vc4: Declare the cpu pointers as being modified in NEON asm.

Otherwise, the compiler is free to reuse the register containing the input
for another call and assume that the value hasn't been modified.  Fixes
crashes on texture upload/download with current gcc.

We now have to have a temporary for the cpu2 value, since outputs must be
lvalues.

(commit message by anholt)

Fixes: 4d30024238ef ("vc4: Use NEON to speed up utile loads on Pi2.")
(cherry picked from commit 300d3ae8b1445b5060f92c77c0f577f4b7b2c7d6)
[Emil: apply the patch to vc4_tiling_lt.c instead of v3d_cpu_tiling.h]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Conflicts:
	src/broadcom/common/v3d_cpu_tiling.h

Squashed with commit:

vc4: Declare the last cpu pointer as being modified in NEON asm.

Earlier commit addressed 7 of the 8 instances available.

v2: Rebase patch back to master (by anholt)

Cc: Carsten Haitzler (Rasterman) <raster@rasterman.com>
Cc: Eric Anholt <eric@anholt.net>
Fixes: 300d3ae8b14 ("vc4: Declare the cpu pointers as being modified in NEON asm.")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 385843ac3ce1b868d9e24fcb2dbc0c8d5f5a7c99)

Conflicts:
	src/broadcom/common/v3d_cpu_tiling.h

diff --git a/src/gallium/drivers/vc4/vc4_tiling_lt.c b/src/gallium/drivers/vc4/vc4_tiling_lt.c
index df6236b..324a633 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_tiling_lt.c
+++ b/src/gallium/drivers/vc4/vc4_tiling_lt.c
@@ -85,13 +85,13 @@ vc4_load_utile(void *cpu, void *gpu, uint32_t cpu_stride, uint32_t cpp)
                         "vst1.8 d5, [%[cpu]], %[cpu_stride]\n"
                         "vst1.8 d6, [%[cpu]], %[cpu_stride]\n"
                         "vst1.8 d7, [%[cpu]]\n"
-                        :
+                        : [cpu]         "+r"(cpu)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
                           [cpu_stride]  "r"(cpu_stride)
                         : "q0", "q1", "q2", "q3");
         } else {
                 assert(gpu_stride == 16);
+                void *cpu2 = cpu + 8;
                 __asm__ volatile (
                         /* Load from the GPU in one shot, no interleave, to
                          * d0-d7.
@@ -109,10 +109,9 @@ vc4_load_utile(void *cpu, void *gpu, uint32_t cpu_stride, uint32_t cpp)
                         "vst1.8 d5, [%[cpu2]],%[cpu_stride]\n"
                         "vst1.8 d6, [%[cpu]]\n"
                         "vst1.8 d7, [%[cpu2]]\n"
-                        :
+                        : [cpu]         "+r"(cpu),
+                          [cpu2]        "+r"(cpu2)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
-                          [cpu2]        "r"(cpu + 8),
                           [cpu_stride]  "r"(cpu_stride)
                         : "q0", "q1", "q2", "q3");
         }
@@ -134,13 +133,13 @@ vc4_load_utile(void *cpu, void *gpu, uint32_t cpu_stride, uint32_t cpp)
                         "st1 {v2.D}[1], [%[cpu]], %[cpu_stride]\n"
                         "st1 {v3.D}[0], [%[cpu]], %[cpu_stride]\n"
                         "st1 {v3.D}[1], [%[cpu]]\n"
-                       :
+                        : [cpu]         "+r"(cpu)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
                           [cpu_stride]  "r"(cpu_stride)
                         : "v0", "v1", "v2", "v3");
         } else {
                 assert(gpu_stride == 16);
+                void *cpu2 = cpu + 8;
                 __asm__ volatile (
                         /* Load from the GPU in one shot, no interleave, to
                          * d0-d7.
@@ -158,10 +157,9 @@ vc4_load_utile(void *cpu, void *gpu, uint32_t cpu_stride, uint32_t cpp)
                         "st1 {v2.D}[1], [%[cpu2]],%[cpu_stride]\n"
                         "st1 {v3.D}[0], [%[cpu]]\n"
                         "st1 {v3.D}[1], [%[cpu2]]\n"
-                        :
+                        : [cpu]         "+r"(cpu),
+                          [cpu2]        "+r"(cpu2)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
-                          [cpu2]        "r"(cpu + 8),
                           [cpu_stride]  "r"(cpu_stride)
                         : "v0", "v1", "v2", "v3");
         }
@@ -196,13 +194,13 @@ vc4_store_utile(void *gpu, void *cpu, uint32_t cpu_stride, uint32_t cpp)
                          * d0-d7.
                          */
                         "vstm %[gpu], {q0, q1, q2, q3}\n"
-                        :
+                        : [cpu]         "r"(cpu)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
                           [cpu_stride]  "r"(cpu_stride)
                         : "q0", "q1", "q2", "q3");
         } else {
                 assert(gpu_stride == 16);
+                void *cpu2 = cpu + 8;
                 __asm__ volatile (
                         /* Load each 16-byte line in 2 parts from the cpu-side
                          * destination.  (vld1 can only store one d-register
@@ -218,10 +216,9 @@ vc4_store_utile(void *gpu, void *cpu, uint32_t cpu_stride, uint32_t cpp)
                         "vld1.8 d7, [%[cpu2]]\n"
                         /* Store to the GPU in one shot, no interleave. */
                         "vstm %[gpu], {q0, q1, q2, q3}\n"
-                        :
+                        : [cpu]         "+r"(cpu),
+                          [cpu2]        "+r"(cpu2)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
-                          [cpu2]        "r"(cpu + 8),
                           [cpu_stride]  "r"(cpu_stride)
                         : "q0", "q1", "q2", "q3");
         }
@@ -241,13 +238,13 @@ vc4_store_utile(void *gpu, void *cpu, uint32_t cpu_stride, uint32_t cpp)
                         "ld1 {v3.D}[1], [%[cpu]]\n"
                         /* Store to the GPU in one shot, no interleave. */
                         "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%[gpu]]\n"
-                        :
+                        : [cpu]         "+r"(cpu)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
                           [cpu_stride]  "r"(cpu_stride)
                         : "v0", "v1", "v2", "v3");
         } else {
                 assert(gpu_stride == 16);
+                void *cpu2 = cpu + 8;
                 __asm__ volatile (
                         /* Load each 16-byte line in 2 parts from the cpu-side
                          * destination.  (vld1 can only store one d-register
@@ -263,10 +260,9 @@ vc4_store_utile(void *gpu, void *cpu, uint32_t cpu_stride, uint32_t cpp)
                         "ld1 {v3.D}[1], [%[cpu2]]\n"
                         /* Store to the GPU in one shot, no interleave. */
                         "st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%[gpu]]\n"
-                        :
+                        : [cpu]         "+r"(cpu),
+                          [cpu2]        "+r"(cpu2)
                         : [gpu]         "r"(gpu),
-                          [cpu]         "r"(cpu),
-                          [cpu2]        "r"(cpu + 8),
                           [cpu_stride]  "r"(cpu_stride)
                         : "v0", "v1", "v2", "v3");
         }