Module: Mesa Branch: main Commit: d09ad16fd4a0596fb6c97cffaf0fdf031053b5a4 URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=d09ad16fd4a0596fb6c97cffaf0fdf031053b5a4
Author: Timur Kristóf <timur.kris...@gmail.com> Date: Wed Nov 8 10:22:23 2023 +0100 ac: Remove CIK prefix from SDMA opcodes. The vast majority of AMD GPUs (except the very first GCN) have the same SDMA packet format, so let's just call it SDMA instead of CIK_SDMA. (And leave the oldest GPUs with SI_SDMA as they are now.) Signed-off-by: Timur Kristóf <timur.kris...@gmail.com> Reviewed-by: Tatsuyuki Ishi <ishitatsuy...@gmail.com> Reviewed-by: Samuel Pitoiset <samuel.pitoi...@gmail.com> Reviewed-by: Marek Olšák <marek.ol...@amd.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26110> --- src/amd/common/ac_debug.c | 16 +++++----- src/amd/common/sid.h | 38 +++++++++++------------ src/amd/vulkan/radv_cs.h | 4 +-- src/amd/vulkan/radv_sdma.c | 12 +++---- src/amd/vulkan/si_cmd_buffer.c | 2 +- src/gallium/drivers/radeonsi/si_sdma_copy_image.c | 14 ++++----- 6 files changed, 43 insertions(+), 43 deletions(-) diff --git a/src/amd/common/ac_debug.c b/src/amd/common/ac_debug.c index 48040f85726..22b001c2a11 100644 --- a/src/amd/common/ac_debug.c +++ b/src/amd/common/ac_debug.c @@ -735,7 +735,7 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) const uint32_t sub_op = (header >> 8) & 0xff; switch (opcode) { - case CIK_SDMA_OPCODE_NOP: { + case SDMA_OPCODE_NOP: { fprintf(f, "NOP\n"); const uint32_t count = header >> 16; @@ -745,7 +745,7 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) } break; } - case CIK_SDMA_OPCODE_CONSTANT_FILL: { + case SDMA_OPCODE_CONSTANT_FILL: { fprintf(f, "CONSTANT_FILL\n"); ac_ib_get(ib); fprintf(f, "\n"); @@ -764,7 +764,7 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) break; } - case CIK_SDMA_OPCODE_WRITE: { + case SDMA_OPCODE_WRITE: { fprintf(f, "WRITE\n"); /* VA */ @@ -783,9 +783,9 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) break; } - case CIK_SDMA_OPCODE_COPY: { + case SDMA_OPCODE_COPY: { switch (sub_op) { - case CIK_SDMA_COPY_SUB_OPCODE_LINEAR: { + case SDMA_COPY_SUB_OPCODE_LINEAR: { fprintf(f, "COPY LINEAR\n"); uint32_t copy_bytes = ac_ib_get(ib) + (ib->gfx_level >= GFX9 ? 1 : 0); @@ -803,7 +803,7 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) break; } - case CIK_SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW: { + case SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW: { fprintf(f, "COPY LINEAR_SUB_WINDOW\n"); for (unsigned i = 0; i < 12; ++i) { @@ -812,7 +812,7 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) } break; } - case CIK_SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW: { + case SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW: { fprintf(f, "COPY TILED_SUB_WINDOW %s\n", header >> 31 ? "t2l" : "l2t"); uint32_t dcc = (header >> 19) & 1; @@ -860,7 +860,7 @@ static void parse_sdma_ib(FILE *f, struct ac_ib_parser *ib) } break; } - case CIK_SDMA_COPY_SUB_OPCODE_T2T_SUB_WINDOW: { + case SDMA_COPY_SUB_OPCODE_T2T_SUB_WINDOW: { fprintf(f, "COPY T2T_SUB_WINDOW\n"); uint32_t dcc = (header >> 19) & 1; diff --git a/src/amd/common/sid.h b/src/amd/common/sid.h index 20a318a2ba2..09755147610 100644 --- a/src/amd/common/sid.h +++ b/src/amd/common/sid.h @@ -321,36 +321,36 @@ #define SI_DMA_PACKET_NOP 0xf /* CIK async DMA packets */ -#define CIK_SDMA_PACKET(op, sub_op, n) \ +#define SDMA_PACKET(op, sub_op, n) \ ((((unsigned)(n)&0xFFFF) << 16) | (((unsigned)(sub_op)&0xFF) << 8) | \ (((unsigned)(op)&0xFF) << 0)) /* CIK async DMA packet types */ -#define CIK_SDMA_OPCODE_NOP 0x0 -#define CIK_SDMA_OPCODE_COPY 0x1 -#define CIK_SDMA_COPY_SUB_OPCODE_LINEAR 0x0 -#define CIK_SDMA_COPY_SUB_OPCODE_TILED 0x1 -#define CIK_SDMA_COPY_SUB_OPCODE_SOA 0x3 -#define CIK_SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW 0x4 -#define CIK_SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW 0x5 -#define CIK_SDMA_COPY_SUB_OPCODE_T2T_SUB_WINDOW 