Adding new trace decoder record types and new ATT parameters (#195)
* Adding new trace decoder record types and new ATT parameters * Add compatiblity with decoder 0.1.2 * Added RT * Format * Add logging to sdata values * Review comment * Review comments * Update projects/rocprofiler-sdk/source/include/rocprofiler-sdk/experimental/thread-trace/trace_decoder_types.h
This commit is contained in:
Giovanni Lenzi Baraldi
committed by
GitHub
GitHub
parent
83971da05e
commit
cb77f5af5c
@@ -1663,7 +1663,9 @@ ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_PARAMETER_SIMD_SELECT);
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTERS_CTRL);
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTER);
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_PARAMETER_SERIALIZE_ALL);
|
||||
static_assert(ROCPROFILER_THREAD_TRACE_PARAMETER_LAST == 7);
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTER_EXCLUDE_MASK);
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_PARAMETER_NO_DETAIL);
|
||||
static_assert(ROCPROFILER_THREAD_TRACE_PARAMETER_LAST == 9);
|
||||
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_CONTROL_NONE);
|
||||
ROCPROFILER_ENUM_LABEL(ROCPROFILER_THREAD_TRACE_CONTROL_START_AND_STOP);
|
||||
|
||||
+8
-3
@@ -47,9 +47,14 @@ typedef enum rocprofiler_thread_trace_parameter_type_t
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_BUFFER_SIZE, ///< Size of combined GPU buffer for ATT
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_SIMD_SELECT, ///< Bitmask (GFX9) or ID (Navi) of SIMDs
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTERS_CTRL, ///< Period [1,32] or disable (0) perfmon
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTER, ///< Perfmon ID and SIMD mask
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_SERIALIZE_ALL, ///< Serializes kernels not under thread
|
||||
///< trace
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTER, ///< Perfmon ID and SIMD mask. gfx9 only
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_SERIALIZE_ALL, ///< Serializes also kernels not under
|
||||
///< thread trace
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_PERFCOUNTER_EXCLUDE_MASK, ///< Bitmask of which compute
|
||||
///< units to exclude from
|
||||
///< perfcounters. gfx9 only
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_NO_DETAIL, ///< Dont collect instruction timing,
|
||||
///< only shader-wide information
|
||||
ROCPROFILER_THREAD_TRACE_PARAMETER_LAST
|
||||
} rocprofiler_thread_trace_parameter_type_t;
|
||||
|
||||
|
||||
+51
-1
@@ -40,6 +40,7 @@ typedef enum rocprofiler_thread_trace_decoder_info_t
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_INFO_NONE = 0,
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_INFO_DATA_LOST,
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_INFO_STITCH_INCOMPLETE,
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_INFO_WAVE_INCOMPLETE,
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_INFO_LAST
|
||||
} rocprofiler_thread_trace_decoder_info_t;
|
||||
|
||||
@@ -76,7 +77,7 @@ typedef struct rocprofiler_thread_trace_decoder_occupancy_t
|
||||
uint8_t reserved; ///< Reserved
|
||||
uint8_t cu; ///< Compute unit ID (gfx9) or WGP ID (gfx10+).
|
||||
uint8_t simd; ///< SIMD ID [0,3] within compute unit
|
||||
uint8_t slot; ///< Wave slot ID within SIMD
|
||||
uint8_t wave_id; ///< Wave slot ID within SIMD
|
||||
uint32_t start : 1; ///< 1 if wave_start, 0 if a wave_end
|
||||
uint32_t _rsvd : 31;
|
||||
} rocprofiler_thread_trace_decoder_occupancy_t;
|
||||
@@ -168,6 +169,49 @@ typedef struct rocprofiler_thread_trace_decoder_wave_t
|
||||
rocprofiler_thread_trace_decoder_inst_t* instructions_array; ///< Instructions executed
|
||||
} rocprofiler_thread_trace_decoder_wave_t;
|
||||
|
||||
/**
|
||||
* @brief Matches the reference (realtime) clock with the shader clock
|
||||
* Added in rocprof-trace-decoder 0.1.3. Requires aqlprofile for rocm 7.1+.
