( struct cso_context *cso ); -void cso_restore_sampler_textures( struct cso_context *cso ); - - enum pipe_error cso_set_vertex_elements(struct cso_context *ctx, unsigned count, const struct pipe_vertex_element *states); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_depth.c b/src/gallium/auxiliary/gallivm/lp_bld_depth.c index e4500e5aef..4ce1a27a06 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_depth.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_depth.c @@ -52,7 +52,14 @@ * Z31 Z32 Z41 Z42 Z33 Z34 Z43 Z44 ... * ... ... ... ... ... ... ... ... ... * - * FIXME: Code generate stencil test + * + * Stencil test: + * Two-sided stencil test is supported but probably not as efficient as + * it could be. Currently, we use if/then/else constructs to do the + * operations for front vs. back-facing polygons. We could probably do + * both the front and back arithmetic then use a Select() instruction to + * choose the result depending on polyon orientation. We'd have to + * measure performance both ways and see which is better. * * @author Jose Fonseca <jfonseca@vmware.com> */ @@ -70,23 +77,35 @@ #include "lp_bld_swizzle.h" +/** Used to select fields from pipe_stencil_state */ +enum stencil_op { + S_FAIL_OP, + Z_FAIL_OP, + Z_PASS_OP +}; + + /** - * Do the stencil test comparison (compare fb Z values against ref value. - * \param stencilVals vector of stencil values from framebuffer + * Do the stencil test comparison (compare FB stencil values against ref value). + * This will be used twice when generating two-sided stencil code. + * \param stencil the front/back stencil state * \param stencilRef the stencil reference value, replicated as a vector - * \return mask of pass/fail values + * \param stencilVals vector of stencil values from framebuffer + * \return vector mask of pass/fail values (~0 or 0) */ static LLVMValueRef -lp_build_stencil_test(struct lp_build_context *bld, - const struct pipe_stencil_state *stencil, - LLVMValueRef stencilRef, - LLVMValueRef stencilVals) +lp_build_stencil_test_single(struct lp_build_context *bld, + const struct pipe_stencil_state *stencil, + LLVMValueRef stencilRef, + LLVMValueRef stencilVals) { const unsigned stencilMax = 255; /* XXX fix */ struct lp_type type = bld->type; LLVMValueRef res; + assert(type.sign); + assert(stencil->enabled); if (stencil->valuemask != stencilMax) { @@ -104,23 +123,101 @@ lp_build_stencil_test(struct lp_build_context *bld, /** + * Do the one or two-sided stencil test comparison. + * \sa lp_build_stencil_test_single + * \param face an integer indicating front (+) or back (-) facing polygon. + * If NULL, assume front-facing. + */ +static LLVMValueRef +lp_build_stencil_test(struct lp_build_context *bld, + const struct pipe_stencil_state stencil[2], + LLVMValueRef stencilRefs[2], + LLVMValueRef stencilVals, + LLVMValueRef face) +{ + LLVMValueRef res; + + assert(stencil[0].enabled); + + if (stencil[1].enabled && face) { + /* do two-sided test */ + struct lp_build_flow_context *flow_ctx; + struct lp_build_if_state if_ctx; + LLVMValueRef front_facing; + LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0); + LLVMValueRef result = bld->undef; + + flow_ctx = lp_build_flow_create(bld->builder); + lp_build_flow_scope_begin(flow_ctx); + + lp_build_flow_scope_declare(flow_ctx, &result); + + /* front_facing = face > 0.0 */ + front_facing = LLVMBuildFCmp(bld->builder, LLVMRealUGT, face, zero, ""); + + lp_build_if(&if_ctx, flow_ctx, bld->builder, front_facing); + { + result = lp_build_stencil_test_single(bld, &stencil[0], + stencilRefs[0], stencilVals); + } + lp_build_else(&if_ctx); + { + result = lp_build_stencil_test_single(bld, &stencil[1], + stencilRefs[1], stencilVals); + } + lp_build_endif(&if_ctx); + + lp_build_flow_scope_end(flow_ctx); + lp_build_flow_destroy(flow_ctx); + + res = result; + } + else { + /* do single-side test */ + res = lp_build_stencil_test_single(bld, &stencil[0], + stencilRefs[0], stencilVals); + } + + return res; +} + + +/** * Apply the stencil operator (add/sub/keep/etc) to the given vector * of stencil values. * \return new stencil values vector */ static LLVMValueRef -lp_build_stencil_op(struct lp_build_context *bld, - const struct pipe_stencil_state *stencil, - unsigned stencil_op, - LLVMValueRef stencilRef, - LLVMValueRef stencilVals, - LLVMValueRef mask) +lp_build_stencil_op_single(struct lp_build_context *bld, + const struct pipe_stencil_state *stencil, + enum stencil_op op, + LLVMValueRef stencilRef, + LLVMValueRef stencilVals, + LLVMValueRef mask) { const unsigned stencilMax = 255; /* XXX fix */ struct lp_type type = bld->type; LLVMValueRef res; LLVMValueRef max = lp_build_const_int_vec(type, stencilMax); + unsigned stencil_op; + + assert(type.sign); + + switch (op) { + case S_FAIL_OP: + stencil_op = stencil->fail_op; + break; + case Z_FAIL_OP: + stencil_op = stencil->zfail_op; + break; + case Z_PASS_OP: + stencil_op = stencil->zpass_op; + break; + default: + assert(0 && "Invalid stencil_op mode"); + stencil_op = PIPE_STENCIL_OP_KEEP; + } switch (stencil_op) { case PIPE_STENCIL_OP_KEEP: @@ -151,6 +248,7 @@ lp_build_stencil_op(struct lp_build_context *bld, break; case PIPE_STENCIL_OP_INVERT: res = LLVMBuildNot(bld->builder, stencilVals, ""); + res = LLVMBuildAnd(bld->builder, res, max, ""); break; default: assert(0 && "bad stencil op mode"); @@ -174,6 +272,63 @@ lp_build_stencil_op(struct lp_build_context *bld, /** + * Do the one or two-sided stencil test op/update. + */ +static LLVMValueRef +lp_build_stencil_op(struct lp_build_context *bld, + const struct pipe_stencil_state stencil[2], + enum stencil_op op, + LLVMValueRef stencilRefs[2], + LLVMValueRef stencilVals, + LLVMValueRef mask, + LLVMValueRef face) + +{ + assert(stencil[0].enabled); + + if (stencil[1].enabled && face) { + /* do two-sided op */ + struct lp_build_flow_context *flow_ctx; + struct lp_build_if_state if_ctx; + LLVMValueRef front_facing; + LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0); + LLVMValueRef result = bld->undef; + + flow_ctx = lp_build_flow_create(bld->builder); + lp_build_flow_scope_begin(flow_ctx); + + lp_build_flow_scope_declare(flow_ctx, &result); + + /* front_facing = face > 0.0 */ + front_facing = LLVMBuildFCmp(bld->builder, LLVMRealUGT, face, zero, ""); + + lp_build_if(&if_ctx, flow_ctx, bld->builder, front_facing); + { + result = lp_build_stencil_op_single(bld, &stencil[0], op, + stencilRefs[0], stencilVals, mask); + } + lp_build_else(&if_ctx); + { + result = lp_build_stencil_op_single(bld, &stencil[1], op, + stencilRefs[1], stencilVals, mask); + } + lp_build_endif(&if_ctx); + + lp_build_flow_scope_end(flow_ctx); + lp_build_flow_destroy(flow_ctx); + + return result; + } + else { + /* do single-sided op */ + return lp_build_stencil_op_single(bld, &stencil[0], op, + stencilRefs[0], stencilVals, mask); + } +} + + + +/** * Return a type appropriate for depth/stencil testing. */ struct lp_type @@ -213,36 +368,96 @@ lp_depth_type(const struct util_format_description *format_desc, } -static LLVMValueRef -lp_build_get_stencil_ref(struct lp_build_context *bld, - struct lp_type type, LLVMValueRef stencil_refs_ptr) +/** + * Compute bitmask and bit shift to apply to the incoming fragment