radeonsi: fix UINT/SINT clamping for 10-bit formats on <= CIK

The same PS epilog workaround as for 8-bit integer formats is required,
since the CB doesn't do clamping.

Fixes GL45-CTS.gtf32.GL3Tests.packed_pixels.packed_pixels*.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

src/gallium/drivers/radeon/r600_pipe_common.h

@@ -316,6 +316,7 @@ struct r600_surface {
 	bool alphatest_bypass;
 	bool export_16bpc;
 	bool color_is_int8;
+	bool color_is_int10;
 
 	/* Color registers. */
 	unsigned cb_color_info;

src/gallium/drivers/radeonsi/si_pipe.h

@@ -169,7 +169,8 @@ struct si_framebuffer {
 	unsigned			spi_shader_col_format_alpha;
 	unsigned			spi_shader_col_format_blend;
 	unsigned			spi_shader_col_format_blend_alpha;
-	unsigned			color_is_int8; /* bitmask */
+	unsigned			color_is_int8;
+	unsigned			color_is_int10;
 	unsigned			dirty_cbufs;
 	bool				dirty_zsbuf;
 	bool				any_dst_linear;

src/gallium/drivers/radeonsi/si_shader.c

@@ -1735,7 +1735,7 @@ static void si_llvm_init_export_args(struct lp_build_tgsi_context *bld_base,
 	LLVMValueRef val[4];
 	unsigned spi_shader_col_format = V_028714_SPI_SHADER_32_ABGR;
 	unsigned chan;
-	bool is_int8;
+	bool is_int8, is_int10;
 
 	/* Default is 0xf. Adjusted below depending on the format. */
 	args[0] = lp_build_const_int32(base->gallivm, 0xf); /* writemask */
@@ -1757,6 +1757,7 @@ static void si_llvm_init_export_args(struct lp_build_tgsi_context *bld_base,
 		assert(cbuf >= 0 && cbuf < 8);
 		spi_shader_col_format = (col_formats >> (cbuf * 4)) & 0xf;
 		is_int8 = (key->part.ps.epilog.color_is_int8 >> cbuf) & 0x1;
+		is_int10 = (key->part.ps.epilog.color_is_int10 >> cbuf) & 0x1;
 	}
 
 	args[4] = uint->zero; /* COMPR flag */
@@ -1856,13 +1857,17 @@ static void si_llvm_init_export_args(struct lp_build_tgsi_context *bld_base,
 		break;
 
 	case V_028714_SPI_SHADER_UINT16_ABGR: {
-		LLVMValueRef max = lp_build_const_int32(gallivm, is_int8 ?
+		LLVMValueRef max_rgb = lp_build_const_int32(gallivm,
-							255 : 65535);
+			is_int8 ? 255 : is_int10 ? 1023 : 65535);
+		LLVMValueRef max_alpha =
+			!is_int10 ? max_rgb : lp_build_const_int32(gallivm, 3);
+
 		/* Clamp. */
 		for (chan = 0; chan < 4; chan++) {
 			val[chan] = bitcast(bld_base, TGSI_TYPE_UNSIGNED, values[chan]);
 			val[chan] = lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_UMIN,
-							      val[chan], max);
+					val[chan],
+					chan == 3 ? max_alpha : max_rgb);
 		}
 
 		args[4] = uint->one; /* COMPR flag */
@@ -1874,19 +1879,24 @@ static void si_llvm_init_export_args(struct lp_build_tgsi_context *bld_base,
 	}
 
 	case V_028714_SPI_SHADER_SINT16_ABGR: {
-		LLVMValueRef max = lp_build_const_int32(gallivm, is_int8 ?
+		LLVMValueRef max_rgb = lp_build_const_int32(gallivm,
-							127 : 32767);
+			is_int8 ? 127 : is_int10 ? 511 : 32767);
-		LLVMValueRef min = lp_build_const_int32(gallivm, is_int8 ?
+		LLVMValueRef min_rgb = lp_build_const_int32(gallivm,
-							-128 : -32768);
+			is_int8 ? -128 : is_int10 ? -512 : -32768);
+		LLVMValueRef max_alpha =
+			!is_int10 ? max_rgb : lp_build_const_int32(gallivm, 1);
+		LLVMValueRef min_alpha =
+			!is_int10 ? min_rgb : lp_build_const_int32(gallivm, -2);
+
 		/* Clamp. */
 		for (chan = 0; chan < 4; chan++) {
 			val[chan] = bitcast(bld_base, TGSI_TYPE_UNSIGNED, values[chan]);
 			val[chan] = lp_build_emit_llvm_binary(bld_base,
-							      TGSI_OPCODE_IMIN,
+					TGSI_OPCODE_IMIN,
-							      val[chan], max);
+					val[chan], chan == 3 ? max_alpha : max_rgb);
 			val[chan] = lp_build_emit_llvm_binary(bld_base,
-							      TGSI_OPCODE_IMAX,
+					TGSI_OPCODE_IMAX,
-							      val[chan], min);
+					val[chan], chan == 3 ? min_alpha : min_rgb);
 		}
 
