radeonsi: add fine derivate control (v2.1)

This adds support for fine derivatives and enables
ARB_derivative_control on radeonsi.

(just fell out of my working out interpolation)

v2: cleanup some bits, write a comment
v2.1: take Michel's comment from the mailing list

Reviewed-by: Marek Olšák <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

4 files changed, 50 insertions, 7 deletions

diff --git a/docs/GL3.txt b/docs/GL3.txt
index e3fa1a1ec1b..15bb57f441a 100644
--- a/docs/GL3.txt
+++ b/docs/GL3.txt
@@ -191,7 +191,7 @@ GL 4.5, GLSL 4.50:
   GL_ARB_clip_control                                  DONE (i965, nv50, nvc0, r600, radeonsi, llvmpipe, softpipe)
   GL_ARB_conditional_render_inverted                   DONE (i965, nv50, nvc0, llvmpipe, softpipe)
   GL_ARB_cull_distance                                 in progress (Tobias)
-  GL_ARB_derivative_control                            DONE (i965, nv50, nvc0, r600)
+  GL_ARB_derivative_control                            DONE (i965, nv50, nvc0, r600, radeonsi)
   GL_ARB_direct_state_access                           DONE (all drivers)
   GL_ARB_get_texture_sub_image                         DONE (all drivers)
   GL_ARB_shader_texture_image_samples                  not started
diff --git a/docs/relnotes/10.7.0.html b/docs/relnotes/10.7.0.html
index 26615a83814..afef5256b75 100644
--- a/docs/relnotes/10.7.0.html
+++ b/docs/relnotes/10.7.0.html
@@ -45,6 +45,7 @@ Note: some of the new features are only available with certain drivers.
 
 <ul>
 <li>GL_AMD_vertex_shader_viewport_index on radeonsi</li>
+<li>GL_ARB_derivative_control on radeonsi</li>
 <li>GL_ARB_fragment_layer_viewport on radeonsi</li>
 <li>GL_ARB_framebuffer_no_attachments on i965</li>
 <li>GL_ARB_get_texture_sub_image for all drivers</li>
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index c2985b8d224..ebe1f5a68f1 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -249,6 +249,7 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 	case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
 	case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
 	case PIPE_CAP_TGSI_TEXCOORD:
+	case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 		return 1;
 
 	case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
@@ -289,7 +290,6 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 	case PIPE_CAP_USER_VERTEX_BUFFERS:
 	case PIPE_CAP_FAKE_SW_MSAA:
 	case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
-	case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
 	case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
 	case PIPE_CAP_SAMPLER_VIEW_TARGET:
 	case PIPE_CAP_VERTEXID_NOBASE:
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index bc0f2171a52..27b3c72d1c9 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2854,6 +2854,35 @@ static void build_txq_intrinsic(const struct lp_build_tgsi_action * action,
 	}
 }
 
+/*
+ * SI implements derivatives using the local data store (LDS)
+ * All writes to the LDS happen in all executing threads at
+ * the same time. TID is the Thread ID for the current
+ * thread and is a value between 0 and 63, representing
+ * the thread's position in the wavefront.
+ *
+ * For the pixel shader threads are grouped into quads of four pixels.
+ * The TIDs of the pixels of a quad are:
+ *
+ *  +------+------+
+ *  |4n + 0|4n + 1|
+ *  +------+------+
+ *  |4n + 2|4n + 3|
+ *  +------+------+
+ *
+ * So, masking the TID with 0xfffffffc yields the TID of the top left pixel
+ * of the quad, masking with 0xfffffffd yields the TID of the top pixel of
+ * the current pixel's column, and masking with 0xfffffffe yields the TID
+ * of the left pixel of the current pixel's row.
+ *
+ * Adding 1 yields the TID of the pixel to the right of the left pixel, and
+ * adding 2 yields the TID of the pixel below the top pixel.
+ */
+/* masks for thread ID. */
+#define TID_MASK_TOP_LEFT 0xfffffffc
+#define TID_MASK_TOP      0xfffffffd
+#define TID_MASK_LEFT     0xfffffffe
+
 static void si_llvm_emit_ddxy(
 	const struct lp_build_tgsi_action * action,
 	struct lp_build_tgsi_context * bld_base,
@@ -2870,6 +2899,8 @@ static void si_llvm_emit_ddxy(
 	LLVMTypeRef i32;
 	unsigned swizzle[4];
 	unsigned c;
+	int idx;
+	unsigned mask;
 
 	i32 = LLVMInt32TypeInContext(gallivm->context);
 
@@ -2879,15 +2910,22 @@ static void si_llvm_emit_ddxy(
 	store_ptr = LLVMBuildGEP(gallivm->builder, si_shader_ctx->lds,
 				 indices, 2, "");
 
+	if (opcode == TGSI_OPCODE_DDX_FINE)
+		mask = TID_MASK_LEFT;
+	else if (opcode == TGSI_OPCODE_DDY_FINE)
+		mask = TID_MASK_TOP;
+	else
+		mask = TID_MASK_TOP_LEFT;
+
 	indices[1] = LLVMBuildAnd(gallivm->builder, indices[1],
-				  lp_build_const_int32(gallivm, 0xfffffffc), "");
+				  lp_build_const_int32(gallivm, mask), "");
 	load_ptr0 = LLVMBuildGEP(gallivm->builder, si_shader_ctx->lds,
 				 indices, 2, "");
 
+	/* for DDX we want to next X pixel, DDY next Y pixel. */
+	idx = (opcode == TGSI_OPCODE_DDX || opcode == TGSI_OPCODE_DDX_FINE) ? 1 : 2;
 	indices[1] = LLVMBuildAdd(gallivm->builder, indices[1],
-				  lp_build_const_int32(gallivm,
-						       opcode == TGSI_OPCODE_DDX ? 1 : 2),
-				  "");
+				  lp_build_const_int32(gallivm, idx), "");
 	load_ptr1 = LLVMBuildGEP(gallivm->builder, si_shader_ctx->lds,
 				 indices, 2, "");
 
@@ -3229,7 +3267,9 @@ static void create_function(struct si_shader_context *si_shader_ctx)
 
 	if (bld_base->info &&
 	    (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_DDY] > 0 ||
+	     bld_base->info->opcode_count[TGSI_OPCODE_DDX_FINE] > 0 ||
+	     bld_base->info->opcode_count[TGSI_OPCODE_DDY_FINE] > 0))
 		si_shader_ctx->lds =
 			LLVMAddGlobalInAddressSpace(gallivm->module,
 						    LLVMArrayType(i32, 64),
@@ -3722,6 +3762,8 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 
 	bld_base->op_actions[TGSI_OPCODE_DDX].emit = si_llvm_emit_ddxy;
 	bld_base->op_actions[TGSI_OPCODE_DDY].emit = si_llvm_emit_ddxy;
+	bld_base->op_actions[TGSI_OPCODE_DDX_FINE].emit = si_llvm_emit_ddxy;
+	bld_base->op_actions[TGSI_OPCODE_DDY_FINE].emit = si_llvm_emit_ddxy;
 
 	bld_base->op_actions[TGSI_OPCODE_EMIT].emit = si_llvm_emit_vertex;
 	bld_base->op_actions[TGSI_OPCODE_ENDPRIM].emit = si_llvm_emit_primitive;