Z values + * and the Z buffer values needed before doing the Z comparison. + * + * Note that we leave the Z bits in the position that we find them + * in the Z buffer (typically 0xffffff00 or 0x00ffffff). That lets us + * get by with fewer bit twiddling steps. + */ +static boolean +get_z_shift_and_mask(const struct util_format_description *format_desc, + unsigned *shift, unsigned *mask) +{ + const unsigned total_bits = format_desc->block.bits; + unsigned z_swizzle; + int chan; + unsigned padding_left, padding_right; + + assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS); + assert(format_desc->block.width == 1); + assert(format_desc->block.height == 1); + + z_swizzle = format_desc->swizzle[0]; + + if (z_swizzle == UTIL_FORMAT_SWIZZLE_NONE) + return FALSE; + + padding_right = 0; + for (chan = 0; chan < z_swizzle; ++chan) + padding_right += format_desc->channel[chan].size; + + padding_left = + total_bits - (padding_right + format_desc->channel[z_swizzle].size); + + if (padding_left || padding_right) { + unsigned long long mask_left = (1ULL << (total_bits - padding_left)) - 1; + unsigned long long mask_right = (1ULL << (padding_right)) - 1; + *mask = mask_left ^ mask_right; + } + else { + *mask = 0xffffffff; + } + + *shift = padding_left; + + return TRUE; +} + + +/** + * Compute bitmask and bit shift to apply to the framebuffer pixel values + * to put the stencil bits in the least significant position. + * (i.e. 0x000000ff) + */ +static boolean +get_s_shift_and_mask(const struct util_format_description *format_desc, + unsigned *shift, unsigned *mask) { - LLVMValueRef indexes[2], ptr, ref, ref_vec; + unsigned s_swizzle; + int chan, sz; + + s_swizzle = format_desc->swizzle[1]; - /* load 0th element of the array */ - indexes[0] = indexes[1] = LLVMConstInt(LLVMInt32Type(), 0, 0); - ptr = LLVMBuildGEP(bld->builder, stencil_refs_ptr, indexes, 2, ""); - ref = LLVMBuildLoad(bld->builder, ptr, ""); + if (s_swizzle == UTIL_FORMAT_SWIZZLE_NONE) + return FALSE; - /* convert int8 value to i32 */ - ref = LLVMBuildZExt(bld->builder, ref, LLVMIntType(type.width), ""); + *shift = 0; + for (chan = 0; chan < s_swizzle; chan++) + *shift += format_desc->channel[chan].size; - /* make scalar into vector */ - ref_vec = lp_build_broadcast_scalar(bld, ref); + sz = format_desc->channel[s_swizzle].size; + *mask = (1U << sz) - 1U; - return ref_vec; + return TRUE; } + /** * Generate code for performing depth and/or stencil tests. * We operate on a vector of values (typically a 2x2 quad). * + * \param depth the depth test state + * \param stencil the front/back stencil state * \param type the data type of the fragment depth/stencil values * \param format_desc description of the depth/stencil surface - * \param mask the alive/dead pixel mask for the quad - * \param src the incoming depth/stencil values (a 2x2 quad) - * \param dst_ptr the outgoing/updated depth/stencil values + * \param mask the alive/dead pixel mask for the quad (vector) + * \param stencil_refs the front/back stencil ref values (scalar) + * \param z_src the incoming depth/stencil values (a 2x2 quad) + * \param zs_dst_ptr pointer to depth/stencil values in framebuffer + * \param facing contains float value indicating front/back facing polygon */ void lp_build_depth_stencil_test(LLVMBuilderRef builder, @@ -251,119 +466,131 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder, struct lp_type type, const struct util_format_description *format_desc, struct lp_build_mask_context *mask, - LLVMValueRef stencil_refs, + LLVMValueRef stencil_refs[2], LLVMValueRef z_src, - LLVMValueRef