Re: [Mesa-dev] [PATCH] vulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]
Bas Nieuwenhuizen writes: > Reviewed-by: Bas Nieuwenhuizen Thanks to you, Jason and Lionel for reviewing the code and helping improve it. -- -keith signature.asc Description: PGP signature ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [Mesa-dev] [PATCH] vulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]
Reviewed-by: Bas Nieuwenhuizen
On Wed, Oct 17, 2018 at 6:49 PM Keith Packard wrote:
>
> Offers three clocks, device, clock monotonic and clock monotonic
> raw. Could use some kernel support to reduce the deviation between
> clock values.
>
> v2:
> Ensure deviation is at least as big as the GPU time interval.
>
> v3:
> Set device->lost when returning DEVICE_LOST.
> Use MAX2 and DIV_ROUND_UP instead of open coding these.
> Delete spurious TIMESTAMP in radv version.
>
> Suggested-by: Jason Ekstrand
> Suggested-by: Lionel Landwerlin
>
> v4:
> Add anv_gem_reg_read to anv_gem_stubs.c
>
> Suggested-by: Jason Ekstrand
>
> v5:
> Adjust maxDeviation computation to max(sampled_clock_period) +
> sample_interval.
>
> Suggested-by: Bas Nieuwenhuizen
> Suggested-by: Jason Ekstrand
>
> Signed-off-by: Keith Packard
> ---
> src/amd/vulkan/radv_device.c | 119 +++
> src/amd/vulkan/radv_extensions.py | 1 +
> src/intel/vulkan/anv_device.c | 127 +
> src/intel/vulkan/anv_extensions.py | 1 +
> src/intel/vulkan/anv_gem.c | 13 +++
> src/intel/vulkan/anv_gem_stubs.c | 7 ++
> src/intel/vulkan/anv_private.h | 2 +
> 7 files changed, 270 insertions(+)
>
> diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
> index 174922780fc..4a705a724ef 100644
> --- a/src/amd/vulkan/radv_device.c
> +++ b/src/amd/vulkan/radv_device.c
> @@ -4955,3 +4955,122 @@ radv_GetDeviceGroupPeerMemoryFeatures(
>VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
>VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
> }
> +
> +static const VkTimeDomainEXT radv_time_domains[] = {
> + VK_TIME_DOMAIN_DEVICE_EXT,
> + VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT,
> + VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT,
> +};
> +
> +VkResult radv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
> + VkPhysicalDevice physicalDevice,
> + uint32_t *pTimeDomainCount,
> + VkTimeDomainEXT *pTimeDomains)
> +{
> + int d;
> + VK_OUTARRAY_MAKE(out, pTimeDomains, pTimeDomainCount);
> +
> + for (d = 0; d < ARRAY_SIZE(radv_time_domains); d++) {
> + vk_outarray_append(, i) {
> + *i = radv_time_domains[d];
> + }
> + }
> +
> + return vk_outarray_status();
> +}
> +
> +static uint64_t
> +radv_clock_gettime(clockid_t clock_id)
> +{
> + struct timespec current;
> + int ret;
> +
> + ret = clock_gettime(clock_id, );
> + if (ret < 0 && clock_id == CLOCK_MONOTONIC_RAW)
> + ret = clock_gettime(CLOCK_MONOTONIC, );
> + if (ret < 0)
> + return 0;
> +
> + return (uint64_t) current.tv_sec * 10ULL + current.tv_nsec;
> +}
> +
> +VkResult radv_GetCalibratedTimestampsEXT(
> + VkDevice _device,
> + uint32_t timestampCount,
> + const VkCalibratedTimestampInfoEXT *pTimestampInfos,
> + uint64_t *pTimestamps,
> + uint64_t *pMaxDeviation)
> +{
> + RADV_FROM_HANDLE(radv_device, device, _device);
> + uint32_t clock_crystal_freq =
> device->physical_device->rad_info.clock_crystal_freq;
> + int d;
> + uint64_t begin, end;
> +uint64_t max_clock_period = 0;
> +
> + begin = radv_clock_gettime(CLOCK_MONOTONIC_RAW);
> +
> + for (d = 0; d < timestampCount; d++) {
> + switch (pTimestampInfos[d].timeDomain) {
> + case VK_TIME_DOMAIN_DEVICE_EXT:
> + pTimestamps[d] = device->ws->query_value(device->ws,
> +
> RADEON_TIMESTAMP);
> +uint64_t device_period = DIV_ROUND_UP(100,
> clock_crystal_freq);
> +max_clock_period = MAX2(max_clock_period,
> device_period);
> + break;
> + case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT:
> + pTimestamps[d] = radv_clock_gettime(CLOCK_MONOTONIC);
> +max_clock_period = MAX2(max_clock_period, 1);
> + break;
> +
> + case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT:
> + pTimestamps[d] = begin;
> + break;
> + default:
> + pTimestamps[d] = 0;
> + break;
> + }
> + }
> +
> + end = radv_clock_gettime(CLOCK_MONOTONIC_RAW);
> +
> +/*
> + * The maximum deviation is the sum of the interval over which we
> + * perform the sampling and the maximum period of any
Re: [Mesa-dev] [PATCH] vulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]
Jason Ekstrand writes: > I like it When the comments are longer than the code, you know you're done? -- -keith signature.asc Description: PGP signature ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [Mesa-dev] [PATCH] vulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]
I like it
Reviewed-by: Jason Ekstrand
On Wed, Oct 17, 2018 at 11:49 AM Keith Packard wrote:
> Offers three clocks, device, clock monotonic and clock monotonic
> raw. Could use some kernel support to reduce the deviation between
> clock values.
