[Mesa-dev] [PATCH 4/4] radeonsi: add support for interpolateAt functions

Dave Airlie airlied at gmail.com
Tue Jul 21 15:51:37 PDT 2015


From: Dave Airlie <airlied at redhat.com>

This is part of ARB_gpu_shader5, and this passes
all the piglit tests currently available.

Signed-off-by: Dave Airlie <airlied at redhat.com>
---
 docs/GL3.txt                             |   2 +-
 src/gallium/drivers/radeonsi/si_shader.c | 232 ++++++++++++++++++++++++++++++-
 2 files changed, 232 insertions(+), 2 deletions(-)

diff --git a/docs/GL3.txt b/docs/GL3.txt
index 4f6c415..d74ae63 100644
--- a/docs/GL3.txt
+++ b/docs/GL3.txt
@@ -107,7 +107,7 @@ GL 4.0, GLSL 4.00:
   - Geometry shader instancing                         DONE (r600, radeonsi, llvmpipe, softpipe)
   - Geometry shader multiple streams                   DONE ()
   - Enhanced per-sample shading                        DONE (r600, radeonsi)
-  - Interpolation functions                            DONE (r600)
+  - Interpolation functions                            DONE (r600, radeonsi)
   - New overload resolution rules                      DONE
   GL_ARB_gpu_shader_fp64                               DONE (nvc0, radeonsi, llvmpipe, softpipe)
   GL_ARB_sample_shading                                DONE (i965, nv50, nvc0, r600, radeonsi)
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index c5d80f0..0c01c90 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2263,6 +2263,225 @@ static void si_llvm_emit_ddxy(
 	emit_data->output[0] = lp_build_gather_values(gallivm, result, 4);
 }
 
