summaryrefslogtreecommitdiffstats
path: root/src/broadcom/common
diff options
context:
space:
mode:
authorEric Anholt <[email protected]>2018-12-17 11:10:11 -0800
committerEric Anholt <[email protected]>2018-12-19 10:27:26 -0800
commitf6a0f4f41e9a4d3f0ea9c3bf82497c384e684843 (patch)
tree71e7093327dc7ccbc2d6bb87cd258cb3bb6f572d /src/broadcom/common
parent8ee752194c1bc4d9ad145f6bf206ee119c9498b8 (diff)
vc4: Move the utile load/store functions to a header for reuse by v3d.
These implementations of whole-utile load/stores would be the same for v3d, though the layouts of blocks of utiles has changed.
Diffstat (limited to 'src/broadcom/common')
-rw-r--r--src/broadcom/common/v3d_cpu_tiling.h222
1 files changed, 222 insertions, 0 deletions
diff --git a/src/broadcom/common/v3d_cpu_tiling.h b/src/broadcom/common/v3d_cpu_tiling.h
new file mode 100644
index 00000000000..15678f9e925
--- /dev/null
+++ b/src/broadcom/common/v3d_cpu_tiling.h
@@ -0,0 +1,222 @@
+/*
+ * Copyright © 2017 Broadcom
+ *
+ * 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.
+ */
+
+/** @file v3d_cpu_tiling.h
+ *
+ * Contains load/store functions common to both v3d and vc4. The utile layout
+ * stayed the same, though the way utiles get laid out has changed.
+ */
+
+static inline void
+v3d_load_utile(void *cpu, uint32_t cpu_stride,
+ void *gpu, uint32_t gpu_stride)
+{
+#if defined(V3D_BUILD_NEON) && defined(PIPE_ARCH_ARM)
+ if (gpu_stride == 8) {
+ __asm__ volatile (
+ /* Load from the GPU in one shot, no interleave, to
+ * d0-d7.
+ */
+ "vldm %0, {q0, q1, q2, q3}\n"
+ /* Store each 8-byte line to cpu-side destination,
+ * incrementing it by the stride each time.
+ */
+ "vst1.8 d0, [%1], %2\n"
+ "vst1.8 d1, [%1], %2\n"
+ "vst1.8 d2, [%1], %2\n"
+ "vst1.8 d3, [%1], %2\n"
+ "vst1.8 d4, [%1], %2\n"
+ "vst1.8 d5, [%1], %2\n"
+ "vst1.8 d6, [%1], %2\n"
+ "vst1.8 d7, [%1]\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : "q0", "q1", "q2", "q3");
+ } else {
+ assert(gpu_stride == 16);
+ __asm__ volatile (
+ /* Load from the GPU in one shot, no interleave, to
+ * d0-d7.
+ */
+ "vldm %0, {q0, q1, q2, q3};\n"
+ /* Store each 16-byte line in 2 parts to the cpu-side
+ * destination. (vld1 can only store one d-register
+ * at a time).
+ */
+ "vst1.8 d0, [%1], %3\n"
+ "vst1.8 d1, [%2], %3\n"
+ "vst1.8 d2, [%1], %3\n"
+ "vst1.8 d3, [%2], %3\n"
+ "vst1.8 d4, [%1], %3\n"
+ "vst1.8 d5, [%2], %3\n"
+ "vst1.8 d6, [%1]\n"
+ "vst1.8 d7, [%2]\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : "q0", "q1", "q2", "q3");
+ }
+#elif defined (PIPE_ARCH_AARCH64)
+ if (gpu_stride == 8) {
+ __asm__ volatile (
+ /* Load from the GPU in one shot, no interleave, to
+ * d0-d7.
+ */
+ "ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ /* Store each 8-byte line to cpu-side destination,
+ * incrementing it by the stride each time.
+ */
+ "st1 {v0.D}[0], [%1], %2\n"
+ "st1 {v0.D}[1], [%1], %2\n"
+ "st1 {v1.D}[0], [%1], %2\n"
+ "st1 {v1.D}[1], [%1], %2\n"
+ "st1 {v2.D}[0], [%1], %2\n"
+ "st1 {v2.D}[1], [%1], %2\n"
+ "st1 {v3.D}[0], [%1], %2\n"
+ "st1 {v3.D}[1], [%1]\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : "v0", "v1", "v2", "v3");
+ } else {
+ assert(gpu_stride == 16);
+ __asm__ volatile (
+ /* Load from the GPU in one shot, no interleave, to
+ * d0-d7.
+ */
+ "ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ /* Store each 16-byte line in 2 parts to the cpu-side
+ * destination. (vld1 can only store one d-register
+ * at a time).
