gem_pwrite_pread: Use mmap to bypass pwrite/pread for comparison

Just to help sanity test alternate paths.

diff --git a/tests/gem_pwrite_pread.c b/tests/gem_pwrite_pread.c
index b400e1f..9f7c227 100644 (file)
--- a/tests/gem_pwrite_pread.c
+++ b/tests/gem_pwrite_pread.c
@@ -93,7 +93,112 @@ static void copy(int fd, uint32_t src, uint32_t dst, void *buf, int len, int loo
        gem_close(fd, exec[2].handle);
 }
 
-static void test(int fd, uint32_t src, uint32_t dst, uint32_t *buf, int len)
+static void as_gtt_mmap(int fd, uint32_t src, uint32_t dst, void *buf, int len, int loops)
+{
+       struct drm_i915_gem_relocation_entry reloc[] = {
+               { dst, 0, 4*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER },
+               { src, 0, 7*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, 0 },
+       };
+       struct drm_i915_gem_exec_object2 exec[] = {
+               { src },
+               { dst },
+               { gem_create(fd, 4096), 2, (uintptr_t)reloc }
+       };
+       uint32_t batch[] = {
+               COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB,
+               0xcc << 16 | 1 << 25 | 1 << 24 | len,
+               0,
+               1 << 16 | (len / 4),
+               0, /* dst */
+               0,
+               len,
+               0, /* src */
+               MI_BATCH_BUFFER_END,
+               0
+       };
+       struct drm_i915_gem_execbuffer2 execbuf = {
+               (uintptr_t)exec, 3,
+               0, sizeof(batch),
+               0, 0, 0, 0,
+               HAS_BLT_RING(intel_get_drm_devid(fd)) ? I915_EXEC_BLT : 0,
+       };
+       uint32_t *src_ptr, *dst_ptr;
+
+       gem_write(fd, exec[2].handle, 0, batch, sizeof(batch));
+
+       src_ptr = gem_mmap__gtt(fd, src, OBJECT_SIZE, PROT_WRITE);
+       dst_ptr = gem_mmap__gtt(fd, dst, OBJECT_SIZE, PROT_READ);
+
+       while (loops--) {
+               gem_set_domain(fd, src,
+                              I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT);
+               memcpy(src_ptr, buf, len);
+
+               do_or_die(drmIoctl(fd, DRM_IOCTL_I915_GEM_EXECBUFFER2, &execbuf));
+               gem_set_domain(fd, dst,
+                              I915_GEM_DOMAIN_GTT, 0);
+               memcpy(buf, dst_ptr, len);
+       }
+
+       munmap(dst_ptr, len);
+       munmap(src_ptr, len);
+       gem_close(fd, exec[2].handle);
+}
+
+
+static void as_cpu_mmap(int fd, uint32_t src, uint32_t dst, void *buf, int len, int loops)
+{
+       struct drm_i915_gem_relocation_entry reloc[] = {
+               { dst, 0, 4*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER },
+               { src, 0, 7*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, 0 },
+       };
+       struct drm_i915_gem_exec_object2 exec[] = {
+               { src },
+               { dst },
+               { gem_create(fd, 4096), 2, (uintptr_t)reloc }
+       };
+       uint32_t batch[] = {
+               COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB,
+               0xcc << 16 | 1 << 25 | 1 << 24 | len,
+               0,
+               1 << 16 | (len / 4),
+               0, /* dst */
+               0,
+               len,
+               0, /* src */
+               MI_BATCH_BUFFER_END,
+               0
+       };
+       struct drm_i915_gem_execbuffer2 execbuf = {
+               (uintptr_t)exec, 3,
+               0, sizeof(batch),
+               0, 0, 0, 0,
+               HAS_BLT_RING(intel_get_drm_devid(fd)) ? I915_EXEC_BLT : 0,
+       };
+       uint32_t *src_ptr, *dst_ptr;
+
+       gem_write(fd, exec[2].handle, 0, batch, sizeof(batch));
+
+       src_ptr = gem_mmap__cpu(fd, src, OBJECT_SIZE, PROT_WRITE);
+       dst_ptr = gem_mmap__cpu(fd, dst, OBJECT_SIZE, PROT_READ);
+
+       while (loops--) {
+               gem_set_domain(fd, src,
+                              I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
+               memcpy(src_ptr, buf, len);
+
+               do_or_die(drmIoctl(fd, DRM_IOCTL_I915_GEM_EXECBUFFER2, &execbuf));
+               gem_set_domain(fd, dst,
+                              I915_GEM_DOMAIN_CPU, 0);
+               memcpy(buf, dst_ptr, len);
+       }
+
+       munmap(dst_ptr, len);
+       munmap(src_ptr, len);
+       gem_close(fd, exec[2].handle);
+}
+
+static void test_copy(int fd, uint32_t src, uint32_t dst, uint32_t *buf, int len)
 {
        struct drm_i915_gem_relocation_entry reloc[] = {
                { dst, 0, 4*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER },
@@ -141,6 +246,110 @@ static void test(int fd, uint32_t src, uint32_t dst, uint32_t *buf, int len)
                assert(buf[i] == i);
 }
 
