.. meta:: :description: ROCm Compute Profiler - RDNA3.5 WGP / roofline metrics :keywords: ROCm Compute Profiler, RDNA, gfx115x, WGP, roofline .. _rdna-wgp: ========================= Workgroup processor (WGP) ========================= Within each shader engine, Workgroup Processors (WGPs) pair two Compute Units (CUs) that share resources and execute dispatched waves after the Workgroup Manager (SPI) hands off work. On RDNA3.5 architecture-based GPUs/APUs, compute kernels are typically tracked with wave32-oriented waves; the gfx115x WGP panels cover occupancy, dispatch, instruction mix, and local caches at that WGP/CU-pair granularity. The sections below list RDNA3.5 (gfx115x) metric descriptions. .. Note:: The CDNA architecture-based AMD Instinct GPUs use a different execution hierarchy and panel grouping. For Instinct-only pipeline metrics (for example, VALU / VMEM / MFMA-style tables), see :doc:`../cdna/cdna-performance-model`-without assuming RDNA WGPs or CUs map directly to those layouts. Roofline ======== Roofline performance rates -------------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-roofline-performance-rates-gfx115x :file: _templates/metrics_table.j2 Roofline plot points -------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-roofline-plot-points-gfx115x :file: _templates/metrics_table.j2 WGP block metrics ================= WGP utilization --------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wgp-utilization-gfx115x :file: _templates/metrics_table.j2 Wavefront launch stats ---------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wavefront-launch-stats-gfx115x :file: _templates/metrics_table.j2 Wave dispatch ------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wave-dispatch-gfx115x :file: _templates/metrics_table.j2 Wave life --------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wave-life-gfx115x :file: _templates/metrics_table.j2 Wave instruction mix -------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wave-instruction-mix-gfx115x :file: _templates/metrics_table.j2 .. note:: The **Instructions - Dual VALU (VOPD)** row counts ``SQ_INSTS_DUAL_VALU_WAVE32`` — the number of VOPD-encoded wave32 instructions issued per normalization unit. Each unit corresponds to one dual-issue VALU slot consumed; kernels that use VOPD heavily typically show a large value relative to the **Instructions - VALU** row. The exact ratio depends on ISA counting semantics for paired operations, so treat the two rows as complementary signals rather than a fixed ceiling relationship. For peak-rate theory and FLOPs ceilings see :ref:`rdna-dual-issue-valu`. VMEM instruction mix -------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-vmem-instruction-mix-gfx115x :file: _templates/metrics_table.j2 LDS instruction mix ------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-lds-instruction-mix-gfx115x :file: _templates/metrics_table.j2 Wait state analysis ------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wait-state-analysis-gfx115x :file: _templates/metrics_table.j2 WGP instruction cache --------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wgp-instruction-cache-gfx115x :file: _templates/metrics_table.j2 WGP scalar data cache --------------------- .. tab-set:: .. tab-item:: RDNA3.5 (gfx115x) :selected: .. jinja:: rdna115x-wgp-scalar-data-cache-gfx115x :file: _templates/metrics_table.j2