diff options
author | Marek Olšák <[email protected]> | 2018-08-02 17:40:05 -0400 |
---|---|---|
committer | Marek Olšák <[email protected]> | 2018-08-14 21:20:31 -0400 |
commit | ac72a6bd0b6fb9aad879a2d43833fb77dd3f5b7d (patch) | |
tree | 85befc25e0263ffd9f80fbd6f0f97311f37d3508 /src/gallium/drivers/radeonsi/si_query.c | |
parent | 0ca8294ecefe466005363c0d1dabd4f462225495 (diff) |
radeonsi: move internal TGSI shaders into si_shaderlib_tgsi.c
Tested-by: Dieter Nützel <[email protected]>
Reviewed-by: Samuel Pitoiset <[email protected]>
Diffstat (limited to 'src/gallium/drivers/radeonsi/si_query.c')
-rw-r--r-- | src/gallium/drivers/radeonsi/si_query.c | 219 |
1 files changed, 1 insertions, 218 deletions
diff --git a/src/gallium/drivers/radeonsi/si_query.c b/src/gallium/drivers/radeonsi/si_query.c index e29ce7150bd..f768b531139 100644 --- a/src/gallium/drivers/radeonsi/si_query.c +++ b/src/gallium/drivers/radeonsi/si_query.c @@ -30,7 +30,6 @@ #include "util/u_upload_mgr.h" #include "util/os_time.h" #include "util/u_suballoc.h" -#include "tgsi/tgsi_text.h" #include "amd/common/sid.h" #define SI_MAX_STREAMS 4 @@ -1389,222 +1388,6 @@ bool si_query_hw_get_result(struct si_context *sctx, return true; } -/* Create the compute shader that is used to collect the results. - * - * One compute grid with a single thread is launched for every query result - * buffer. The thread (optionally) reads a previous summary buffer, then - * accumulates data from the query result buffer, and writes the result either - * to a summary buffer to be consumed by the next grid invocation or to the - * user-supplied buffer. - * - * Data layout: - * - * CONST - * 0.x = end_offset - * 0.y = result_stride - * 0.z = result_count - * 0.w = bit field: - * 1: read previously accumulated values - * 2: write accumulated values for chaining - * 4: write result available - * 8: convert result to boolean (0/1) - * 16: only read one dword and use that as result - * 32: apply timestamp conversion - * 64: store full 64 bits result - * 128: store signed 32 bits result - * 256: SO_OVERFLOW mode: take the difference of two successive half-pairs - * 1.x = fence_offset - * 1.y = pair_stride - * 1.z = pair_count - * - * BUFFER[0] = query result buffer - * BUFFER[1] = previous summary buffer - * BUFFER[2] = next summary buffer or user-supplied buffer - */ -static void si_create_query_result_shader(struct si_context *sctx) -{ - /* TEMP[0].xy = accumulated result so far - * TEMP[0].z = result not available - * - * TEMP[1].x = current result index - * TEMP[1].y = current pair index - */ - static const char text_tmpl[] = - "COMP\n" - "PROPERTY