+static void test_as_gtt_mmap(int fd, uint32_t src, uint32_t dst, int len)
+{
+       struct drm_i915_gem_relocation_entry reloc[] = {
+               { dst, 0, 4*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER },
+               { src, 0, 7*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, 0 },
+       };
+       struct drm_i915_gem_exec_object2 exec[] = {
+               { src },
+               { dst },
+               { gem_create(fd, 4096), 2, (uintptr_t)reloc }
+       };
+       uint32_t batch[] = {
+               COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB,
+               0xcc << 16 | 1 << 25 | 1 << 24 | len,
+               0,
+               1 << 16 | (len / 4),
+               0, /* dst */
+               0,
+               len,
+               0, /* src */
+               MI_BATCH_BUFFER_END,
+               0
+       };
+       struct drm_i915_gem_execbuffer2 execbuf = {
+               (uintptr_t)exec, 3,
+               0, sizeof(batch),
+               0, 0, 0, 0,
+               HAS_BLT_RING(intel_get_drm_devid(fd)) ? I915_EXEC_BLT : 0,
+       };
+       uint32_t *src_ptr, *dst_ptr;
+       int i;
+
+       gem_write(fd, exec[2].handle, 0, batch, sizeof(batch));
+
+       src_ptr = gem_mmap__gtt(fd, src, OBJECT_SIZE, PROT_WRITE);
+       dst_ptr = gem_mmap__gtt(fd, dst, OBJECT_SIZE, PROT_READ);
+
+       gem_set_domain(fd, src, I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT);
+       for (i = 0; i < len/4; i++)
+               src_ptr[i] = i;
+
+       do_or_die(drmIoctl(fd, DRM_IOCTL_I915_GEM_EXECBUFFER2, &execbuf));
+       gem_close(fd, exec[2].handle);
+
+       gem_set_domain(fd, dst, I915_GEM_DOMAIN_GTT, 0);
+       for (i = 0; i < len/4; i++)
+               assert(dst_ptr[i] == i);
+
+       munmap(dst_ptr, len);
+       munmap(src_ptr, len);
+}
+
+static void test_as_cpu_mmap(int fd, uint32_t src, uint32_t dst, int len)
+{
+       struct drm_i915_gem_relocation_entry reloc[] = {
+               { dst, 0, 4*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER },
+               { src, 0, 7*sizeof(uint32_t), 0, I915_GEM_DOMAIN_RENDER, 0 },
+       };
+       struct drm_i915_gem_exec_object2 exec[] = {
+               { src },
+               { dst },
+               { gem_create(fd, 4096), 2, (uintptr_t)reloc }
+       };
+       uint32_t batch[] = {
+               COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB,
+               0xcc << 16 | 1 << 25 | 1 << 24 | len,
+               0,
+               1 << 16 | (len / 4),
+               0, /* dst */
+               0,
+               len,
+               0, /* src */
+               MI_BATCH_BUFFER_END,
+               0
+       };
+       struct drm_i915_gem_execbuffer2 execbuf = {
+               (uintptr_t)exec, 3,
+               0, sizeof(batch),
+               0, 0, 0, 0,
+               HAS_BLT_RING(intel_get_drm_devid(fd)) ? I915_EXEC_BLT : 0,
+       };
+       uint32_t *src_ptr, *dst_ptr;
+       int i;
+
+       gem_write(fd, exec[2].handle, 0, batch, sizeof(batch));
+
+       src_ptr = gem_mmap__cpu(fd, src, OBJECT_SIZE, PROT_WRITE);
+       dst_ptr = gem_mmap__cpu(fd, dst, OBJECT_SIZE, PROT_READ);
+
+       gem_set_domain(fd, src, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
+       for (i = 0; i < len/4; i++)
+               src_ptr[i] = i;
+
+       do_or_die(drmIoctl(fd, DRM_IOCTL_I915_GEM_EXECBUFFER2, &execbuf));
+       gem_close(fd, exec[2].handle);
+
+       gem_set_domain(fd, dst, I915_GEM_DOMAIN_CPU, 0);
+       for (i = 0; i < len/4; i++)
+               assert(dst_ptr[i] == i);
+
+       munmap(dst_ptr, len);
+       munmap(src_ptr, len);
+}
+
 static double elapsed(const struct timeval *start,
                      const struct timeval *end,
                      int loop)
@@ -190,7 +399,7 @@ int main(int argc, char **argv)
 
