radv: clean up the sample locations codebase

Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Bas Nieuwenhuizen <[email protected]>

4 files changed, 76 insertions, 98 deletions

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index e4e5966da54..4f592bc7f68 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -669,7 +669,7 @@ radv_update_multisample_state(struct radv_cmd_buffer *cmd_buffer,
 
 	radeon_set_context_reg(cmd_buffer->cs, R_028A48_PA_SC_MODE_CNTL_0, ms->pa_sc_mode_cntl_0);
 
-	radv_cayman_emit_msaa_sample_locs(cmd_buffer->cs, num_samples);
+	radv_emit_default_sample_locations(cmd_buffer->cs, num_samples);
 
 	/* GFX9: Flush DFSM when the AA mode changes. */
 	if (cmd_buffer->device->dfsm_allowed) {
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index c89a6f139ba..56fd65bec29 100644
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -1128,7 +1128,7 @@ radv_pipeline_init_multisample_state(struct radv_pipeline *pipeline,
 			S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples) |
 			S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples);
 		ms->pa_sc_aa_config |= S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
-			S_028BE0_MAX_SAMPLE_DIST(radv_cayman_get_maxdist(log_samples)) |
+			S_028BE0_MAX_SAMPLE_DIST(radv_get_default_max_sample_dist(log_samples)) |
 			S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples); /* CM_R_028BE0_PA_SC_AA_CONFIG */
 		ms->pa_sc_mode_cntl_1 |= S_028A4C_PS_ITER_SAMPLE(ps_iter_samples > 1);
 		if (ps_iter_samples > 1)
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 9563e86a680..7834a505562 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -1212,8 +1212,8 @@ void radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer *cmd_buffer);
 void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer *cmd_buffer);
 void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer *cmd_buffer);
 void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer *cmd_buffer);
-void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf *cs, int nr_samples);
-unsigned radv_cayman_get_maxdist(int log_samples);
+void radv_emit_default_sample_locations(struct radeon_cmdbuf *cs, int nr_samples);
+unsigned radv_get_default_max_sample_dist(int log_samples);
 void radv_device_init_msaa(struct radv_device *device);
 
 void radv_update_ds_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index b93cd88a1d1..56d7a9a3682 100644
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -1300,144 +1300,122 @@ void si_cp_dma_wait_for_idle(struct radv_cmd_buffer *cmd_buffer)
 
 /* For MSAA sample positions. */
 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y)  \
-	(((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) |		   \
-	(((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) |	   \
-	(((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) |	   \
+	((((unsigned)(s0x) & 0xf) << 0)  | (((unsigned)(s0y) & 0xf) << 4)  | \
+	 (((unsigned)(s1x) & 0xf) << 8)  | (((unsigned)(s1y) & 0xf) << 12) | \
+	 (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) | \
 	 (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
 
-
-/* 2xMSAA
- * There are two locations (4, 4), (-4, -4). */
-const uint32_t eg_sample_locs_2x[4] = {
-	FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
-	FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
-	FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
-	FILL_SREG(4, 4, -4, -4, 4, 4, -4, -4),
-};
-const unsigned eg_max_dist_2x = 4;
-/* 4xMSAA
- * There are 4 locations: (-2, 6), (6, -2), (-6, 2), (2, 6). */
-const uint32_t eg_sample_locs_4x[4] = {
-	FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
-	FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
-	FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
-	FILL_SREG(-2, -6, 6, -2, -6, 2, 2, 6),
-};
-const unsigned eg_max_dist_4x = 6;
-
-/* Cayman 8xMSAA */
-static const uint32_t cm_sample_locs_8x[] = {
-	FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
-	FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
-	FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
-	FILL_SREG( 1, -3, -1,  3, 5,  1, -3, -5),
-	FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
-	FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
-	FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
-	FILL_SREG(-5,  5, -7, -1, 3,  7,  7, -7),
+/* For obtaining location coordinates from registers */
+#define SEXT4(x)		((int)((x) | ((x) & 0x8 ? 0xfffffff0 : 0)))
+#define GET_SFIELD(reg, index)	SEXT4(((reg) >> ((index) * 4)) & 0xf)
+#define GET_SX(reg, index)	GET_SFIELD((reg)[(index) / 4], ((index) % 4) * 2)
+#define GET_SY(reg, index)	GET_SFIELD((reg)[(index) / 4], ((index) % 4) * 2 + 1)
+
+/* 1x MSAA */
+static const uint32_t sample_locs_1x =
+	FILL_SREG(0, 0,   0, 0,   0, 0,   0, 0);
+static const unsigned max_dist_1x = 0;
+
+/* 2xMSAA */
+static const uint32_t sample_locs_2x =
+	FILL_SREG(4,4,   -4, -4,   0, 0,   0, 0);
+static const unsigned max_dist_2x = 4;
+
+/* 4xMSAA */
+static const uint32_t sample_locs_4x =
+	FILL_SREG(-2,-6,   6, -2,   -6, 2,  2, 6);
+static const unsigned max_dist_4x = 6;
+
+/* 8xMSAA */
+static const uint32_t sample_locs_8x[] = {
+	FILL_SREG( 1,-3,  -1, 3,   5, 1,  -3,-5),
+	FILL_SREG(-5, 5,  -7,-1,   3, 7,   7,-7),
+	/* The following are unused by hardware, but we emit them to IBs
+	 * instead of multiple SET_CONTEXT_REG packets. */
+	0,
+	0,
 };
-static const unsigned cm_max_dist_8x = 8;
+static const unsigned max_dist_8x = 8;
 