CS_FIXED_BLOCK_WIDTH 1\n" - "PROPERTY CS_FIXED_BLOCK_HEIGHT 1\n" - "PROPERTY CS_FIXED_BLOCK_DEPTH 1\n" - "DCL BUFFER[0]\n" - "DCL BUFFER[1]\n" - "DCL BUFFER[2]\n" - "DCL CONST[0][0..1]\n" - "DCL TEMP[0..5]\n" - "IMM[0] UINT32 {0, 31, 2147483647, 4294967295}\n" - "IMM[1] UINT32 {1, 2, 4, 8}\n" - "IMM[2] UINT32 {16, 32, 64, 128}\n" - "IMM[3] UINT32 {1000000, 0, %u, 0}\n" /* for timestamp conversion */ - "IMM[4] UINT32 {256, 0, 0, 0}\n" - - "AND TEMP[5], CONST[0][0].wwww, IMM[2].xxxx\n" - "UIF TEMP[5]\n" - /* Check result availability. */ - "LOAD TEMP[1].x, BUFFER[0], CONST[0][1].xxxx\n" - "ISHR TEMP[0].z, TEMP[1].xxxx, IMM[0].yyyy\n" - "MOV TEMP[1], TEMP[0].zzzz\n" - "NOT TEMP[0].z, TEMP[0].zzzz\n" - - /* Load result if available. */ - "UIF TEMP[1]\n" - "LOAD TEMP[0].xy, BUFFER[0], IMM[0].xxxx\n" - "ENDIF\n" - "ELSE\n" - /* Load previously accumulated result if requested. */ - "MOV TEMP[0], IMM[0].xxxx\n" - "AND TEMP[4], CONST[0][0].wwww, IMM[1].xxxx\n" - "UIF TEMP[4]\n" - "LOAD TEMP[0].xyz, BUFFER[1], IMM[0].xxxx\n" - "ENDIF\n" - - "MOV TEMP[1].x, IMM[0].xxxx\n" - "BGNLOOP\n" - /* Break if accumulated result so far is not available. */ - "UIF TEMP[0].zzzz\n" - "BRK\n" - "ENDIF\n" - - /* Break if result_index >= result_count. */ - "USGE TEMP[5], TEMP[1].xxxx, CONST[0][0].zzzz\n" - "UIF TEMP[5]\n" - "BRK\n" - "ENDIF\n" - - /* Load fence and check result availability */ - "UMAD TEMP[5].x, TEMP[1].xxxx, CONST[0][0].yyyy, CONST[0][1].xxxx\n" - "LOAD TEMP[5].x, BUFFER[0], TEMP[5].xxxx\n" - "ISHR TEMP[0].z, TEMP[5].xxxx, IMM[0].yyyy\n" - "NOT TEMP[0].z, TEMP[0].zzzz\n" - "UIF TEMP[0].zzzz\n" - "BRK\n" - "ENDIF\n" - - "MOV TEMP[1].y, IMM[0].xxxx\n" - "BGNLOOP\n" - /* Load start and end. */ - "UMUL TEMP[5].x, TEMP[1].xxxx, CONST[0][0].yyyy\n" - "UMAD TEMP[5].x, TEMP[1].yyyy, CONST[0][1].yyyy, TEMP[5].xxxx\n" - "LOAD TEMP[2].xy, BUFFER[0], TEMP[5].xxxx\n" - - "UADD TEMP[5].y, TEMP[5].xxxx, CONST[0][0].xxxx\n" - "LOAD TEMP[3].xy, BUFFER[0], TEMP[5].yyyy\n" - - "U64ADD TEMP[4].xy, TEMP[3], -TEMP[2]\n" - - "AND TEMP[5].z, CONST[0][0].wwww, IMM[4].xxxx\n" - "UIF TEMP[5].zzzz\n" - /* Load second start/end half-pair and - * take the difference - */ - "UADD TEMP[5].xy, TEMP[5], IMM[1].wwww\n" - "LOAD TEMP[2].xy, BUFFER[0], TEMP[5].xxxx\n" - "LOAD TEMP[3].xy, BUFFER[0], TEMP[5].yyyy\n" - - "U64ADD TEMP[3].xy, TEMP[3], -TEMP[2]\n" - "U64ADD TEMP[4].xy, TEMP[4], -TEMP[3]\n" - "ENDIF\n" - - "U64ADD TEMP[0].xy, TEMP[0], TEMP[4]\n" - - /* Increment pair index */ - "UADD TEMP[1].y, TEMP[1].yyyy, IMM[1].xxxx\n" - "USGE TEMP[5], TEMP[1].yyyy, CONST[0][1].zzzz\n" - "UIF