+ */
+ "st1 {v0.D}[0], [%1], %3\n"
+ "st1 {v0.D}[1], [%2], %3\n"
+ "st1 {v1.D}[0], [%1], %3\n"
+ "st1 {v1.D}[1], [%2], %3\n"
+ "st1 {v2.D}[0], [%1], %3\n"
+ "st1 {v2.D}[1], [%2], %3\n"
+ "st1 {v3.D}[0], [%1]\n"
+ "st1 {v3.D}[1], [%2]\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : "v0", "v1", "v2", "v3");
+ }
+#else
+ for (uint32_t gpu_offset = 0; gpu_offset < 64; gpu_offset += gpu_stride) {
+ memcpy(cpu, gpu + gpu_offset, gpu_stride);
+ cpu += cpu_stride;
+ }
+#endif
+}
+
+static inline void
+v3d_store_utile(void *gpu, uint32_t gpu_stride,
+ void *cpu, uint32_t cpu_stride)
+{
+#if defined(V3D_BUILD_NEON) && defined(PIPE_ARCH_ARM)
+ if (gpu_stride == 8) {
+ __asm__ volatile (
+ /* Load each 8-byte line from cpu-side source,
+ * incrementing it by the stride each time.
+ */
+ "vld1.8 d0, [%1], %2\n"
+ "vld1.8 d1, [%1], %2\n"
+ "vld1.8 d2, [%1], %2\n"
+ "vld1.8 d3, [%1], %2\n"
+ "vld1.8 d4, [%1], %2\n"
+ "vld1.8 d5, [%1], %2\n"
+ "vld1.8 d6, [%1], %2\n"
+ "vld1.8 d7, [%1]\n"
+ /* Load from the GPU in one shot, no interleave, to
+ * d0-d7.
+ */
+ "vstm %0, {q0, q1, q2, q3}\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : "q0", "q1", "q2", "q3");
+ } else {
+ assert(gpu_stride == 16);
+ __asm__ volatile (
+ /* Load each 16-byte line in 2 parts from the cpu-side
+ * destination. (vld1 can only store one d-register
+ * at a time).
+ */
+ "vld1.8 d0, [%1], %3\n"
+ "vld1.8 d1, [%2], %3\n"
+ "vld1.8 d2, [%1], %3\n"
+ "vld1.8 d3, [%2], %3\n"
+ "vld1.8 d4, [%1], %3\n"
+ "vld1.8 d5, [%2], %3\n"
+ "vld1.8 d6, [%1]\n"
+ "vld1.8 d7, [%2]\n"
+ /* Store to the GPU in one shot, no interleave. */
+ "vstm %0, {q0, q1, q2, q3}\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : "q0", "q1", "q2", "q3");
+ }
+#elif defined (PIPE_ARCH_AARCH64)
+ if (gpu_stride == 8) {
+ __asm__ volatile (
+ /* Load each 8-byte line from cpu-side source,
+ * incrementing it by the stride each time.
+ */
+ "ld1 {v0.D}[0], [%1], %2\n"
+ "ld1 {v0.D}[1], [%1], %2\n"
+ "ld1 {v1.D}[0], [%1], %2\n"
+ "ld1 {v1.D}[1], [%1], %2\n"
+ "ld1 {v2.D}[0], [%1], %2\n"
+ "ld1 {v2.D}[1], [%1], %2\n"
+ "ld1 {v3.D}[0], [%1], %2\n"
+ "ld1 {v3.D}[1], [%1]\n"
+ /* Store to the GPU in one shot, no interleave. */
+ "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : "v0", "v1", "v2", "v3");
+ } else {
+ assert(gpu_stride == 16);
+ __asm__ volatile (
+ /* Load each 16-byte line in 2 parts from the cpu-side
+ * destination. (vld1 can only store one d-register
+ * at a time).
+ */
+ "ld1 {v0.D}[0], [%1], %3\n"
+ "ld1 {v0.D}[1], [%2], %3\n"
+ "ld1 {v1.D}[0], [%1], %3\n"
+ "ld1 {v1.D}[1], [%2], %3\n"
+ "ld1 {v2.D}[0], [%1], %3\n"
+ "ld1 {v2.D}[1], [%2], %3\n"
+ "ld1 {v3.D}[0], [%1]\n"
+ "ld1 {v3.D}[1], [%2]\n"
+ /* Store to the GPU in one shot, no interleave. */
+ "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ :
+ : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : "v0", "v1", "v2", "v3");
+ }
+#else
+ for (uint32_t gpu_offset = 0; gpu_offset < 64; gpu_offset += gpu_stride) {
+ memcpy(gpu + gpu_offset, cpu, gpu_stride);
+ cpu += cpu_stride;
+ }
+#endif
+}