 		args[4] = uint->one; /* COMPR flag */
@@ -6336,6 +6346,7 @@ static void si_dump_shader_key(unsigned shader, struct si_shader_key *key,
 		fprintf(f, "  part.ps.prolog.bc_optimize_for_linear = %u\n", key->part.ps.prolog.bc_optimize_for_linear);
 		fprintf(f, "  part.ps.epilog.spi_shader_col_format = 0x%x\n", key->part.ps.epilog.spi_shader_col_format);
 		fprintf(f, "  part.ps.epilog.color_is_int8 = 0x%X\n", key->part.ps.epilog.color_is_int8);
+		fprintf(f, "  part.ps.epilog.color_is_int10 = 0x%X\n", key->part.ps.epilog.color_is_int10);
 		fprintf(f, "  part.ps.epilog.last_cbuf = %u\n", key->part.ps.epilog.last_cbuf);
 		fprintf(f, "  part.ps.epilog.alpha_func = %u\n", key->part.ps.epilog.alpha_func);
 		fprintf(f, "  part.ps.epilog.alpha_to_one = %u\n", key->part.ps.epilog.alpha_to_one);

src/gallium/drivers/radeonsi/si_shader.h

@@ -375,6 +375,7 @@ struct si_ps_prolog_bits {
 struct si_ps_epilog_bits {
 	unsigned	spi_shader_col_format;
 	unsigned	color_is_int8:8;
+	unsigned	color_is_int10:8;
 	unsigned	last_cbuf:3;
 	unsigned	alpha_func:3;
 	unsigned	alpha_to_one:1;

src/gallium/drivers/radeonsi/si_state.c

@@ -2137,11 +2137,15 @@ static void si_initialize_color_surface(struct si_context *sctx,
 		blend_bypass = 1;
 	}
 
-	if ((ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT) &&
+	if (ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT) {
-	    (format == V_028C70_COLOR_8 ||
+		if (format == V_028C70_COLOR_8 ||
-	     format == V_028C70_COLOR_8_8 ||
+		    format == V_028C70_COLOR_8_8 ||
-	     format == V_028C70_COLOR_8_8_8_8))
+		    format == V_028C70_COLOR_8_8_8_8)
-		surf->color_is_int8 = true;
+			surf->color_is_int8 = true;
+		else if (format == V_028C70_COLOR_10_10_10_2 ||
+			 format == V_028C70_COLOR_2_10_10_10)
+			surf->color_is_int10 = true;
+	}
 
 	color_info = S_028C70_FORMAT(format) |
 		S_028C70_COMP_SWAP(swap) |
@@ -2405,6 +2409,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
 	sctx->framebuffer.spi_shader_col_format_blend = 0;
 	sctx->framebuffer.spi_shader_col_format_blend_alpha = 0;
 	sctx->framebuffer.color_is_int8 = 0;
+	sctx->framebuffer.color_is_int10 = 0;
 
 	sctx->framebuffer.compressed_cb_mask = 0;
 	sctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
@@ -2434,6 +2439,8 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
 
 		if (surf->color_is_int8)
 			sctx->framebuffer.color_is_int8 |= 1 << i;
+		if (surf->color_is_int10)
+			sctx->framebuffer.color_is_int10 |= 1 << i;
 
 		if (rtex->fmask.size) {
 			sctx->framebuffer.compressed_cb_mask |= 1 << i;

src/gallium/drivers/radeonsi/si_state_shaders.c

@@ -1051,13 +1051,16 @@ static inline void si_shader_selector_key(struct pipe_context *ctx,
 		 * to the range supported by the type if a channel has less
 		 * than 16 bits and the export format is 16_ABGR.
 		 */
-		if (sctx->b.chip_class <= CIK && sctx->b.family != CHIP_HAWAII)
+		if (sctx->b.chip_class <= CIK && sctx->b.family != CHIP_HAWAII) {
 			key->part.ps.epilog.color_is_int8 = sctx->framebuffer.color_is_int8;
+			key->part.ps.epilog.color_is_int10 = sctx->framebuffer.color_is_int10;
+		}
 
 		/* Disable unwritten outputs (if WRITE_ALL_CBUFS isn't enabled). */
 		if (!key->part.ps.epilog.last_cbuf) {
 			key->part.ps.epilog.spi_shader_col_format &= sel->colors_written_4bit;
 			key->part.ps.epilog.color_is_int8 &= sel->info.colors_written;
+			key->part.ps.epilog.color_is_int10 &= sel->info.colors_written;
 		}
 
 		if (rs) {