X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fintel_batchbuffer.h;h=fbb51582ee0fb2a4308e6d1c883978e97c0190ea;hb=8bed1adfc144d9ae8d55ccb9b277942da8a78064;hp=1a6d1aa4946416eb93c713d69e0a8d033e3c5695;hpb=3dbba95b72262344b82fba018b7c2c1208754cd2;p=mesa.git diff --git a/src/mesa/drivers/dri/i965/intel_batchbuffer.h b/src/mesa/drivers/dri/i965/intel_batchbuffer.h index 1a6d1aa4946..fbb51582ee0 100644 --- a/src/mesa/drivers/dri/i965/intel_batchbuffer.h +++ b/src/mesa/drivers/dri/i965/intel_batchbuffer.h @@ -3,9 +3,8 @@ #include "main/mtypes.h" -#include "intel_context.h" +#include "brw_context.h" #include "intel_bufmgr.h" -#include "intel_reg.h" #ifdef __cplusplus extern "C" { @@ -19,17 +18,35 @@ extern "C" { * - Optional MI_NOOP for ensuring the batch length is qword aligned (4 bytes) * - Any state emitted by vtbl->finish_batch(): * - Gen4-5 record ending occlusion query values (4 * 4 = 16 bytes) + * - Disabling OA counters on Gen6+ (3 DWords = 12 bytes) + * - Ending MI_REPORT_PERF_COUNT on Gen5+, plus associated PIPE_CONTROLs: + * - Two sets of PIPE_CONTROLs, which become 4 PIPE_CONTROLs each on SNB, + * which are 5 DWords each ==> 2 * 4 * 5 * 4 = 160 bytes + * - 3 DWords for MI_REPORT_PERF_COUNT itself on Gen6+. ==> 12 bytes. + * On Ironlake, it's 6 DWords, but we have some slack due to the lack of + * Sandybridge PIPE_CONTROL madness. + * - CC_STATE workaround on HSW (17 * 4 = 68 bytes) + * - 10 dwords for initial mi_flush + * - 2 dwords for CC state setup + * - 5 dwords for the required pipe control at the end + * - Restoring L3 configuration: (24 dwords = 96 bytes) + * - 2*6 dwords for two PIPE_CONTROL flushes. + * - 7 dwords for L3 configuration set-up. + * - 5 dwords for L3 atomic set-up (on HSW). */ -#define BATCH_RESERVED 24 +#define BATCH_RESERVED 308 struct intel_batchbuffer; -void intel_batchbuffer_init(struct intel_context *intel); -void intel_batchbuffer_free(struct intel_context *intel); -void intel_batchbuffer_save_state(struct intel_context *intel); -void intel_batchbuffer_reset_to_saved(struct intel_context *intel); +void intel_batchbuffer_emit_render_ring_prelude(struct brw_context *brw); +void intel_batchbuffer_init(struct brw_context *brw); +void intel_batchbuffer_free(struct brw_context *brw); +void intel_batchbuffer_save_state(struct brw_context *brw); +void intel_batchbuffer_reset_to_saved(struct brw_context *brw); +void intel_batchbuffer_require_space(struct brw_context *brw, GLuint sz, + enum brw_gpu_ring ring); -int _intel_batchbuffer_flush(struct intel_context *intel, +int _intel_batchbuffer_flush(struct brw_context *brw, const char *file, int line); #define intel_batchbuffer_flush(intel) \ @@ -41,25 +58,26 @@ int _intel_batchbuffer_flush(struct intel_context *intel, * Consider it a convenience function wrapping multple * intel_buffer_dword() calls. */ -void intel_batchbuffer_data(struct intel_context *intel, - const void *data, GLuint bytes, bool is_blit); - -bool intel_batchbuffer_emit_reloc(struct intel_context *intel, - drm_intel_bo *buffer, - uint32_t read_domains, - uint32_t write_domain, - uint32_t offset); -bool intel_batchbuffer_emit_reloc_fenced(struct intel_context *intel, - drm_intel_bo *buffer, - uint32_t read_domains, - uint32_t write_domain, - uint32_t offset); -void intel_batchbuffer_emit_mi_flush(struct intel_context *intel); -void intel_emit_post_sync_nonzero_flush(struct intel_context *intel); -void intel_emit_depth_stall_flushes(struct intel_context *intel); -void gen7_emit_vs_workaround_flush(struct intel_context *intel); - -static INLINE uint32_t float_as_int(float f) +void intel_batchbuffer_data(struct brw_context *brw, + const void *data, GLuint bytes, + enum brw_gpu_ring ring); + +uint32_t intel_batchbuffer_reloc(struct brw_context *brw, + drm_intel_bo *buffer, + uint32_t offset, + uint32_t read_domains, + uint32_t write_domain, + uint32_t delta); +uint64_t intel_batchbuffer_reloc64(struct brw_context *brw, + drm_intel_bo *buffer, + uint32_t offset, + uint32_t read_domains, + uint32_t write_domain, + uint32_t delta); + +#define USED_BATCH(batch) ((uintptr_t)((batch).map_next - (batch).map)) + +static inline uint32_t float_as_int(float f) { union { float f; @@ -75,65 +93,47 @@ static INLINE uint32_t float_as_int(float f) * be passed as structs rather than dwords, but that's a little bit of * work... */ -static INLINE unsigned -intel_batchbuffer_space(struct