zs_dst_ptr) + LLVMValueRef zs_dst_ptr, + LLVMValueRef face) { struct lp_build_context bld; - unsigned z_swizzle, s_swizzle; + struct lp_build_context sbld; + struct lp_type s_type; LLVMValueRef zs_dst, z_dst = NULL; LLVMValueRef stencil_vals = NULL; - LLVMValueRef z_bitmask = NULL, s_bitmask = NULL; + LLVMValueRef z_bitmask = NULL, stencil_shift = NULL; LLVMValueRef z_pass = NULL, s_pass_mask = NULL; LLVMValueRef orig_mask = mask->value; - assert(depth->enabled || stencil[0].enabled); + /* Sanity checking */ + { + const unsigned z_swizzle = format_desc->swizzle[0]; + const unsigned s_swizzle = format_desc->swizzle[1]; - assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS); - assert(format_desc->block.width == 1); - assert(format_desc->block.height == 1); + assert(z_swizzle != UTIL_FORMAT_SWIZZLE_NONE || + s_swizzle != UTIL_FORMAT_SWIZZLE_NONE); - z_swizzle = format_desc->swizzle[0]; - s_swizzle = format_desc->swizzle[1]; + assert(depth->enabled || stencil[0].enabled); - assert(z_swizzle != UTIL_FORMAT_SWIZZLE_NONE || - s_swizzle != UTIL_FORMAT_SWIZZLE_NONE); + assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS); + assert(format_desc->block.width == 1); + assert(format_desc->block.height == 1); - /* Sanity checking */ - assert(z_swizzle < 4); - assert(format_desc->block.bits == type.width); - if(type.floating) { - assert(z_swizzle == 0); - assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_FLOAT); - assert(format_desc->channel[z_swizzle].size == format_desc->block.bits); - } - else { - assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED); - assert(format_desc->channel[z_swizzle].normalized); - assert(!type.fixed); - assert(!type.sign); - assert(type.norm); + if (stencil[0].enabled) { + assert(format_desc->format == PIPE_FORMAT_Z24S8_UNORM || + format_desc->format == PIPE_FORMAT_S8Z24_UNORM); + } + + assert(z_swizzle < 4); + assert(format_desc->block.bits == type.width); + if (type.floating) { + assert(z_swizzle == 0); + assert(format_desc->channel[z_swizzle].type == + UTIL_FORMAT_TYPE_FLOAT); + assert(format_desc->channel[z_swizzle].size == + format_desc->block.bits); + } + else { + assert(format_desc->channel[z_swizzle].type == + UTIL_FORMAT_TYPE_UNSIGNED); + assert(format_desc->channel[z_swizzle].normalized); + assert(!type.fixed); + assert(!type.sign); + assert(type.norm); + } } - /* Setup build context */ + + /* Setup build context for Z vals */ lp_build_context_init(&bld, builder, type); + /* Setup build context for stencil vals */ + s_type = lp_type_int_vec(type.width); + lp_build_context_init(&sbld, builder, s_type); + /* Load current z/stencil value from z/stencil buffer */ zs_dst = LLVMBuildLoad(builder, zs_dst_ptr, ""); lp_build_name(zs_dst, "zsbufval"); - /* Align the source depth bits with the destination's, and mask out any - * stencil or padding bits from both */ - if(format_desc->channel[z_swizzle].size == format_desc->block.bits) { - assert(z_swizzle == 0); - z_dst = zs_dst; - } - else { - /* shift/mask bits to right-justify the Z bits */ - unsigned padding_left; - unsigned padding_right; - unsigned chan; - - assert(format_desc->layout == UTIL_FORMAT_LAYOUT_PLAIN); - assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED); - assert(format_desc->channel[z_swizzle].size <= format_desc->block.bits); - assert(format_desc->channel[z_swizzle].normalized); - - padding_right = 0; - for(chan = 0; chan < z_swizzle; ++chan) - padding_right += format_desc->channel[chan].size; - padding_left = format_desc->block.bits - - (padding_right + format_desc->channel[z_swizzle].size); - - if(padding_left || padding_right) { - const