radeonsi: apply a mask to gl_SampleMaskIn in the PS prolog

gl_SampleMaskIn is supposed to contain set bits only for the samples that
are covered by the current fragment shader invocation, but the VGPR
initialization hardware loads the set of all bits that are covered at the
current pixel.

Fixes various tests in
dEQP-GLES31.functional.shaders.sample_variables.sample_mask_in.*

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
(cherry picked from commit 92c4277990799641d4986ce66a62366228713945)
[Juan A. Suarez: resolve trivial conflicts]
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>

Conflicts:
	src/gallium/drivers/radeonsi/si_shader.c

3 files changed, 76 insertions, 5 deletions

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index bba97cce60c..f77a5087b94 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -85,10 +85,10 @@ static void si_build_ps_prolog_function(struct si_shader_context *ctx,
 static void si_build_ps_epilog_function(struct si_shader_context *ctx,
 					union si_shader_part_key *key);
 
-/* Ideally pass the sample mask input to the PS epilog as v13, which
+/* Ideally pass the sample mask input to the PS epilog as v14, which
  * is its usual location, so that the shader doesn't have to add v_mov.
  */
-#define PS_EPILOG_SAMPLEMASK_MIN_LOC 13
+#define PS_EPILOG_SAMPLEMASK_MIN_LOC 14
 
 /* The VS location of the PrimitiveID input is the same in the epilog,
  * so that the main shader part doesn't have to move it.
@@ -5809,6 +5809,7 @@ static void create_function(struct si_shader_context *ctx)
 		params[SI_PARAM_FRONT_FACE] = ctx->i32;
 		shader->info.face_vgpr_index = 20;
 		params[SI_PARAM_ANCILLARY] = ctx->i32;
+		shader->info.ancillary_vgpr_index = 21;
 		params[SI_PARAM_SAMPLE_COVERAGE] = ctx->f32;
 		params[SI_PARAM_POS_FIXED_PT] = ctx->i32;
 		num_params = SI_PARAM_POS_FIXED_PT+1;
@@ -5876,6 +5877,7 @@ static void create_function(struct si_shader_context *ctx)
 				      S_0286D0_LINEAR_CENTER_ENA(1) |
 				      S_0286D0_LINEAR_CENTROID_ENA(1) |
 				      S_0286D0_FRONT_FACE_ENA(1) |
+				      S_0286D0_ANCILLARY_ENA(1) |
 				      S_0286D0_POS_FIXED_PT_ENA(1));
 	} else if (ctx->type == PIPE_SHADER_COMPUTE) {
 		si_llvm_add_attribute(ctx->main_fn,
@@ -7145,6 +7147,7 @@ static void si_get_ps_prolog_key(struct si_shader *shader,
 		 key->ps_prolog.states.force_linear_center_interp ||
 		 key->ps_prolog.states.bc_optimize_for_persp ||
 		 key->ps_prolog.states.bc_optimize_for_linear);
+	key->ps_prolog.ancillary_vgpr_index = shader->info.ancillary_vgpr_index;
 
 	if (info->colors_read) {
 		unsigned *color = shader->selector->color_attr_index;
@@ -7254,7 +7257,8 @@ static bool si_need_ps_prolog(const union si_shader_part_key *key)
 	       key->ps_prolog.states.force_linear_center_interp ||
 	       key->ps_prolog.states.bc_optimize_for_persp ||
 	       key->ps_prolog.states.bc_optimize_for_linear ||
-	       key->ps_prolog.states.poly_stipple;
+	       key->ps_prolog.states.poly_stipple ||
+	       key->ps_prolog.states.samplemask_log_ps_iter;
 }
 
 /**
@@ -7720,6 +7724,7 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 	if (ctx.type == PIPE_SHADER_FRAGMENT) {
 		shader->info.num_input_vgprs = 0;
 		shader->info.face_vgpr_index = -1;
+		shader->info.ancillary_vgpr_index = -1;
 
 		if (G_0286CC_PERSP_SAMPLE_ENA(shader->config.spi_ps_input_addr))
 			shader->info.num_input_vgprs += 2;
@@ -7749,8 +7754,10 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 			shader->info.face_vgpr_index = shader->info.num_input_vgprs;
 			shader->info.num_input_vgprs += 1;
 		}
-		if (G_0286CC_ANCILLARY_ENA(shader->config.spi_ps_input_addr))
+		if (G_0286CC_ANCILLARY_ENA(shader->config.spi_ps_input_addr)) {
+			shader->info.ancillary_vgpr_index = shader->info.num_input_vgprs;
 			shader->info.num_input_vgprs += 1;
+		}
 		if (G_0286CC_SAMPLE_COVERAGE_ENA(shader->config.spi_ps_input_addr))
 			shader->info.num_input_vgprs += 1;
 		if (G_0286CC_POS_FIXED_PT_ENA(shader->config.spi_ps_input_addr))
@@ -8395,6 +8402,54 @@ static void si_build_ps_prolog_function(struct si_shader_context *ctx,
 		}
 	}
 
