[Mesa-dev] [PATCH 7/8] i965/fs: Add support for TXD with shadow comparisons on gen4-6.
Kenneth Graunke
kenneth at whitecape.org
Wed Jun 15 01:24:54 PDT 2011
Gen4-6 don't have a sample_d_c message, so we have to do a regular
sample_d and emit instructions to manually perform the comparison.
This requires a state dependent recompile whenever ctx->Depth.Func
changes. do_wm_prog looks for a compiled program in the cache based off
of brw_wm_prog_key, and if it doesn't find one, recompiles. So we
simply need to add the depth comparison function to the key; the
_NEW_DEPTH dirty bit was already in-place.
Of course, this is unnecessary in every other case, so I leave
key->depth_comparison_func as 0 unless it's actually relevant.
Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
---
src/mesa/drivers/dri/i965/brw_fs_emit.cpp | 4 +-
src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 51 ++++++++++++++++++++++++--
src/mesa/drivers/dri/i965/brw_wm.h | 1 +
3 files changed, 51 insertions(+), 5 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
index 636ef4e..8bd8b01 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
@@ -276,7 +276,7 @@ fs_visitor::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src)
if (inst->shadow_compare && intel->gen >= 7) {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_DERIVS_COMPARE;
} else {
- assert(intel->gen >= 7 || !inst->shadow_compare); // shadow comparisons not supported on gen5-6 yet.
+ /* TXD with shadow comparison is handled specially pre-Ivybridge */
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS;
}
break;
@@ -316,7 +316,7 @@ fs_visitor::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src)
}
break;
case FS_OPCODE_TXD:
- assert(!inst->shadow_compare); // not supported yet
+ /* TXD with shadow comparison is handled specially pre-Ivybridge */
assert(inst->mlen == 7 || inst->mlen == 10);
msg_type = BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_GRADIENTS;
break;
diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
index 03687ce..67344bd 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
@@ -744,7 +744,7 @@ fs_visitor::emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,
}
mlen += ir->coordinate->type->vector_elements * reg_width;
- if (ir->shadow_comparitor) {
+ if (ir->shadow_comparitor && ir->op != ir_txd) {
mlen = MAX2(mlen, header_present + 4 * reg_width);
this->result = reg_undef;
@@ -934,6 +934,21 @@ fs_visitor::visit(ir_texture *ir)
int sampler = _mesa_get_sampler_uniform_value(ir->sampler, prog, &fp->Base);
sampler = fp->Base.SamplerUnits[sampler];
+ /* Pre-Ivybridge doesn't have a sample_d_c message, so shadow compares
+ * for textureGrad/TXD need to be emulated with instructions.
+ */
+ bool hw_compare_supported = ir->op != ir_txd || intel->gen > 7;
+ if (ir->shadow_comparitor && !hw_compare_supported) {
+ /* Mark that this program is only valid for the current glDepthFunc */
+ c->key.depth_compare_func = ctx->Depth.Func;
+
+ /* No need to even sample for GL_ALWAYS or GL_NEVER...bail early */
+ if (ctx->Depth.Func == GL_ALWAYS)
+ return swizzle_shadow_result(ir, fs_reg(1.0f), sampler);
+ else if (ctx->Depth.Func == GL_NEVER)
+ return swizzle_shadow_result(ir, fs_reg(0.0f), sampler);
+ }
+
this->result = reg_undef;
ir->coordinate->accept(this);
fs_reg coordinate = this->result;
@@ -1044,8 +1059,38 @@ fs_visitor::visit(ir_texture *ir)
this->result = dst;
- if (ir->shadow_comparitor)
- inst->shadow_compare = true;
+ if (ir->shadow_comparitor) {
+ if (hw_compare_supported) {
+ inst->shadow_compare = true;
+ } else {
+ fs_reg result = fs_reg(this, glsl_type::vec4_type);
+ fs_reg z = coordinate;
+ z.reg_offset = 2;
+
+ /* FINISHME: This needs to be done pre-filtering. */
+
+ uint32_t conditional = 0;
+ switch (ctx->Depth.Func) {
+ /* GL_ALWAYS and GL_NEVER were handled at the top of the function */
+ case GL_LESS: conditional = BRW_CONDITIONAL_L; break;
+ case GL_GREATER: conditional = BRW_CONDITIONAL_G; break;
+ case GL_LEQUAL: conditional = BRW_CONDITIONAL_LE; break;
+ case GL_GEQUAL: conditional = BRW_CONDITIONAL_GE; break;
+ case GL_EQUAL: conditional = BRW_CONDITIONAL_EQ; break;
+ case GL_NOTEQUAL: conditional = BRW_CONDITIONAL_NEQ; break;
+ default: assert(!"Should not get here: bad ctx->Depth.Func");
+ }
+
+ /* r = (coordinate.z < sample result) ? 1.0 : 0.0 */
+ inst = emit(BRW_OPCODE_MOV, result, fs_reg(0.0f));
+ inst = emit(BRW_OPCODE_CMP, reg_null_f, z, dst);
+ inst->conditional_mod = conditional;
+ inst = emit(BRW_OPCODE_MOV, result, fs_reg(1.0f));
+ inst->predicated = true;
+
+ dst = result;
+ }
+ }
swizzle_shadow_result(ir, dst, sampler);
}
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index e244b55..0b284b6 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -67,6 +67,7 @@ struct brw_wm_prog_key {
GLuint alpha_test:1;
GLuint clamp_fragment_color:1;
GLuint line_aa:2;
+ uint8_t depth_compare_func; /* 0 if irrelevant, GL_LESS etc. otherwise */
GLbitfield proj_attrib_mask; /**< one bit per fragment program attribute */
GLuint yuvtex_mask:16;
--
1.7.5.4
More information about the mesa-dev
mailing list