unsigned long long mask_left = (1ULL << (format_desc->block.bits - padding_left)) - 1; - const unsigned long long mask_right = (1ULL << (padding_right)) - 1; - z_bitmask = lp_build_const_int_vec(type, mask_left ^ mask_right); - } - - s_bitmask = LLVMBuildNot(builder, z_bitmask, ""); - stencil_vals = LLVMBuildAnd(builder, zs_dst, s_bitmask, ""); + /* Compute and apply the Z/stencil bitmasks and shifts. + */ + { + unsigned z_shift, z_mask; + unsigned s_shift, s_mask; + + if (get_z_shift_and_mask(format_desc, &z_shift, &z_mask)) { + if (z_shift) { + LLVMValueRef shift = lp_build_const_int_vec(type, z_shift); + z_src = LLVMBuildLShr(builder, z_src, shift, ""); + } + + if (z_mask != 0xffffffff) { + LLVMValueRef mask = lp_build_const_int_vec(type, z_mask); + z_src = LLVMBuildAnd(builder, z_src, mask, ""); + z_dst = LLVMBuildAnd(builder, zs_dst, mask, ""); + z_bitmask = mask; /* used below */ + } + else { + z_dst = zs_dst; + } + + lp_build_name(z_dst, "zsbuf.z"); + } - if(padding_left) - z_src = LLVMBuildLShr(builder, z_src, - lp_build_const_int_vec(type, padding_left), ""); - if(padding_right) - z_src = LLVMBuildAnd(builder, z_src, z_bitmask, ""); - if(padding_left || padding_right) - z_dst = LLVMBuildAnd(builder, zs_dst, z_bitmask, ""); - else - z_dst = zs_dst; + if (get_s_shift_and_mask(format_desc, &s_shift, &s_mask)) { + if (s_shift) { + LLVMValueRef shift = lp_build_const_int_vec(type, s_shift); + stencil_vals = LLVMBuildLShr(builder, zs_dst, shift, ""); + stencil_shift = shift; /* used below */ + } + else { + stencil_vals = zs_dst; + } + + if (s_mask != 0xffffffff) { + LLVMValueRef mask = lp_build_const_int_vec(type, s_mask); + stencil_vals = LLVMBuildAnd(builder, stencil_vals, mask, ""); + } + + lp_build_name(stencil_vals, "stencil"); + } } - lp_build_name(z_dst, "zsbuf.z"); - - /* - printf("build depth %d stencil %d\n", - depth->enabled, - stencil[0].enabled); - */ if (stencil[0].enabled) { - /* Incoming stencil_refs is ptr to int8[2]. Get/convert to int32[4]. */ - stencil_refs = lp_build_get_stencil_ref(&bld, type, stencil_refs); + /* convert scalar stencil refs into vectors */ + stencil_refs[0] = lp_build_broadcast_scalar(&bld, stencil_refs[0]); + stencil_refs[1] = lp_build_broadcast_scalar(&bld, stencil_refs[1]); - s_pass_mask = lp_build_stencil_test(&bld, stencil, - stencil_refs, stencil_vals); + s_pass_mask = lp_build_stencil_test(&sbld, stencil, + stencil_refs, stencil_vals, face); /* apply stencil-fail operator */ { LLVMValueRef s_fail_mask = lp_build_andc(&bld, orig_mask, s_pass_mask); - stencil_vals = lp_build_stencil_op(&bld, stencil, stencil[0].fail_op, + stencil_vals = lp_build_stencil_op(&sbld, stencil, S_FAIL_OP, stencil_refs, stencil_vals, - s_fail_mask); + s_fail_mask, face); } } @@ -394,15 +621,15 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder, /* apply Z-fail operator */ z_fail_mask = lp_build_andc(&bld, orig_mask, z_pass); - stencil_vals = lp_build_stencil_op(&bld, stencil, stencil[0].zfail_op, + stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_FAIL_OP, stencil_refs, stencil_vals, - z_fail_mask); + z_fail_mask, face); /* apply Z-pass operator */ z_pass_mask = LLVMBuildAnd(bld.builder, orig_mask, z_pass, ""); - stencil_vals = lp_build_stencil_op(&bld, stencil, stencil[0].zpass_op, + stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_PASS_OP, stencil_refs, stencil_vals, - z_pass_mask); + z_pass_mask, face); } } else { @@ -410,10 +637,18 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder, * passed the stencil test. */ s_pass_mask = LLVMBuildAnd(bld.builder, orig_mask, s_pass_mask, ""); - stencil_vals = lp_build_stencil_op(&bld, stencil, stencil[0].zpass_op, - stencil_refs, stencil_vals, s_pass_mask); + stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_PASS_OP, + stencil_refs, stencil_vals, + s_pass_mask, face); } + /* The Z bits are already in the right place but we may need to shift the + * stencil bits before ORing Z with Stencil to make the final pixel value. + */ + if (stencil_vals && stencil_shift) + stencil_vals = LLVMBuildShl(bld.builder, stencil_vals, + stencil_shift, ""); + /* Finally, merge/store the z/stencil values */ if ((depth->enabled && depth->writemask) || (stencil[0].enabled && stencil[0].writemask)) { diff --git a/src/gallium/auxiliary/gallivm/lp_bld_depth.h b/src/gallium/auxiliary/gallivm/lp_bld_depth.h index eedc1e419b..27dd46b625 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_depth.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_depth.h @@ -57,9 +57,10 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder, struct lp_type type, const struct util_format_description *format_desc, struct lp_build_mask_context *mask, - LLVMValueRef stencil_refs, + LLVMValueRef stencil_refs[2], LLVMValueRef zs_src, - LLVMValueRef zs_dst_ptr); + LLVMValueRef zs_dst_ptr, + LLVMValueRef facing); #endif /* !LP_BLD_DEPTH_H */ diff --git a/src/gallium/auxiliary/util/u_format.csv b/src/gallium/auxiliary/util/u_format.csv index 96a0fa6550..11243e7349 100644 --- a/src/gallium/auxiliary/util/u_format.csv +++ b/src/gallium/auxiliary/util/u_format.csv @@ -63,6 +63,7 @@ PIPE_FORMAT_A8R8G8B8_UNORM , plain, 1, 1, un8 , un8 , un8 , un8 , yzwx, r PIPE_FORMAT_X8R8G8B8_UNORM , plain, 1, 1, un8 , un8 , un8 , un8 , yzw1, rgb PIPE_FORMAT_A8B8G8R8_UNORM , plain, 1, 1, un8 , un8 , un8 , un8 , wzyx, rgb PIPE_FORMAT_X8B8G8R8_UNORM , plain, 1, 1, un8 , un8 , un8 , un8 , wzy1, rgb +PIPE_FORMAT_B5G5R5X1_UNORM , plain, 1, 1, un5 , un5 , un5 , un1 , zyx1, rgb PIPE_FORMAT_B5G5R5A1_UNORM , plain, 1, 1, un5 , un5 , un5 , un1 , zyxw, rgb PIPE_FORMAT_B4G4R4A4_UNORM , plain, 1, 1, un4 , un4 , un4 , un4 , zyxw, rgb PIPE_FORMAT_B5G6R5_UNORM , plain, 1, 1, un5 , un6 , un5 , , zyx1, rgb diff --git a/src/gallium/auxiliary/util/u_format_tests.c b/src/gallium/auxiliary/util/u_format_tests.c index 182a474044..9d6debcd8c 100644 --- a/src/gallium/auxiliary/util/u_format_tests.c +++ b/src/gallium/auxiliary/util/u_format_tests.c @@ -120,6 +120,13 @@ util_format_test_cases[] = * 16-bit rendertarget formats */ + {PIPE_FORMAT_B5G5R5X1_UNORM, PACKED_1x16(0x7fff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_B5G5R5X1_UNORM, PACKED_1x16(0x7fff), PACKED_1x16(0x001f), {0.0, 0.0, 1.0, 0.0}}, + {PIPE_FORMAT_B5G5R5X1_UNORM, PACKED_1x16(0x7fff), PACKED_1x16(0x03e0), {0.0, 1.0, 0.0, 0.0}}, + {PIPE_FORMAT_B5G5R5X1_UNORM, PACKED_1x16(0x7fff), PACKED_1x16(0x7c00), {1.0, 0.0, 0.0, 0.0}}, + {PIPE_FORMAT_B5G5R5X1_UNORM, PACKED_1x16(0x7fff), PACKED_1x16(0x8000), {0.0, 0.0, 0.0, 1.0}}, + {PIPE_FORMAT_B5G5R5X1_UNORM, PACKED_1x16(0x7fff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}}, + {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}}, {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x001f), {0.0, 0.0, 1.0, 0.0}}, {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x03e0), {0.0, 1.0, 0.0, 0.0}}, diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c index 61d64cff6d..509d38754f 100644 --- a/src/gallium/auxiliary/util/u_gen_mipmap.c +++ b/src/gallium/auxiliary/util/u_gen_mipmap.c @@ -938,6 +938,7 @@ format_to_type_comps(enum pipe_format pformat, *datatype = DTYPE_UBYTE; *comps = 4; return; + case PIPE_FORMAT_B5G5R5X1_UNORM: case