freedreno: Introduce a "cpp_shift" value for cpp divs/muls.

This only converts part of the driver to use it, leaving the rest to the
following commit (which inspired this one).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4558>

src/freedreno/fdl/fd6_layout.c

@@ -93,6 +93,8 @@ fdl6_layout(struct fdl_layout *layout,
 
 	layout->cpp = util_format_get_blocksize(format);
 	layout->cpp *= nr_samples;
+	layout->cpp_shift = ffs(layout->cpp) - 1;
+
 	layout->format = format;
 	layout->nr_samples = nr_samples;
 	layout->layer_first = !is_3d;

src/freedreno/fdl/freedreno_layout.c

@@ -36,6 +36,7 @@ fdl_layout_buffer(struct fdl_layout *layout, uint32_t size)
 	layout->height0 = 1;
 	layout->depth0 = 1;
 	layout->cpp = 1;
+	layout->cpp_shift = 0;
 	layout->size = size;
 	layout->format = PIPE_FORMAT_R8_UINT;
 	layout->nr_samples = 1;

src/freedreno/fdl/freedreno_layout.h

@@ -109,6 +109,12 @@ struct fdl_layout {
 	 */
 	uint8_t cpp;
 
+	/**
+	 * Left shift necessary to multiply by cpp.  Invalid for NPOT cpp, please
+	 * use fdl_cpp_shift() to sanity check you aren't hitting that case.
+	 */
+	uint8_t cpp_shift;
+
 	uint32_t width0, height0, depth0;
 	uint32_t nr_samples;
 	enum pipe_format format;
@@ -117,6 +123,13 @@ struct fdl_layout {
 	uint32_t base_align; /* Alignment of the base address, in bytes. */
 };
 
+static inline uint32_t
+fdl_cpp_shift(const struct fdl_layout *layout)
+{
+	assert(util_is_power_of_two_or_zero(layout->cpp));
+	return layout->cpp_shift;
+}
+
 static inline uint32_t
 fdl_layer_stride(const struct fdl_layout *layout, unsigned level)
 {

src/gallium/drivers/freedreno/a3xx/fd3_gmem.c

@@ -94,7 +94,7 @@ emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
 			swap = rsc->layout.tile_mode ? WZYX : fd3_pipe2swap(pformat);
 
 			if (bin_w) {
-				stride = bin_w * rsc->layout.cpp;
+				stride = bin_w << fdl_cpp_shift(&rsc->layout);
 
 				if (bases) {
 					base = bases[i];
@@ -1009,11 +1009,13 @@ fd3_emit_tile_renderprep(struct fd_batch *batch, const struct fd_tile *tile)
 	OUT_RING(ring, reg);
 	if (pfb->zsbuf) {
 		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
-		OUT_RING(ring, A3XX_RB_DEPTH_PITCH(rsc->layout.cpp * gmem->bin_w));
+		OUT_RING(ring, A3XX_RB_DEPTH_PITCH(gmem->bin_w <<
+						fdl_cpp_shift(&rsc->layout)));
 		if (rsc->stencil) {
 			OUT_PKT0(ring, REG_A3XX_RB_STENCIL_INFO, 2);
 			OUT_RING(ring, A3XX_RB_STENCIL_INFO_STENCIL_BASE(gmem->zsbuf_base[1]));
-			OUT_RING(ring, A3XX_RB_STENCIL_PITCH(rsc->stencil->layout.cpp * gmem->bin_w));
+			OUT_RING(ring, A3XX_RB_STENCIL_PITCH(gmem->bin_w <<
+							fdl_cpp_shift(&rsc->stencil->layout)));
 		}
 	} else {
 		OUT_RING(ring, 0x00000000);

src/gallium/drivers/freedreno/a4xx/fd4_gmem.c

@@ -97,7 +97,7 @@ emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
 					psurf->u.tex.first_layer);
 
 			if (bin_w) {
-				stride = bin_w * rsc->layout.cpp;
+				stride = bin_w << fdl_cpp_shift(&rsc->layout);
 
 				if (bases) {
 					base = bases[i];

src/gallium/drivers/freedreno/a5xx/fd5_gmem.c

@@ -498,7 +498,7 @@ emit_mem2gmem_surf(struct fd_batch *batch, uint32_t base,
 		buf = BLIT_MRT0;
 	}
 
-	stride = gmem->bin_w * rsc->layout.cpp;
+	stride = gmem->bin_w << fdl_cpp_shift(&rsc->layout);
 	size = stride * gmem->bin_h;
 
 	OUT_PKT4(ring, REG_A5XX_RB_BLIT_FLAG_DST_LO, 4);

src/gallium/drivers/freedreno/freedreno_resource.c

@@ -886,6 +886,7 @@ fd_resource_layout_init(struct pipe_resource *prsc)
 
 	layout->cpp = util_format_get_blocksize(prsc->format);
 	layout->cpp *= fd_resource_nr_samples(prsc);
+	layout->cpp_shift = ffs(layout->cpp) - 1;
 }
 
 /**