+/* return 4 values - v2i32 DDX, v2i32 DDY */
+static LLVMValueRef si_llvm_emit_ddxy_interp(
+	struct lp_build_tgsi_context * bld_base,
+	LLVMValueRef interp_ij)
+{
+	struct si_shader_context *si_shader_ctx = si_shader_context(bld_base);
+	struct gallivm_state *gallivm = bld_base->base.gallivm;
+	struct lp_build_context * base = &bld_base->base;
+	LLVMValueRef indices[2];
+	LLVMValueRef store_ptr, load_ptr_x, load_ptr_y, load_ptr_ddx, load_ptr_ddy, temp, temp2;
+	LLVMValueRef tl, tr, bl, result[4];
+	LLVMTypeRef i32;
+	unsigned c;
+
+	i32 = LLVMInt32TypeInContext(gallivm->context);
+
+	indices[0] = bld_base->uint_bld.zero;
+	indices[1] = build_intrinsic(gallivm->builder, "llvm.SI.tid", i32,
+				     NULL, 0, LLVMReadNoneAttribute);
+	store_ptr = LLVMBuildGEP(gallivm->builder, si_shader_ctx->ddxy_lds,
+				 indices, 2, "");
+
+        temp = LLVMBuildAnd(gallivm->builder, indices[1],
+			    lp_build_const_int32(gallivm, 0xfffffffe), "");
+
+	temp2 = LLVMBuildAnd(gallivm->builder, indices[1],
+			     lp_build_const_int32(gallivm, 0xfffffffd), "");
+
+	indices[1] = temp;
+	load_ptr_x = LLVMBuildGEP(gallivm->builder, si_shader_ctx->ddxy_lds,
+				  indices, 2, "");
+
+	indices[1] = temp2;
+	load_ptr_y = LLVMBuildGEP(gallivm->builder, si_shader_ctx->ddxy_lds,
+				  indices, 2, "");
+
+	indices[1] = LLVMBuildAdd(gallivm->builder, temp,
+				  lp_build_const_int32(gallivm,
+						       1),
+				  "");
+	load_ptr_ddx = LLVMBuildGEP(gallivm->builder, si_shader_ctx->ddxy_lds,
+				 indices, 2, "");
+
+	indices[1] = LLVMBuildAdd(gallivm->builder, temp2,
+				  lp_build_const_int32(gallivm,
+						       2),
+				  "");
+	load_ptr_ddy = LLVMBuildGEP(gallivm->builder, si_shader_ctx->ddxy_lds,
+				 indices, 2, "");
+
+	for (c = 0; c < 2; ++c) {
+		LLVMValueRef store_val;
+		LLVMValueRef c_ll = lp_build_const_int32(gallivm, c);
+
+		store_val = LLVMBuildExtractElement(gallivm->builder,
+						    interp_ij, c_ll, "");
+		LLVMBuildStore(gallivm->builder,
+			       store_val,
+			       store_ptr);
+
+		tl = LLVMBuildLoad(gallivm->builder, load_ptr_x, "");
+		tl = LLVMBuildBitCast(gallivm->builder, tl, base->elem_type, "");
+
+		tr = LLVMBuildLoad(gallivm->builder, load_ptr_ddx, "");
+		tr = LLVMBuildBitCast(gallivm->builder, tr, base->elem_type, "");
+
+		result[c] = LLVMBuildFSub(gallivm->builder, tr, tl, "");
+
+		tl = LLVMBuildLoad(gallivm->builder, load_ptr_y, "");
+		tl = LLVMBuildBitCast(gallivm->builder, tl, base->elem_type, "");
+
+		bl = LLVMBuildLoad(gallivm->builder, load_ptr_ddy, "");
+		bl = LLVMBuildBitCast(gallivm->builder, bl, base->elem_type, "");
+
+		result[c + 2] = LLVMBuildFSub(gallivm->builder, bl, tl, "");
+	}
+
+	return lp_build_gather_values(gallivm, result, 4);
+}
+
+static void interp_fetch_args(
+	struct lp_build_tgsi_context * bld_base,
+	struct lp_build_emit_data * emit_data)
+{
+	struct si_shader_context *si_shader_ctx = si_shader_context(bld_base);
+	struct gallivm_state *gallivm = bld_base->base.gallivm;
+	const struct tgsi_full_instruction * inst = emit_data->inst;
+
+	if (inst->Instruction.Opcode == TGSI_OPCODE_INTERP_OFFSET) {
+		/* offset is in second src, first two channels */
+		emit_data->args[0] = lp_build_emit_fetch(bld_base,
+							 emit_data->inst, 1,
+							 0);
+		emit_data->args[1] = lp_build_emit_fetch(bld_base,
+							 emit_data->inst, 1,
+							 1);
+		emit_data->arg_count = 2;
+	} else if (inst->Instruction.Opcode == TGSI_OPCODE_INTERP_SAMPLE) {
+		LLVMValueRef sample_position;
+		LLVMValueRef sample_id;
+		LLVMValueRef halfval = lp_build_const_float(gallivm, 0.5f);
+		/* fetch sample ID, then fetch it's sample position */
+		sample_id = lp_build_emit_fetch(bld_base,
+						emit_data->inst, 1, 0);
+		sample_id = LLVMBuildBitCast(gallivm->builder, sample_id,
+					     LLVMInt32TypeInContext(gallivm->context),
+					     "");
+		sample_position = load_sample_position(&si_shader_ctx->radeon_bld, sample_id);
+
+		emit_data->args[0] = LLVMBuildExtractElement(gallivm->builder,
+							     sample_position,
+							     lp_build_const_int32(gallivm, 0), "");
+
+		emit_data->args[0] = LLVMBuildFSub(gallivm->builder, emit_data->args[0], halfval, "");
+		emit_data->args[1] = LLVMBuildExtractElement(gallivm->builder,
+							     sample_position,
+							     lp_build_const_int32(gallivm, 1), "");
+		emit_data->args[1] = LLVMBuildFSub(gallivm->builder, emit_data->args[1], halfval, "");
+		emit_data->arg_count = 2;
+	}
+}
+
+static void build_interp_intrinsic(const struct lp_build_tgsi_action * action,
+				struct lp_build_tgsi_context * bld_base,
+				struct lp_build_emit_data * emit_data)
+{
+	struct si_shader_context *si_shader_ctx = si_shader_context(bld_base);
+	struct si_shader *shader = si_shader_ctx->shader;
+	struct gallivm_state *gallivm = bld_base->base.gallivm;
+	LLVMValueRef interp_param;
+	const struct tgsi_full_instruction * inst = emit_data->inst;
+	const char * intr_name;
+	int input_index;
+	int chan;
+	int i;
+	LLVMValueRef attr_number;
+	LLVMTypeRef input_type = LLVMFloatTypeInContext(gallivm->context);
+
+	LLVMValueRef params = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, SI_PARAM_PRIM_MASK);
+	int interp_param_idx;
+	unsigned location;
+
+	assert(inst->Src[0].Register.File == TGSI_FILE_INPUT);
+	input_index = inst->Src[0].Register.Index;
+
+	if (inst->Instruction.Opcode == TGSI_OPCODE_INTERP_OFFSET ||
+	    inst->Instruction.Opcode == TGSI_OPCODE_INTERP_SAMPLE)
+		location = TGSI_INTERPOLATE_LOC_CENTER;
+	else
+		location = TGSI_INTERPOLATE_LOC_CENTROID;
+
+	interp_param_idx = lookup_interp_param_index(shader->ps_input_interpolate[input_index],
+						     location);
+	if (interp_param_idx == -1)
+		return;
+	else if (interp_param_idx)
+		interp_param = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, interp_param_idx);
+	else
+		interp_param = NULL;
+
+	attr_number = lp_build_const_int32(gallivm,
+					   shader->ps_input_param_offset[input_index]);
+
+	if (inst->Instruction.Opcode == TGSI_OPCODE_INTERP_OFFSET ||
+	    inst->Instruction.Opcode == TGSI_OPCODE_INTERP_SAMPLE) {
+		LLVMValueRef ij_out[2];
+
+		LLVMValueRef ddxy_out = si_llvm_emit_ddxy_interp(bld_base, interp_param);
+
+		for (i = 0; i < 2; i++) {
+			LLVMValueRef ix_ll = lp_build_const_int32(gallivm, i);
+			LLVMValueRef iy_ll = lp_build_const_int32(gallivm, i + 2);
+			LLVMValueRef ddx_el = LLVMBuildExtractElement(gallivm->builder,
+								      ddxy_out, ix_ll, "");
+			LLVMValueRef ddy_el = LLVMBuildExtractElement(gallivm->builder,
+								      ddxy_out, iy_ll, "");
+			LLVMValueRef interp_el = LLVMBuildExtractElement(gallivm->builder,
+									 interp_param, ix_ll, "");
+			LLVMValueRef temp1, temp2;
+
+			interp_el = LLVMBuildBitCast(gallivm->builder, interp_el,
+						     LLVMFloatTypeInContext(gallivm->context), "");
+
+			temp1 = LLVMBuildFMul(gallivm->builder, ddx_el, emit_data->args[0], "");
+
+			temp1 = LLVMBuildFAdd(gallivm->builder, temp1, interp_el, "");
+
+			temp2 = LLVMBuildFMul(gallivm->builder, ddy_el, emit_data->args[1], "");
+
+			temp2 = LLVMBuildFAdd(gallivm->builder, temp2, temp1, "");
+
+			ij_out[i] = LLVMBuildBitCast(gallivm->builder,
+						     temp2,
+						     LLVMIntTypeInContext(gallivm->context, 32), "");
+		}
+		interp_param = lp_build_gather_values(bld_base->base.gallivm, ij_out, 2);
+	}
+
+	intr_name = interp_param ? "llvm.SI.fs.interp" : "llvm.SI.fs.constant";
+	for (chan = 0; chan < 2; chan++) {
+		LLVMValueRef args[4];
+		LLVMValueRef llvm_chan;
+		unsigned schan;
+
+		schan = tgsi_util_get_full_src_register_swizzle(&inst->Src[0], chan);
+		llvm_chan = lp_build_const_int32(gallivm, schan);
+
+		args[0] = llvm_chan;
+		args[1] = attr_number;
+		args[2] = params;
+		args[3] = interp_param;
+
+		emit_data->output[chan] =
+			build_intrinsic(gallivm->builder, intr_name,
+					input_type, args, args[3] ? 4 : 3,
+					LLVMReadNoneAttribute | LLVMNoUnwindAttribute);
+	}
+}
+
 /* Emit one vertex from the geometry shader */
 static void si_llvm_emit_vertex(
 	const struct lp_build_tgsi_action *action,
@@ -2365,6 +2584,11 @@ static const struct lp_build_tgsi_action txq_action = {
 	.intr_name = "llvm.SI.resinfo"
 };
 