+	/* Section 15.2.2 (Shader Inputs) of the OpenGL 4.5 (Core Profile) spec
+	 * says:
+	 *
+	 *    "When per-sample shading is active due to the use of a fragment
+	 *     input qualified by sample or due to the use of the gl_SampleID
+	 *     or gl_SamplePosition variables, only the bit for the current
+	 *     sample is set in gl_SampleMaskIn. When state specifies multiple
+	 *     fragment shader invocations for a given fragment, the sample
+	 *     mask for any single fragment shader invocation may specify a
+	 *     subset of the covered samples for the fragment. In this case,
+	 *     the bit corresponding to each covered sample will be set in
+	 *     exactly one fragment shader invocation."
+	 *
+	 * The samplemask loaded by hardware is always the coverage of the
+	 * entire pixel/fragment, so mask bits out based on the sample ID.
+	 */
+	if (key->ps_prolog.states.samplemask_log_ps_iter) {
+		/* The bit pattern matches that used by fixed function fragment
+		 * processing. */
+		static const uint16_t ps_iter_masks[] = {
+			0xffff, /* not used */
+			0x5555,
+			0x1111,
+			0x0101,
+			0x0001,
+		};
+		assert(key->ps_prolog.states.samplemask_log_ps_iter < ARRAY_SIZE(ps_iter_masks));
+
+		uint32_t ps_iter_mask = ps_iter_masks[key->ps_prolog.states.samplemask_log_ps_iter];
+		unsigned ancillary_vgpr = key->ps_prolog.num_input_sgprs +
+					  key->ps_prolog.ancillary_vgpr_index;
+		LLVMValueRef sampleid = unpack_param(ctx, ancillary_vgpr, 8, 4);
+		LLVMValueRef samplemask = LLVMGetParam(func, ancillary_vgpr + 1);
+
+		samplemask = LLVMBuildBitCast(gallivm->builder, samplemask, ctx->i32, "");
+		samplemask = LLVMBuildAnd(
+			gallivm->builder,
+			samplemask,
+			LLVMBuildShl(gallivm->builder,
+				     LLVMConstInt(ctx->i32, ps_iter_mask, false),
+				     sampleid, ""),
+			"");
+		samplemask = LLVMBuildBitCast(gallivm->builder, samplemask, ctx->f32, "");
+
+		ret = LLVMBuildInsertValue(gallivm->builder, ret, samplemask,
+					   ancillary_vgpr + 1, "");
+	}
+
 	/* Tell LLVM to insert WQM instruction sequence when needed. */
 	if (key->ps_prolog.wqm) {
 		LLVMAddTargetDependentFunctionAttr(func,
@@ -8592,6 +8647,12 @@ static bool si_shader_select_ps_parts(struct si_screen *sscreen,
 		assert(G_0286CC_LINEAR_CENTER_ENA(shader->config.spi_ps_input_addr));
 	}
 
+	/* Samplemask fixup requires the sample ID. */
+	if (shader->key.part.ps.prolog.samplemask_log_ps_iter) {
+		shader->config.spi_ps_input_ena |= S_0286CC_ANCILLARY_ENA(1);
+		assert(G_0286CC_ANCILLARY_ENA(shader->config.spi_ps_input_addr));
+	}
+
 	/* The sample mask input is always enabled, because the API shader always
 	 * passes it through to the epilog. Disable it here if it's unused.
 	 */
@@ -8669,6 +8730,7 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 		shader->info.num_input_sgprs = mainp->info.num_input_sgprs;
 		shader->info.num_input_vgprs = mainp->info.num_input_vgprs;
 		shader->info.face_vgpr_index = mainp->info.face_vgpr_index;
+		shader->info.ancillary_vgpr_index = mainp->info.ancillary_vgpr_index;
 		memcpy(shader->info.vs_output_param_offset,
 		       mainp->info.vs_output_param_offset,
 		       sizeof(mainp->info.vs_output_param_offset));
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 2aac4242d77..a2cf98272d4 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -371,6 +371,7 @@ struct si_ps_prolog_bits {
 	unsigned	force_linear_center_interp:1;
 	unsigned	bc_optimize_for_persp:1;
 	unsigned	bc_optimize_for_linear:1;
+	unsigned	samplemask_log_ps_iter:3;
 };
 
 /* Common PS bits between the shader key and the epilog key. */
@@ -409,6 +410,7 @@ union si_shader_part_key {
 		unsigned	colors_read:8; /* color input components read */
 		unsigned	num_interp_inputs:5; /* BCOLOR is at this location */
 		unsigned	face_vgpr_index:5;
+		unsigned	ancillary_vgpr_index:5;
 		unsigned	wqm:1;
 		char		color_attr_index[2];
 		char		color_interp_vgpr_index[2]; /* -1 == constant */
@@ -503,7 +505,8 @@ struct si_shader_info {
 	ubyte			vs_output_param_offset[SI_MAX_VS_OUTPUTS];
 	ubyte			num_input_sgprs;
 	ubyte			num_input_vgprs;
-	char			face_vgpr_index;
+	signed char		face_vgpr_index;
+	signed char		ancillary_vgpr_index;
 	bool			uses_instanceid;
 	ubyte			nr_pos_exports;
 	ubyte			nr_param_exports;
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index f2413907047..8b1827b3440 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1154,6 +1154,12 @@ static inline void si_shader_selector_key(struct pipe_context *ctx,
 							     sctx->framebuffer.nr_samples <= 1;
 			key->part.ps.epilog.clamp_color = rs->clamp_fragment_color;
 
+			if (sctx->ps_iter_samples > 1 &&
+			    sel->info.reads_samplemask) {
+				key->part.ps.prolog.samplemask_log_ps_iter =
+					util_logbase2(util_next_power_of_two(sctx->ps_iter_samples));
+			}
+
 			if (rs->force_persample_interp &&
 			    rs->multisample_enable &&
 			    sctx->framebuffer.nr_samples > 1 &&