Mesa (master): radeonsi: Handle position input parameter for pixel shaders v2

[Tom Stellard]

Module: Mesa
Branch: master
Commit: 0fb1e68a0ba0dc58e0b97f5bb3f7a46d9b3eae29
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=0fb1e68a0ba0dc58e0b97f5bb3f7a46d9b3eae29

Author: Tom Stellard <thomas.stellard at amd.com>
Date:   Thu Sep  6 16:18:11 2012 -0400

radeonsi: Handle position input parameter for pixel shaders v2

v2:
  - Don't increment ninterp or set any of the have_* flags for
    TGSI_SEMANTIC_POSITION

Reviewed-by: Michel Dänzer <michel.daenzer at amd.com>

---

 src/gallium/drivers/radeon/SIInstructions.td   |   20 ++++++++++++++++++++
 src/gallium/drivers/radeon/SIIntrinsics.td     |    2 ++
 src/gallium/drivers/radeonsi/radeonsi_shader.c |   16 +++++++++++++++-
 src/gallium/drivers/radeonsi/si_state_draw.c   |   16 ++++++++++++++++
 4 files changed, 53 insertions(+), 1 deletions(-)

diff --git a/src/gallium/drivers/radeon/SIInstructions.td b/src/gallium/drivers/radeon/SIInstructions.td
index 8f9de52..ffc9537 100644
--- a/src/gallium/drivers/radeon/SIInstructions.td
+++ b/src/gallium/drivers/radeon/SIInstructions.td
@@ -1102,6 +1102,26 @@ def : Pat <
              imm:$attr, SReg_32:$params)
 >;
 
+def : Pat <
+  (int_SI_fs_read_pos 0),
+  (f32 POS_X_FLOAT)
+>;
+
+def : Pat <
+  (int_SI_fs_read_pos 1),
+  (f32 POS_Y_FLOAT)
+>;
+
+def : Pat <
+  (int_SI_fs_read_pos 2),
+  (f32 POS_Z_FLOAT)
+>;
+
+def : Pat <
+  (int_SI_fs_read_pos 3),
+  (f32 POS_W_FLOAT)
+>;
+
 /********** ================== **********/
 /********** Intrinsic Patterns **********/
 /********** ================== **********/
diff --git a/src/gallium/drivers/radeon/SIIntrinsics.td b/src/gallium/drivers/radeon/SIIntrinsics.td
index fbb8dc9..f5c3f76 100644
--- a/src/gallium/drivers/radeon/SIIntrinsics.td
+++ b/src/gallium/drivers/radeon/SIIntrinsics.td
@@ -34,4 +34,6 @@ let TargetPrefix = "SI", isTarget = 1 in {
   def int_SI_fs_interp_persp_center : Interp;
   def int_SI_fs_interp_persp_centroid : Interp;
   def int_SI_fs_interp_constant : Interp;
+
+  def int_SI_fs_read_pos : Intrinsic <[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
 }
diff --git a/src/gallium/drivers/radeonsi/radeonsi_shader.c b/src/gallium/drivers/radeonsi/radeonsi_shader.c
index 0994fac..466d565 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_shader.c
+++ b/src/gallium/drivers/radeonsi/radeonsi_shader.c
@@ -261,6 +261,7 @@ static void declare_input_fs(
 	struct lp_build_context * base =
 				&si_shader_ctx->radeon_bld.soa.bld_base.base;
 	struct gallivm_state * gallivm = base->gallivm;
+	LLVMTypeRef input_type = LLVMFloatTypeInContext(gallivm->context);
 
 	/* This value is:
 	 * [15:0] NewPrimMask (Bit mask for each quad.  It is set it the
@@ -278,6 +279,20 @@ static void declare_input_fs(
 	/* XXX: Is this the input_index? */
 	LLVMValueRef attr_number = lp_build_const_int32(gallivm, input_index);
 
+	if (decl->Semantic.Name == TGSI_SEMANTIC_POSITION) {
+		for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
+			LLVMValueRef args[1];
+			unsigned soa_index =
+				radeon_llvm_reg_index_soa(input_index, chan);
+			args[0] = lp_build_const_int32(gallivm, chan);
+			si_shader_ctx->radeon_bld.inputs[soa_index] =
+				build_intrinsic(base->gallivm->builder,
+					"llvm.SI.fs.read.pos", input_type,
+					args, 1, LLVMReadNoneAttribute);
+		}
+		return;
+	}
+
 	/* XXX: Handle all possible interpolation modes */
 	switch (decl->Interp.Interpolate) {
 	case TGSI_INTERPOLATE_COLOR:
@@ -332,7 +347,6 @@ static void declare_input_fs(
 		LLVMValueRef args[3];
 		LLVMValueRef llvm_chan = lp_build_const_int32(gallivm, chan);
 		unsigned soa_index = radeon_llvm_reg_index_soa(input_index, chan);
-		LLVMTypeRef input_type = LLVMFloatTypeInContext(gallivm->context);
 		args[0] = llvm_chan;
 		args[1] = attr_number;
 		args[2] = params;
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 8b43f5a..c71ad43 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -103,6 +103,7 @@ static void si_pipe_shader_ps(struct pipe_context *ctx, struct si_pipe_shader *s
 	unsigned num_sgprs, num_user_sgprs;
 	int ninterp = 0;
 	boolean have_linear = FALSE, have_centroid = FALSE, have_perspective = FALSE;
+	unsigned fragcoord_interp_mode = 0;
 	unsigned spi_baryc_cntl, spi_ps_input_ena;
 	uint64_t va;
 
@@ -116,6 +117,20 @@ static void si_pipe_shader_ps(struct pipe_context *ctx, struct si_pipe_shader *s
 
 	db_shader_control = S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
 	for (i = 0; i < shader->shader.ninput; i++) {
+		if (shader->shader.input[i].name == TGSI_SEMANTIC_POSITION) {
+			if (shader->shader.input[i].centroid) {
+				/* fragcoord_interp_mode will be written to
+				 * SPI_BARYC_CNTL.POS_FLOAT_LOCATION
+				 * Possible vaules:
+				 * 0 -> Position = pixel center (default)
+				 * 1 -> Position = pixel centroid
+				 * 2 -> Position = iterated sample number XXX:
+				 *                        What does this mean?
+			 	 */
+				fragcoord_interp_mode = 1;
+			}
+			continue;
+		}
 		ninterp++;
 		/* XXX: Flat shading hangs the GPU */
 		if (shader->shader.input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
@@ -166,6 +181,7 @@ static void si_pipe_shader_ps(struct pipe_context *ctx, struct si_pipe_shader *s
 	if (have_linear)
 		spi_baryc_cntl |= have_centroid ?
 			S_0286E0_LINEAR_CENTROID_CNTL(1) : S_0286E0_LINEAR_CENTER_CNTL(1);
+	spi_baryc_cntl |= S_0286E0_POS_FLOAT_LOCATION(fragcoord_interp_mode);
 
 	si_pm4_set_reg(pm4, R_0286E0_SPI_BARYC_CNTL, spi_baryc_cntl);
 	spi_ps_input_ena = shader->spi_ps_input_ena;