freedreno: Add a helper for computing guardband sizes

This should be much better than the previous method that was more
guesswork-based than anything else. It returns a value within 1 of the
blob for every input value I've tested, and it seems like it returns
slightly better (but still legal) answers when it differs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5950>

src/freedreno/common/freedreno_guardband.h

@@ -0,0 +1,98 @@
+/*
+ * Copyright © 2020 Valve Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef __FREEDRENO_GUARDBAND_H__
+#define __FREEDRENO_GUARDBAND_H__
+
+#include <math.h>
+#include <stdbool.h>
+#include <assert.h>
+
+static inline unsigned
+fd_calc_guardband(float offset, float scale, bool is_a3xx)
+{
+	/* On a3xx, the viewport max is 4k and the docs say the max guardband
+	 * width is 8k. That is, GRAS cannot handle triangle coordinates more than
+	 * 8k, positive or negative. On a4xx+ the viewport width was bumped to
+	 * 16k, and so the guardband width was necessarily also bumped. Note that
+	 * the numbers here should correspond to
+	 * VkPhysicalDeviceLimits::viewportBoundsRange in Vulkan.
+	 */
+	const float gb_min = is_a3xx ? -8192. : -32768.;
+	const float gb_max = is_a3xx ?  8191. :  32767.;
+
+	/* Clipping happens in normalized device coordinates, so we have to
+	 * transform gb_min and gb_max to ndc using the inverse of the viewport
+	 * transform. Avoid flipping min and max by using the absolute value of
+	 * the scale.
+	 */
+	const float gb_min_ndc = (gb_min - offset) / fabsf(scale);
+	const float gb_max_ndc = (gb_max - offset) / fabsf(scale);
+
+	/* There's only one GB_ADJ field, so presumably the guardband is
+	 * [-GB_ADJ, GB_ADJ] like on Radeon. It's always safe to make the
+	 * guardband smaller, so we have to take the min to get the largest range
+	 * contained in [gb_min_ndc, gb_max_ndc].
+	 */
+	const float gb_adj = fminf(-gb_min_ndc, gb_max_ndc);
+
+	/* The viewport should always be contained in the guardband. */
+	assert(gb_adj >= 1.0);
+
+	/* frexp returns an unspecified value if given an infinite value, which
+	 * can happen if scale == 0.
+	 */
+	if (isinf(gb_adj))
+		return 0x1ff;
+
+	/* Convert gb_adj to 3.6 floating point, rounding down since it's always
+	 * safe to make the guard band smaller (but not the other way around!).
+	 *
+	 * Note: After converting back to a float, the value the blob returns here
+	 * is sometimes a little smaller than the value we return. This seems to
+	 * happen around the boundary between two different rounded values. For
+	 * example, using the a6xx blob:
+	 *
+	 * min  | width  | unrounded gb_adj | blob result | mesa result
+	 * ------------------------------------------------------------
+	 * 0    | 510    |          127.498 |        127. |        127.
+	 * 0    | 511    |          127.247 |        126. |        127.
+	 * 0    | 512    |          126.996 |        126. |        126.
+	 *
+	 * The guardband must be 32767 wide, since that's what the blob reports
+	 * for viewportBoundsRange, so I'm guessing that they're rounding slightly
+	 * more conservatively somehow.
+	 */
+	int gb_adj_exp;
+	float gb_adj_mantissa = frexpf(gb_adj, &gb_adj_exp);
+	assert(gb_adj_exp > 0);
+
+	/* Round non-representable numbers down to the largest possible number. */
+	if (gb_adj_exp > 8)
+		return 0x1ff;
+
+	return ((gb_adj_exp - 1) << 6) |
+		((unsigned) truncf(gb_adj_mantissa * (1 << 7)) - (1 << 6));
+}
+
+#endif /* __FREEDRENO_GUARDBAND_H__ */

src/freedreno/common/meson.build

@@ -23,6 +23,7 @@ libfreedreno_common = static_library(
   [
     'freedreno_uuid.c',
     'freedreno_uuid.h',
+    'freedreno_guardband.h',
   ],
   include_directories : [inc_freedreno, inc_include, inc_src, inc_gallium],
   c_args : [no_override_init_args],