0x6 -#define CIK_SDMA_OPCODE_WRITE 0x2 +#define SDMA_OPCODE_NOP 0x0 +#define SDMA_OPCODE_COPY 0x1 +#define SDMA_COPY_SUB_OPCODE_LINEAR 0x0 +#define SDMA_COPY_SUB_OPCODE_TILED 0x1 +#define SDMA_COPY_SUB_OPCODE_SOA 0x3 +#define SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW 0x4 +#define SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW 0x5 +#define SDMA_COPY_SUB_OPCODE_T2T_SUB_WINDOW 0x6 +#define SDMA_OPCODE_WRITE 0x2 #define SDMA_WRITE_SUB_OPCODE_LINEAR 0x0 #define SDMA_WRITE_SUB_OPCODE_TILED 0x1 -#define CIK_SDMA_OPCODE_INDIRECT_BUFFER 0x4 -#define CIK_SDMA_OPCODE_FENCE 0x5 +#define SDMA_OPCODE_INDIRECT_BUFFER 0x4 +#define SDMA_OPCODE_FENCE 0x5 #define SDMA_FENCE_MTYPE_UC 0x3 -#define CIK_SDMA_OPCODE_TRAP 0x6 -#define CIK_SDMA_OPCODE_SEMAPHORE 0x7 -#define CIK_SDMA_OPCODE_POLL_REGMEM 0x8 +#define SDMA_OPCODE_TRAP 0x6 +#define SDMA_OPCODE_SEMAPHORE 0x7 +#define SDMA_OPCODE_POLL_REGMEM 0x8 #define SDMA_POLL_MEM (1 << 31) #define SDMA_POLL_INTERVAL_160_CLK 0xa #define SDMA_POLL_RETRY_INDEFINITELY 0xfff -#define CIK_SDMA_OPCODE_CONSTANT_FILL 0xb -#define CIK_SDMA_OPCODE_TIMESTAMP 0xd +#define SDMA_OPCODE_CONSTANT_FILL 0xb +#define SDMA_OPCODE_TIMESTAMP 0xd #define SDMA_TS_SUB_OPCODE_SET_LOCAL_TIMESTAMP 0x0 #define SDMA_TS_SUB_OPCODE_GET_LOCAL_TIMESTAMP 0x1 #define SDMA_TS_SUB_OPCODE_GET_GLOBAL_TIMESTAMP 0x2 -#define CIK_SDMA_OPCODE_SRBM_WRITE 0xe +#define SDMA_OPCODE_SRBM_WRITE 0xe /* There is apparently an undocumented HW limitation that * prevents the HW from copying the last 255 bytes of (1 << 22) - 1 @@ -358,7 +358,7 @@ #define SDMA_V2_0_COPY_MAX_BYTES 0x3fff00 /* almost 4 MB*/ #define SDMA_V5_2_COPY_MAX_BYTES 0x3fffff00 /* almost 1 GB */ -#define SDMA_NOP_PAD CIK_SDMA_PACKET(CIK_SDMA_OPCODE_NOP, 0, 0) /* header-only version */ +#define SDMA_NOP_PAD SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0) /* header-only version */ enum amd_cmp_class_flags { diff --git a/src/amd/vulkan/radv_cs.h b/src/amd/vulkan/radv_cs.h index d8f525713ae..1c8114fff66 100644 --- a/src/amd/vulkan/radv_cs.h +++ b/src/amd/vulkan/radv_cs.h @@ -223,7 +223,7 @@ radv_cp_wait_mem(struct radeon_cmdbuf *cs, const enum radv_queue_family qf, cons radeon_emit(cs, mask); /* mask */ radeon_emit(cs, 4); /* poll interval */ } else if (qf == RADV_QUEUE_TRANSFER) { - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_POLL_REGMEM, 0, 0) | op << 28 | SDMA_POLL_MEM); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_POLL_REGMEM, 0, 0) | op << 28 | SDMA_POLL_MEM); radeon_emit(cs, va); radeon_emit(cs, va >> 32); radeon_emit(cs, ref); @@ -250,7 +250,7 @@ radv_cs_write_data_head(const struct radv_device *device, struct radeon_cmdbuf * /* Vulkan transfer queues don't support conditional rendering, so we can ignore predication here. * Furthermore, we can ignore the engine selection here, it is meaningless to the SDMA. */ - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0)); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0)); radeon_emit(cs, va); radeon_emit(cs, va >> 32); radeon_emit(cs, count - 1); diff --git a/src/amd/vulkan/radv_sdma.c b/src/amd/vulkan/radv_sdma.c index 9cd829c92f6..b56c5ddc503 100644 --- a/src/amd/vulkan/radv_sdma.c +++ b/src/amd/vulkan/radv_sdma.c @@ -347,7 +347,7 @@ radv_sdma_emit_nop(const struct radv_device *device, struct radeon_cmdbuf *cs) { /* SDMA NOP acts as a fence command and causes the SDMA engine to wait for pending copy operations. */ radeon_check_space(device->ws, cs, 1); - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_NOP, 0, 0)); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0)); } void @@ -380,7 +380,7 @@ radv_sdma_copy_buffer(const struct radv_device *device, struct radeon_cmdbuf *cs for (unsigned i = 0; i < ncopy; i++) { unsigned csize = size >= 4 ? MIN2(size & align, max_size_per_packet) : size; - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, CIK_SDMA_COPY_SUB_OPCODE_LINEAR, 