|
||||
* clock_in_seconds = realtime_clock / ROCPROFILER_THREAD_TRACE_DECODER_RECORD_RT_FREQUENCY
|
||||
* gfx_frequency = delta(shader_clock) / delta(clock_in_seconds)
|
||||
* For best average, use
|
||||
* gfx_frequency[n] = (shader_clock[n]-shader_clock[0]) / (clock_in_seconds[n]-clock_in_seconds[0])
|
||||
*/
|
||||
typedef struct rocprofiler_thread_trace_decoder_realtime_t
|
||||
{
|
||||
int64_t shader_clock; ///< Clock timestamp in gfx clock units
|
||||
uint64_t realtime_clock; ///< Clock timestamp in realtime units
|
||||
uint64_t reserved;
|
||||
} rocprofiler_thread_trace_decoder_realtime_t;
|
||||
|
||||
/**
|
||||
* @brief Bitmask of additional information for shaderdata_t
|
||||
* Added in rocprof-trace-decoder 0.1.3
|
||||
*/
|
||||
typedef enum rocprofiler_thread_trace_decoder_shaderdata_flags_t
|
||||
{
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_SHADERDATA_FLAGS_IMM = 0,
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_SHADERDATA_FLAGS_PRIV ///< Generated by the trap handler
|
||||
|
||||
/// @var ROCPROFILER_THREAD_TRACE_DECODER_SHADERDATA_FLAGS_IMM
|
||||
/// @brief Value comes from s_ttracedata_imm.
|
||||
} rocprofiler_thread_trace_decoder_shaderdata_flags_t;
|
||||
|
||||
/**
|
||||
* @brief Record created by s_ttracedata and s_ttracedata_imm
|
||||
* Added in rocprof-trace-decoder 0.1.3
|
||||
*/
|
||||
typedef struct rocprofiler_thread_trace_decoder_shaderdata_t
|
||||
{
|
||||
int64_t time;
|
||||
uint64_t value; ///< Value written from M0/IMM
|
||||
uint8_t cu; ///< CU id (gfx9) or wgp id (gfx10+). This is always the target_cu.
|
||||
uint8_t simd; ///< SIMD ID [0,3].
|
||||
uint8_t wave_id; ///< Wave slot ID within SIMD.
|
||||
uint8_t flags; ///< bitmask of rocprofiler_thread_trace_decoder_shaderdata_flags_t
|
||||
uint32_t reserved;
|
||||
} rocprofiler_thread_trace_decoder_shaderdata_t;
|
||||
|
||||
/**
|
||||
* @brief Defines the type of payload received by rocprofiler_thread_trace_decoder_callback_t
|
||||
*/
|
||||
@@ -179,7 +223,13 @@ typedef enum rocprofiler_thread_trace_decoder_record_type_t
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_WAVE, ///< rocprofiler_thread_trace_decoder_wave_t*
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_INFO, ///< rocprofiler_thread_trace_decoder_info_t*
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_DEBUG, ///< Debug
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_SHADERDATA, ///< rocprofiler_thread_trace_decoder_shaderdata_t*
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_REALTIME, ///< rocprofiler_thread_trace_decoder_realtime_t*
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_RT_FREQUENCY,
|
||||
ROCPROFILER_THREAD_TRACE_DECODER_RECORD_LAST
|
||||
|
||||
/// @var ROCPROFILER_THREAD_TRACE_DECODER_RECORD_RT_FREQUENCY
|
||||
/// @brief uint64_t*. Realtime clock frequency in Hz.
|
||||
} rocprofiler_thread_trace_decoder_record_type_t;
|
||||
|
||||
/** @} */
|
||||
|
||||
Reference in New Issue
Block a user