TEMP[5]\n" - "BRK\n" - "ENDIF\n" - "ENDLOOP\n" - - /* Increment result index */ - "UADD TEMP[1].x, TEMP[1].xxxx, IMM[1].xxxx\n" - "ENDLOOP\n" - "ENDIF\n" - - "AND TEMP[4], CONST[0][0].wwww, IMM[1].yyyy\n" - "UIF TEMP[4]\n" - /* Store accumulated data for chaining. */ - "STORE BUFFER[2].xyz, IMM[0].xxxx, TEMP[0]\n" - "ELSE\n" - "AND TEMP[4], CONST[0][0].wwww, IMM[1].zzzz\n" - "UIF TEMP[4]\n" - /* Store result availability. */ - "NOT TEMP[0].z, TEMP[0]\n" - "AND TEMP[0].z, TEMP[0].zzzz, IMM[1].xxxx\n" - "STORE BUFFER[2].x, IMM[0].xxxx, TEMP[0].zzzz\n" - - "AND TEMP[4], CONST[0][0].wwww, IMM[2].zzzz\n" - "UIF TEMP[4]\n" - "STORE BUFFER[2].y, IMM[0].xxxx, IMM[0].xxxx\n" - "ENDIF\n" - "ELSE\n" - /* Store result if it is available. */ - "NOT TEMP[4], TEMP[0].zzzz\n" - "UIF TEMP[4]\n" - /* Apply timestamp conversion */ - "AND TEMP[4], CONST[0][0].wwww, IMM[2].yyyy\n" - "UIF TEMP[4]\n" - "U64MUL TEMP[0].xy, TEMP[0], IMM[3].xyxy\n" - "U64DIV TEMP[0].xy, TEMP[0], IMM[3].zwzw\n" - "ENDIF\n" - - /* Convert to boolean */ - "AND TEMP[4], CONST[0][0].wwww, IMM[1].wwww\n" - "UIF TEMP[4]\n" - "U64SNE TEMP[0].x, TEMP[0].xyxy, IMM[4].zwzw\n" - "AND TEMP[0].x, TEMP[0].xxxx, IMM[1].xxxx\n" - "MOV TEMP[0].y, IMM[0].xxxx\n" - "ENDIF\n" - - "AND TEMP[4], CONST[0][0].wwww, IMM[2].zzzz\n" - "UIF TEMP[4]\n" - "STORE BUFFER[2].xy, IMM[0].xxxx, TEMP[0].xyxy\n" - "ELSE\n" - /* Clamping */ - "UIF TEMP[0].yyyy\n" - "MOV TEMP[0].x, IMM[0].wwww\n" - "ENDIF\n" - - "AND TEMP[4], CONST[0][0].wwww, IMM[2].wwww\n" - "UIF TEMP[4]\n" - "UMIN TEMP[0].x, TEMP[0].xxxx, IMM[0].zzzz\n" - "ENDIF\n" - - "STORE BUFFER[2].x, IMM[0].xxxx, TEMP[0].xxxx\n" - "ENDIF\n" - "ENDIF\n" - "ENDIF\n" - "ENDIF\n" - - "END\n"; - - char text[sizeof(text_tmpl) + 32]; - struct tgsi_token tokens[1024]; - struct pipe_compute_state state = {}; - - /* Hard code the frequency into the shader so that the backend can - * use the full range of optimizations for divide-by-constant. - */ - snprintf(text, sizeof(text), text_tmpl, - sctx->screen->info.clock_crystal_freq); - - if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) { - assert(false); - return; - } - - state.ir_type = PIPE_SHADER_IR_TGSI; - state.prog = tokens; - - sctx->query_result_shader = sctx->b.create_compute_state(&sctx->b, &state); -} - static void si_restore_qbo_state(struct si_context *sctx, struct si_qbo_state *st) { @@ -1647,7 +1430,7 @@ static void si_query_hw_get_result_resource(struct si_context *sctx, } consts; if (!sctx->query_result_shader) { - si_create_query_result_shader(sctx); + sctx->query_result_shader = si_create_query_result_cs(sctx); if (!sctx->query_result_shader) return; } |