0)); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0)); radeon_emit(cs, gfx_level >= GFX9 ? csize - 1 : csize); radeon_emit(cs, 0); /* src/dst endian swap */ radeon_emit(cs, src_va); @@ -424,8 +424,8 @@ radv_sdma_emit_copy_linear_sub_window(const struct radv_device *device, struct r ASSERTED unsigned cdw_end = radeon_check_space(device->ws, cs, 13); - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, CIK_SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW, 0) | - util_logbase2(src->bpp) << 29); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW, 0) | util_logbase2(src->bpp) + << 29); radeon_emit(cs, src->va); radeon_emit(cs, src->va >> 32); radeon_emit(cs, src_off.x | src_off.y << 16); @@ -468,8 +468,8 @@ radv_sdma_emit_copy_tiled_sub_window(const struct radv_device *device, struct ra ASSERTED unsigned cdw_end = radeon_check_space(device->ws, cs, 14 + (dcc ? 3 : 0)); - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, CIK_SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, 0) | dcc << 19 | - detile << 31 | tiled->header_dword); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, 0) | dcc << 19 | detile << 31 | + tiled->header_dword); radeon_emit(cs, tiled->va); radeon_emit(cs, tiled->va >> 32); radeon_emit(cs, tiled_off.x | tiled_off.y << 16); diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c index ecb00d98575..73f469c5bf6 100644 --- a/src/amd/vulkan/si_cmd_buffer.c +++ b/src/amd/vulkan/si_cmd_buffer.c @@ -952,7 +952,7 @@ si_cs_emit_write_event_eop(struct radeon_cmdbuf *cs, enum amd_gfx_level gfx_leve uint32_t new_fence, uint64_t gfx9_eop_bug_va) { if (qf == RADV_QUEUE_TRANSFER) { - radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_FENCE, 0, SDMA_FENCE_MTYPE_UC)); + radeon_emit(cs, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, SDMA_FENCE_MTYPE_UC)); radeon_emit(cs, va); radeon_emit(cs, va >> 32); radeon_emit(cs, new_fence); diff --git a/src/gallium/drivers/radeonsi/si_sdma_copy_image.c b/src/gallium/drivers/radeonsi/si_sdma_copy_image.c index e0e282a4d2f..65fa4fca2c9 100644 --- a/src/gallium/drivers/radeonsi/si_sdma_copy_image.c +++ b/src/gallium/drivers/radeonsi/si_sdma_copy_image.c @@ -82,8 +82,8 @@ static bool si_sdma_v4_v5_copy_texture(struct si_context *sctx, struct si_textur radeon_begin(cs); for (int i = 0; i < chunk_count; i++) { uint32_t size = MIN2(chunk_size, bytes); - radeon_emit(CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, - CIK_SDMA_COPY_SUB_OPCODE_LINEAR, + radeon_emit(SDMA_PACKET(SDMA_OPCODE_COPY, + SDMA_COPY_SUB_OPCODE_LINEAR, (tmz ? 4 : 0))); radeon_emit(size - 1); radeon_emit(0); @@ -125,8 +125,8 @@ static bool si_sdma_v4_v5_copy_texture(struct si_context *sctx, struct si_textur radeon_begin(cs); radeon_emit( - CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, - CIK_SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, + SDMA_PACKET(SDMA_OPCODE_COPY, + SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, (tmz ? 4 : 0)) | dcc << 19 | (is_v5 ? 0 : tiled->buffer.b.b.last_level) << 20 | @@ -216,7 +216,7 @@ bool cik_sdma_copy_texture(struct si_context *sctx, struct si_texture *sdst, str struct radeon_cmdbuf *cs = sctx->sdma_cs; radeon_begin(cs); - radeon_emit(CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, CIK_SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW, 0) | + radeon_emit(SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW, 0) | (util_logbase2(bpp) << 29)); radeon_emit(src_address); radeon_emit(src_address >> 32); @@ -338,8 +338,8 @@ bool cik_sdma_copy_texture(struct si_context *sctx, struct si_texture *sdst, str uint32_t direction = linear == sdst ? 1u << 31 : 0; radeon_begin(cs); - radeon_emit(CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY, - CIK_SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, 0) | + radeon_emit(SDMA_PACKET(SDMA_OPCODE_COPY, + SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, 0) | direction); radeon_emit(tiled_address); radeon_emit(tiled_address >> 32);
