[adriadam10]

What

s_memtime (SMEM 0x24) and s_memrealtime (SMEM 0x25) were rejected by the decoder (SMEM opcode is not supported). Each one reads a 64-bit counter into an SGPR pair. Shaders use them for timing and for cheap per-frame seeds.

• Decoder: both decode with their SDST pair and no memory operands.
• IR: ShaderClock / RealtimeClock (U64). The translation writes the low dword to sdst and the high dword to sdst+1.
• SPIR-V: OpReadClockKHR. s_memtime uses Subgroup scope (the per-wave shader clock) and s_memrealtime uses Device scope (the constant real-time clock). The emitter adds ShaderClockKHR / SPV_KHR_shader_clock only when used. If the target lacks them it fails with reads the shader clock, which needs the device's VK_KHR_shader_clock, the same way MinLod does. It does not return a constant.
• VulkanDevice enables VK_KHR_shader_clock when the device has both shaderSubgroupClock and shaderDeviceClock (RADV, lavapipe, NVIDIA and Intel all do).

References: shadPS4 lowers S_MEMTIME to OpReadClockKHR with subgroup scope and returns the constant 1 when the device has no shader clock. KytyPS5 leaves both opcodes unimplemented.

Test agc_driver_shader_clock: each clock is read before and after a 2000-iteration loop. The test checks that both 64-bit values are non-zero and never go backwards, that the loop result is right, and that the values are uniform over the wave. On lavapipe (subgroup size 8) the wave32 spans several host subgroups, so subgroup-scope reads can differ between groups of 8 lanes. The uniformity check is per group of 8 for that reason.

Tested

Linux, full ctest (100) passes.

• RADV (Radeon 660M) and lavapipe: the test passes (5/5 runs on RADV).
• Without the decoder change: SMEM opcode is not supported. Without the device change: opcode=ShaderClock reason=reads the shader clock, which needs the device's VK_KHR_shader_clock.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed in TechnicalDebt
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: ci: run the shader execution tests on lavapipe #251 (test harness), feat(shader): decode cache invalidations and performance hints as no-ops #253 (SMEM cache-control decode); review only the last commit
• AI-assisted: yes

[github-actions]

Progress report (709d7fe → 7d90b1d)

📈 GPU shader instructions: 53.52% (+0.86%, +10 implemented instructions)

✅ 10 implemented

Encoding	Name
MUBUF	BUFFER_GL0_INV
MUBUF	BUFFER_GL1_INV
SMEM	S_DCACHE_INV
SMEM	S_GL1_INV
SMEM	S_MEMREALTIME
SMEM	S_MEMTIME
SOPP	S_DECPERFLEVEL
SOPP	S_ICACHE_INV
SOPP	S_INCPERFLEVEL
SOPP	S_WAIT_IDLE