>
> v2:
> Ensure deviation is at least as big as the GPU time interval.
>
> v3:
> Set device->lost when returning DEVICE_LOST.
> Use MAX2 and DIV_ROUND_UP instead of open coding these.
> Delete spurious TIMESTAMP in radv version.
>
> Suggested-by: Jason Ekstrand
> Suggested-by: Lionel Landwerlin
>
> v4:
> Add anv_gem_reg_read to anv_gem_stubs.c
>
> Suggested-by: Jason Ekstrand
>
> v5:
> Adjust maxDeviation computation to max(sampled_clock_period) +
> sample_interval.
>
> Suggested-by: Bas Nieuwenhuizen
> Suggested-by: Jason Ekstrand
>
> Signed-off-by: Keith Packard
> ---
> src/amd/vulkan/radv_device.c | 119 +++
> src/amd/vulkan/radv_extensions.py | 1 +
> src/intel/vulkan/anv_device.c | 127 +
> src/intel/vulkan/anv_extensions.py | 1 +
> src/intel/vulkan/anv_gem.c | 13 +++
> src/intel/vulkan/anv_gem_stubs.c | 7 ++
> src/intel/vulkan/anv_private.h | 2 +
> 7 files changed, 270 insertions(+)
>
> diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
> index 174922780fc..4a705a724ef 100644
> --- a/src/amd/vulkan/radv_device.c
> +++ b/src/amd/vulkan/radv_device.c
> @@ -4955,3 +4955,122 @@ radv_GetDeviceGroupPeerMemoryFeatures(
>VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
>VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
> }
> +
> +static const VkTimeDomainEXT radv_time_domains[] = {
> + VK_TIME_DOMAIN_DEVICE_EXT,
> + VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT,
> + VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT,
> +};
> +
> +VkResult radv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
> + VkPhysicalDevice physicalDevice,
> + uint32_t *pTimeDomainCount,
> + VkTimeDomainEXT *pTimeDomains)
> +{
> + int d;
> + VK_OUTARRAY_MAKE(out, pTimeDomains, pTimeDomainCount);
> +
> + for (d = 0; d < ARRAY_SIZE(radv_time_domains); d++) {
> + vk_outarray_append(, i) {
> + *i = radv_time_domains[d];
> + }
> + }
> +
> + return vk_outarray_status();
> +}
> +
> +static uint64_t
> +radv_clock_gettime(clockid_t clock_id)
> +{
> + struct timespec current;
> + int ret;
> +
> + ret = clock_gettime(clock_id, );
> + if (ret < 0 && clock_id == CLOCK_MONOTONIC_RAW)
> + ret = clock_gettime(CLOCK_MONOTONIC, );
> + if (ret < 0)
> + return 0;
> +
> + return (uint64_t) current.tv_sec * 10ULL + current.tv_nsec;
> +}
> +
> +VkResult radv_GetCalibratedTimestampsEXT(
> + VkDevice _device,
> + uint32_t timestampCount,
> + const VkCalibratedTimestampInfoEXT *pTimestampInfos,
> + uint64_t *pTimestamps,
> + uint64_t *pMaxDeviation)
> +{
> + RADV_FROM_HANDLE(radv_device, device, _device);
> + uint32_t clock_crystal_freq =
> device->physical_device->rad_info.clock_crystal_freq;
> + int d;
> + uint64_t begin, end;
> +uint64_t max_clock_period = 0;
> +
> + begin = radv_clock_gettime(CLOCK_MONOTONIC_RAW);
> +
> + for (d = 0; d < timestampCount; d++) {
> + switch (pTimestampInfos[d].timeDomain) {
> + case VK_TIME_DOMAIN_DEVICE_EXT:
> + pTimestamps[d] =
> device->ws->query_value(device->ws,
> +
> RADEON_TIMESTAMP);
> +uint64_t device_period = DIV_ROUND_UP(100,
> clock_crystal_freq);
> +max_clock_period = MAX2(max_clock_period,
> device_period);
> + break;
> + case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT:
> + pTimestamps[d] =
> radv_clock_gettime(CLOCK_MONOTONIC);
> +max_clock_period = MAX2(max_clock_period, 1);
> + break;
> +
> + case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT:
> + pTimestamps[d] = begin;
> + break;
> + default:
> + pTimestamps[d] = 0;
> + break;
> + }
> + }
> +
> + end = radv_clock_gettime(CLOCK_MONOTONIC_RAW);
> +
> +/*
> + * The maximum deviation is the sum of the interval over which we
> + * perform the sampling and the maximum period of any sampled
> + * clock. That's because the maximum
[PATCH] vulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]
Offers three clocks, device, clock monotonic and clock monotonic
raw. Could use some kernel support to reduce the deviation between
clock values.
v2:
Ensure deviation is at least as big as the GPU time interval.
v3:
Set device->lost when returning DEVICE_LOST.