+static const struct lp_build_tgsi_action interp_action = {
+	.fetch_args = interp_fetch_args,
+	.emit = build_interp_intrinsic,
+};
+
 static void create_meta_data(struct si_shader_context *si_shader_ctx)
 {
 	struct gallivm_state *gallivm = si_shader_ctx->radeon_bld.soa.bld_base.base.gallivm;
@@ -2516,7 +2740,9 @@ static void create_function(struct si_shader_context *si_shader_ctx)
 	    (bld_base->info->opcode_count[TGSI_OPCODE_DDX] > 0 ||
 	     bld_base->info->opcode_count[TGSI_OPCODE_DDY] > 0 ||
 	     bld_base->info->opcode_count[TGSI_OPCODE_DDX_FINE] > 0 ||
-	     bld_base->info->opcode_count[TGSI_OPCODE_DDY_FINE] > 0))
+	     bld_base->info->opcode_count[TGSI_OPCODE_DDY_FINE] > 0 ||
+	     bld_base->info->opcode_count[TGSI_OPCODE_INTERP_OFFSET] > 0 ||
+	     bld_base->info->opcode_count[TGSI_OPCODE_INTERP_SAMPLE] > 0))
 		si_shader_ctx->ddxy_lds =
 			LLVMAddGlobalInAddressSpace(gallivm->module,
 						    LLVMArrayType(i32, 64),
@@ -2954,6 +3180,10 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 	bld_base->info = poly_stipple ? &stipple_shader_info : &sel->info;
 	bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = fetch_constant;
 
+	bld_base->op_actions[TGSI_OPCODE_INTERP_CENTROID] = interp_action;
+	bld_base->op_actions[TGSI_OPCODE_INTERP_SAMPLE] = interp_action;
+	bld_base->op_actions[TGSI_OPCODE_INTERP_OFFSET] = interp_action;
+
 	bld_base->op_actions[TGSI_OPCODE_TEX] = tex_action;
 	bld_base->op_actions[TGSI_OPCODE_TEX2] = tex_action;
 	bld_base->op_actions[TGSI_OPCODE_TXB] = tex_action;
-- 
2.4.3



More information about the mesa-dev mailing list