-unsigned radv_cayman_get_maxdist(int log_samples)
+unsigned radv_get_default_max_sample_dist(int log_samples)
 {
 	unsigned max_dist[] = {
-		0,
-		eg_max_dist_2x,
-		eg_max_dist_4x,
-		cm_max_dist_8x,
+		max_dist_1x,
+		max_dist_2x,
+		max_dist_4x,
+		max_dist_8x,
 	};
 	return max_dist[log_samples];
 }
 
-void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf *cs, int nr_samples)
+void radv_emit_default_sample_locations(struct radeon_cmdbuf *cs, int nr_samples)
 {
 	switch (nr_samples) {
 	default:
 	case 1:
-		radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 0);
-		radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, 0);
-		radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, 0);
-		radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, 0);
+		radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_1x);
+		radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_1x);
+		radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_1x);
+		radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_1x);
 		break;
 	case 2:
-		radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, eg_sample_locs_2x[0]);
-		radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, eg_sample_locs_2x[1]);
-		radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, eg_sample_locs_2x[2]);
-		radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, eg_sample_locs_2x[3]);
+		radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_2x);
+		radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_2x);
+		radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_2x);
+		radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_2x);
 		break;
 	case 4:
-		radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, eg_sample_locs_4x[0]);
-		radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, eg_sample_locs_4x[1]);
-		radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, eg_sample_locs_4x[2]);
-		radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, eg_sample_locs_4x[3]);
+		radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_4x);
+		radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_4x);
+		radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_4x);
+		radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_4x);
 		break;
 	case 8:
 		radeon_set_context_reg_seq(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, 14);
-		radeon_emit(cs, cm_sample_locs_8x[0]);
-		radeon_emit(cs, cm_sample_locs_8x[4]);
-		radeon_emit(cs, 0);
-		radeon_emit(cs, 0);
-		radeon_emit(cs, cm_sample_locs_8x[1]);
-		radeon_emit(cs, cm_sample_locs_8x[5]);
-		radeon_emit(cs, 0);
-		radeon_emit(cs, 0);
-		radeon_emit(cs, cm_sample_locs_8x[2]);
-		radeon_emit(cs, cm_sample_locs_8x[6]);
-		radeon_emit(cs, 0);
-		radeon_emit(cs, 0);
-		radeon_emit(cs, cm_sample_locs_8x[3]);
-		radeon_emit(cs, cm_sample_locs_8x[7]);
+		radeon_emit_array(cs, sample_locs_8x, 4);
+		radeon_emit_array(cs, sample_locs_8x, 4);
+		radeon_emit_array(cs, sample_locs_8x, 4);
+		radeon_emit_array(cs, sample_locs_8x, 2);
 		break;
 	}
 }
 
-static void radv_cayman_get_sample_position(struct radv_device *device,
-					    unsigned sample_count,
-					    unsigned sample_index, float *out_value)
+static void radv_get_sample_position(struct radv_device *device,
+				     unsigned sample_count,
+				     unsigned sample_index, float *out_value)
 {
-	int offset, index;
-	struct {
-		int idx:4;
-	} val;
+	const uint32_t *sample_locs;
+
 	switch (sample_count) {
 	case 1:
 	default:
-		out_value[0] = out_value[1] = 0.5;
+		sample_locs = &sample_locs_1x;
 		break;
 	case 2:
-		offset = 4 * (sample_index * 2);
-		val.idx = (eg_sample_locs_2x[0] >> offset) & 0xf;
-		out_value[0] = (float)(val.idx + 8) / 16.0f;
-		val.idx = (eg_sample_locs_2x[0] >> (offset + 4)) & 0xf;
-		out_value[1] = (float)(val.idx + 8) / 16.0f;
+		sample_locs = &sample_locs_2x;
 		break;
 	case 4:
-		offset = 4 * (sample_index * 2);
-		val.idx = (eg_sample_locs_4x[0] >> offset) & 0xf;
-		out_value[0] = (float)(val.idx + 8) / 16.0f;
-		val.idx = (eg_sample_locs_4x[0] >> (offset + 4)) & 0xf;
-		out_value[1] = (float)(val.idx + 8) / 16.0f;
+		sample_locs = &sample_locs_4x;
 		break;
 	case 8:
-		offset = 4 * (sample_index % 4 * 2);
-		index = (sample_index / 4) * 4;
-		val.idx = (cm_sample_locs_8x[index] >> offset) & 0xf;
-		out_value[0] = (float)(val.idx + 8) / 16.0f;
-		val.idx = (cm_sample_locs_8x[index] >> (offset + 4)) & 0xf;
-		out_value[1] = (float)(val.idx + 8) / 16.0f;
+		sample_locs = sample_locs_8x;
 		break;
 	}
+
+	out_value[0] = (GET_SX(sample_locs, sample_index) + 8) / 16.0f;
+	out_value[1] = (GET_SY(sample_locs, sample_index) + 8) / 16.0f;
 }
 
 void radv_device_init_msaa(struct radv_device *device)
 {
 	int i;
-	radv_cayman_get_sample_position(device, 1, 0, device->sample_locations_1x[0]);
+
+	radv_get_sample_position(device, 1, 0, device->sample_locations_1x[0]);
 
 	for (i = 0; i < 2; i++)
-		radv_cayman_get_sample_position(device, 2, i, device->sample_locations_2x[i]);
+		radv_get_sample_position(device, 2, i, device->sample_locations_2x[i]);
 	for (i = 0; i < 4; i++)
-		radv_cayman_get_sample_position(device, 4, i, device->sample_locations_4x[i]);
+		radv_get_sample_position(device, 4, i, device->sample_locations_4x[i]);
 	for (i = 0; i < 8; i++)
-		radv_cayman_get_sample_position(device, 8, i, device->sample_locations_8x[i]);
+		radv_get_sample_position(device, 8, i, device->sample_locations_8x[i]);
 }