        gem_set_cacheing(fd, src, 0);
        gem_set_cacheing(fd, dst, 0);
-       test(fd, src, dst, tmp, object_size);
+       test_copy(fd, src, dst, tmp, object_size);
        for (count = 1; count <= 1<<17; count <<= 1) {
                struct timeval start, end;
 
@@ -204,9 +413,23 @@ int main(int argc, char **argv)
                fflush(stdout);
        }
 
+       test_as_gtt_mmap(fd, src, dst, object_size);
+       for (count = 1; count <= 1<<17; count <<= 1) {
+               struct timeval start, end;
+
+               gettimeofday(&start, NULL);
+               as_gtt_mmap(fd, src, dst, tmp, object_size, count);
+               gettimeofday(&end, NULL);
+               printf("** mmap uncached copy %d bytes x %6d:   %7.3fµs, %s\n",
+                      object_size, count,
+                      elapsed(&start, &end, count),
+                      bytes_per_sec((char *)buf, object_size/elapsed(&start, &end, count)*1e6));
+               fflush(stdout);
+       }
+
        gem_set_cacheing(fd, src, 1);
        gem_set_cacheing(fd, dst, 1);
-       test(fd, src, dst, tmp, object_size);
+       test_copy(fd, src, dst, tmp, object_size);
        for (count = 1; count <= 1<<17; count <<= 1) {
                struct timeval start, end;
 
@@ -220,6 +443,20 @@ int main(int argc, char **argv)
                fflush(stdout);
        }
 
+       test_as_cpu_mmap(fd, src, dst, object_size);
+       for (count = 1; count <= 1<<17; count <<= 1) {
+               struct timeval start, end;
+
+               gettimeofday(&start, NULL);
+               as_cpu_mmap(fd, src, dst, tmp, object_size, count);
+               gettimeofday(&end, NULL);
+               printf("** mmap snooped copy %d bytes x %6d:    %7.3fµs, %s\n",
+                      object_size, count,
+                      elapsed(&start, &end, count),
+                      bytes_per_sec((char *)buf, object_size/elapsed(&start, &end, count)*1e6));
+               fflush(stdout);
+       }
+
        free(tmp);
        gem_close(fd, src);
        gem_close(fd, dst);