summaryrefslogtreecommitdiffstats
path: root/src/broadcom/common/v3d_cpu_tiling.h
diff options
context:
space:
mode:
Diffstat (limited to 'src/broadcom/common/v3d_cpu_tiling.h')
-rw-r--r--src/broadcom/common/v3d_cpu_tiling.h180
1 files changed, 100 insertions, 80 deletions
diff --git a/src/broadcom/common/v3d_cpu_tiling.h b/src/broadcom/common/v3d_cpu_tiling.h
index 82ac2aef6a6..246d3d6d3ab 100644
--- a/src/broadcom/common/v3d_cpu_tiling.h
+++ b/src/broadcom/common/v3d_cpu_tiling.h
@@ -37,20 +37,22 @@ v3d_load_utile(void *cpu, uint32_t cpu_stride,
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
- "vldm %0, {q0, q1, q2, q3}\n"
+ "vldm %[gpu], {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"
+ "vst1.8 d0, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d1, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d2, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d3, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d4, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d5, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d6, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d7, [%[cpu]]\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu_stride] "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
} else if (gpu_stride == 16) {
@@ -58,21 +60,24 @@ v3d_load_utile(void *cpu, uint32_t cpu_stride,
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
- "vldm %0, {q0, q1, q2, q3};\n"
+ "vldm %[gpu], {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"
+ "vst1.8 d0, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d1, [%[cpu2]],%[cpu_stride]\n"
+ "vst1.8 d2, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d3, [%[cpu2]],%[cpu_stride]\n"
+ "vst1.8 d4, [%[cpu]], %[cpu_stride]\n"
+ "vst1.8 d5, [%[cpu2]],%[cpu_stride]\n"
+ "vst1.8 d6, [%[cpu]]\n"
+ "vst1.8 d7, [%[cpu2]]\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu2] "r"(cpu + 8),
+ [cpu_stride] "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
}
@@ -82,20 +87,22 @@ v3d_load_utile(void *cpu, uint32_t cpu_stride,
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
- "ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ "ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%[gpu]]\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"
+ "st1 {v0.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v0.D}[1], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v1.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v1.D}[1], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v2.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v2.D}[1], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v3.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v3.D}[1], [%[cpu]]\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu_stride] "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
} else if (gpu_stride == 16) {
@@ -103,21 +110,24 @@ v3d_load_utile(void *cpu, uint32_t cpu_stride,
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
- "ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ "ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%[gpu]]\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"
+ "st1 {v0.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v0.D}[1], [%[cpu2]],%[cpu_stride]\n"
+ "st1 {v1.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v1.D}[1], [%[cpu2]],%[cpu_stride]\n"
+ "st1 {v2.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "st1 {v2.D}[1], [%[cpu2]],%[cpu_stride]\n"
+ "st1 {v3.D}[0], [%[cpu]]\n"
+ "st1 {v3.D}[1], [%[cpu2]]\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu2] "r"(cpu + 8),
+ [cpu_stride] "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
}
@@ -139,20 +149,22 @@ v3d_store_utile(void *gpu, uint32_t gpu_stride,
/* 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"
+ "vld1.8 d0, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d1, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d2, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d3, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d4, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d5, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d6, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d7, [%[cpu]]\n"
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
- "vstm %0, {q0, q1, q2, q3}\n"
+ "vstm %[gpu], {q0, q1, q2, q3}\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu_stride] "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
} else if (gpu_stride == 16) {
@@ -161,18 +173,21 @@ v3d_store_utile(void *gpu, uint32_t gpu_stride,
* 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"
+ "vld1.8 d0, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d1, [%[cpu2]],%[cpu_stride]\n"
+ "vld1.8 d2, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d3, [%[cpu2]],%[cpu_stride]\n"
+ "vld1.8 d4, [%[cpu]], %[cpu_stride]\n"
+ "vld1.8 d5, [%[cpu2]],%[cpu_stride]\n"
+ "vld1.8 d6, [%[cpu]]\n"
+ "vld1.8 d7, [%[cpu2]]\n"
/* Store to the GPU in one shot, no interleave. */
- "vstm %0, {q0, q1, q2, q3}\n"
+ "vstm %[gpu], {q0, q1, q2, q3}\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu2] "r"(cpu + 8),
+ [cpu_stride] "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
}
@@ -182,18 +197,20 @@ v3d_store_utile(void *gpu, uint32_t gpu_stride,
/* 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"
+ "ld1 {v0.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v0.D}[1], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v1.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v1.D}[1], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v2.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v2.D}[1], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v3.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v3.D}[1], [%[cpu]]\n"
/* Store to the GPU in one shot, no interleave. */
- "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%[gpu]]\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu_stride] "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
} else if (gpu_stride == 16) {
@@ -202,18 +219,21 @@ v3d_store_utile(void *gpu, uint32_t gpu_stride,
* 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"
+ "ld1 {v0.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v0.D}[1], [%[cpu2]],%[cpu_stride]\n"
+ "ld1 {v1.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v1.D}[1], [%[cpu2]],%[cpu_stride]\n"
+ "ld1 {v2.D}[0], [%[cpu]], %[cpu_stride]\n"
+ "ld1 {v2.D}[1], [%[cpu2]],%[cpu_stride]\n"
+ "ld1 {v3.D}[0], [%[cpu]]\n"
+ "ld1 {v3.D}[1], [%[cpu2]]\n"
/* Store to the GPU in one shot, no interleave. */
- "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
+ "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%[gpu]]\n"
:
- : "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
+ : [gpu] "r"(gpu),
+ [cpu] "r"(cpu),
+ [cpu2] "r"(cpu + 8),
+ [cpu_stride] "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-19 20:23:50 +0000