PIPE_FORMAT_B5G5R5A1_UNORM: *datatype = DTYPE_USHORT_1_5_5_5_REV; *comps = 4; diff --git a/src/gallium/auxiliary/util/u_pack_color.h b/src/gallium/auxiliary/util/u_pack_color.h index 50f1b1670b..c5fd7a6783 100644 --- a/src/gallium/auxiliary/util/u_pack_color.h +++ b/src/gallium/auxiliary/util/u_pack_color.h @@ -92,6 +92,11 @@ util_pack_color_ub(ubyte r, ubyte g, ubyte b, ubyte a, uc->us = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3); } return; + case PIPE_FORMAT_B5G5R5X1_UNORM: + { + uc->us = ((0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3); + } + return; case PIPE_FORMAT_B5G5R5A1_UNORM: { uc->us = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3); @@ -216,6 +221,15 @@ util_unpack_color_ub(enum pipe_format format, union util_color *uc, *a = (ubyte) 0xff; } return; + case PIPE_FORMAT_B5G5R5X1_UNORM: + { + ushort p = uc->us; + *r = (ubyte) (((p >> 7) & 0xf8) | ((p >> 12) & 0x7)); + *g = (ubyte) (((p >> 2) & 0xf8) | ((p >> 7) & 0x7)); + *b = (ubyte) (((p << 3) & 0xf8) | ((p >> 2) & 0x7)); + *a = (ubyte) 0xff; + } + return; case PIPE_FORMAT_B5G5R5A1_UNORM: { ushort p = uc->us; @@ -361,6 +375,11 @@ util_pack_color(const float rgba[4], enum pipe_format format, union util_color * uc->us = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3); } return; + case PIPE_FORMAT_B5G5R5X1_UNORM: + { + uc->us = ((0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3); + } + return; case PIPE_FORMAT_B5G5R5A1_UNORM: { uc->us = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3); diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c index 82e44192aa..09b2382733 100644 --- a/src/gallium/auxiliary/util/u_tile.c +++ b/src/gallium/auxiliary/util/u_tile.c @@ -295,6 +295,55 @@ r8g8b8a8_put_tile_rgba(unsigned *dst, } +/*** PIPE_FORMAT_B5G5R5X1_UNORM ***/ + +static void +x1r5g5b5_get_tile_rgba(const ushort *src, + unsigned w, unsigned h, + float *p, + unsigned dst_stride) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + float *pRow = p; + for (j = 0; j < w; j++, pRow += 4) { + const ushort pixel = *src++; + pRow[0] = ((pixel >> 10) & 0x1f) * (1.0f / 31.0f); + pRow[1] = ((pixel >> 5) & 0x1f) * (1.0f / 31.0f); + pRow[2] = ((pixel ) & 0x1f) * (1.0f / 31.0f); + pRow[3] = 1.0f; + } + p += dst_stride; + } +} + + +static void +x1r5g5b5_put_tile_rgba(ushort *dst, + unsigned w, unsigned h, + const float *p, + unsigned src_stride) +{ + unsigned i, j; + + for (i = 0; i < h; i++) { + const float *pRow = p; + for (j = 0; j < w; j++, pRow += 4) { + unsigned r, g, b; + r = float_to_ubyte(pRow[0]); + g = float_to_ubyte(pRow[1]); + b = float_to_ubyte(pRow[2]); + r = r >> 3; /* 5 bits */ + g = g >> 3; /* 5 bits */ + b = b >> 3; /* 5 bits */ + *dst++ = (1 << 15) | (r << 10) | (g << 5) | b; + } + p += src_stride; + } +} + + /*** PIPE_FORMAT_B5G5R5A1_UNORM ***/ static void @@ -1174,6 +1223,9 @@ pipe_tile_raw_to_rgba(enum pipe_format format, case PIPE_FORMAT_A8B8G8R8_UNORM: r8g8b8a8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride); break; + case PIPE_FORMAT_B5G5R5X1_UNORM: + x1r5g5b5_get_tile_rgba((ushort *) src, w, h, dst, dst_stride); + break; case PIPE_FORMAT_B5G5R5A1_UNORM: a1r5g5b5_get_tile_rgba((ushort *) src, w, h, dst, dst_stride); break; @@ -1368,6 +1420,9 @@ pipe_put_tile_rgba(struct pipe_context *pipe, case PIPE_FORMAT_A8B8G8R8_UNORM: r8g8b8a8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride); break; + case PIPE_FORMAT_B5G5R5X1_UNORM: + x1r5g5b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride); + break; case PIPE_FORMAT_B5G5R5A1_UNORM: a1r5g5b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride); break; |