Use MAX2 and DIV_ROUND_UP instead of open coding these.
Delete spurious TIMESTAMP in radv version.
Suggested-by: Jason Ekstrand
Suggested-by: Lionel Landwerlin
v4:
Add anv_gem_reg_read to anv_gem_stubs.c
Suggested-by: Jason Ekstrand
v5:
Adjust maxDeviation computation to max(sampled_clock_period) +
sample_interval.
Suggested-by: Bas Nieuwenhuizen
Suggested-by: Jason Ekstrand
Signed-off-by: Keith Packard
---
src/amd/vulkan/radv_device.c | 119 +++
src/amd/vulkan/radv_extensions.py | 1 +
src/intel/vulkan/anv_device.c | 127 +
src/intel/vulkan/anv_extensions.py | 1 +
src/intel/vulkan/anv_gem.c | 13 +++
src/intel/vulkan/anv_gem_stubs.c | 7 ++
src/intel/vulkan/anv_private.h | 2 +
7 files changed, 270 insertions(+)
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 174922780fc..4a705a724ef 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -4955,3 +4955,122 @@ radv_GetDeviceGroupPeerMemoryFeatures(
VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
}
+
+static const VkTimeDomainEXT radv_time_domains[] = {
+ VK_TIME_DOMAIN_DEVICE_EXT,
+ VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT,
+ VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT,
+};
+
+VkResult radv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
+ VkPhysicalDevice physicalDevice,
+ uint32_t *pTimeDomainCount,
+ VkTimeDomainEXT *pTimeDomains)
+{
+ int d;
+ VK_OUTARRAY_MAKE(out, pTimeDomains, pTimeDomainCount);
+
+ for (d = 0; d < ARRAY_SIZE(radv_time_domains); d++) {
+ vk_outarray_append(, i) {
+ *i = radv_time_domains[d];
+ }
+ }
+
+ return vk_outarray_status();
+}
+
+static uint64_t
+radv_clock_gettime(clockid_t clock_id)
+{
+ struct timespec current;
+ int ret;
+
+ ret = clock_gettime(clock_id, );
+ if (ret < 0 && clock_id == CLOCK_MONOTONIC_RAW)
+ ret = clock_gettime(CLOCK_MONOTONIC, );
+ if (ret < 0)
+ return 0;
+
+ return (uint64_t) current.tv_sec * 10ULL + current.tv_nsec;
+}
+
+VkResult radv_GetCalibratedTimestampsEXT(
+ VkDevice _device,
+ uint32_t timestampCount,
+ const VkCalibratedTimestampInfoEXT *pTimestampInfos,
+ uint64_t *pTimestamps,
+ uint64_t *pMaxDeviation)
+{
+ RADV_FROM_HANDLE(radv_device, device, _device);
+ uint32_t clock_crystal_freq =
device->physical_device->rad_info.clock_crystal_freq;
+ int d;
+ uint64_t begin, end;
+uint64_t max_clock_period = 0;
+
+ begin = radv_clock_gettime(CLOCK_MONOTONIC_RAW);
+
+ for (d = 0; d < timestampCount; d++) {
+ switch (pTimestampInfos[d].timeDomain) {
+ case VK_TIME_DOMAIN_DEVICE_EXT:
+ pTimestamps[d] = device->ws->query_value(device->ws,
+
RADEON_TIMESTAMP);
+uint64_t device_period = DIV_ROUND_UP(100,
clock_crystal_freq);
+max_clock_period = MAX2(max_clock_period,
device_period);
+ break;
+ case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT:
+ pTimestamps[d] = radv_clock_gettime(CLOCK_MONOTONIC);
+max_clock_period = MAX2(max_clock_period, 1);
+ break;
+
+ case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT:
+ pTimestamps[d] = begin;
+ break;
+ default:
+ pTimestamps[d] = 0;
+ break;
+ }
+ }
+
+ end = radv_clock_gettime(CLOCK_MONOTONIC_RAW);
+
+/*
+ * The maximum deviation is the sum of the interval over which we
+ * perform the sampling and the maximum period of any sampled
+ * clock. That's because the maximum skew between any two sampled
+ * clock edges is when the sampled clock with the largest period is
+ * sampled at the end of that period but right at the beginning of the
+ * sampling interval and some other clock is sampled right at the
+ * begining of its