intel_context *intel) +static inline unsigned +intel_batchbuffer_space(struct brw_context *brw) { - return (intel->batch.state_batch_offset - intel->batch.reserved_space) - - intel->batch.used*4; + return (brw->batch.state_batch_offset - brw->batch.reserved_space) + - USED_BATCH(brw->batch) * 4; } -static INLINE void -intel_batchbuffer_emit_dword(struct intel_context *intel, GLuint dword) +static inline void +intel_batchbuffer_emit_dword(struct brw_context *brw, GLuint dword) { #ifdef DEBUG - assert(intel_batchbuffer_space(intel) >= 4); + assert(intel_batchbuffer_space(brw) >= 4); #endif - intel->batch.map[intel->batch.used++] = dword; + *brw->batch.map_next++ = dword; + assert(brw->batch.ring != UNKNOWN_RING); } -static INLINE void -intel_batchbuffer_emit_float(struct intel_context *intel, float f) +static inline void +intel_batchbuffer_emit_float(struct brw_context *brw, float f) { - intel_batchbuffer_emit_dword(intel, float_as_int(f)); + intel_batchbuffer_emit_dword(brw, float_as_int(f)); } -static INLINE void -intel_batchbuffer_require_space(struct intel_context *intel, - GLuint sz, int is_blit) +static inline void +intel_batchbuffer_begin(struct brw_context *brw, int n, enum brw_gpu_ring ring) { - - if (intel->gen >= 6 && - intel->batch.is_blit != is_blit && intel->batch.used) { - intel_batchbuffer_flush(intel); - } - - intel->batch.is_blit = is_blit; + intel_batchbuffer_require_space(brw, n * 4, ring); #ifdef DEBUG - assert(sz < intel->maxBatchSize - BATCH_RESERVED); + brw->batch.emit = USED_BATCH(brw->batch); + brw->batch.total = n; #endif - if (intel_batchbuffer_space(intel) < sz) - intel_batchbuffer_flush(intel); } -static INLINE void -intel_batchbuffer_begin(struct intel_context *intel, int n, bool is_blit) +static inline void +intel_batchbuffer_advance(struct brw_context *brw) { - intel_batchbuffer_require_space(intel, n * 4, is_blit); - - intel->batch.emit = intel->batch.used; #ifdef DEBUG - intel->batch.total = n; -#endif -} - -static INLINE void -intel_batchbuffer_advance(struct intel_context *intel) -{ -#ifdef DEBUG - struct intel_batchbuffer *batch = &intel->batch; - unsigned int _n = batch->used - batch->emit; + struct intel_batchbuffer *batch = &brw->batch; + unsigned int _n = USED_BATCH(*batch) - batch->emit; assert(batch->total != 0); if (_n != batch->total) { fprintf(stderr, "ADVANCE_BATCH: %d of %d dwords emitted\n", @@ -141,30 +141,47 @@ intel_batchbuffer_advance(struct intel_context *intel) abort(); } batch->total = 0; +#else + (void) brw; #endif } -void intel_batchbuffer_cached_advance(struct intel_context *intel); - -/* Here are the crusty old macros, to be removed: - */ -#define BATCH_LOCALS - -#define BEGIN_BATCH(n) intel_batchbuffer_begin(intel, n, false) -#define BEGIN_BATCH_BLT(n) intel_batchbuffer_begin(intel, n, true) -#define OUT_BATCH(d) intel_batchbuffer_emit_dword(intel, d) -#define OUT_BATCH_F(f) intel_batchbuffer_emit_float(intel,f) -#define OUT_RELOC(buf, read_domains, write_domain, delta) do { \ - intel_batchbuffer_emit_reloc(intel, buf, \ - read_domains, write_domain, delta); \ +#define BEGIN_BATCH(n) do { \ + intel_batchbuffer_begin(brw, (n), RENDER_RING); \ + uint32_t *__map = brw->batch.map_next; \ + brw->batch.map_next += (n) + +#define BEGIN_BATCH_BLT(n) do { \ + intel_batchbuffer_begin(brw, (n), BLT_RING); \ + uint32_t *__map = brw->batch.map_next; \ + brw->batch.map_next += (n) + +#define OUT_BATCH(d) *__map++ = (d) +#define OUT_BATCH_F(f) OUT_BATCH(float_as_int((f))) + +#define OUT_RELOC(buf, read_domains, write_domain, delta) do { \ + uint32_t __offset = (__map - brw->batch.map) * 4; \ + OUT_BATCH(intel_batchbuffer_reloc(brw, (buf), __offset, \ + (read_domains), \ + (write_domain), \ + (delta))); \ } while (0) -#define OUT_RELOC_FENCED(buf, read_domains, write_domain, delta) do { \ - intel_batchbuffer_emit_reloc_fenced(intel, buf, \ - read_domains, write_domain, delta); \ + +/* Handle 48-bit address relocations for Gen8+ */ +#define OUT_RELOC64(buf, read_domains, write_domain, delta) do { \ + uint32_t __offset = (__map - brw->batch.map) * 4; \ + uint64_t reloc64 = intel_batchbuffer_reloc64(brw, (buf), __offset, \ + (read_domains), \ + (write_domain), \ + (delta)); \ + OUT_BATCH(reloc64); \ + OUT_BATCH(reloc64 >> 32); \ } while (0) -#define ADVANCE_BATCH() intel_batchbuffer_advance(intel); -#define CACHED_BATCH() intel_batchbuffer_cached_advance(intel); +#define ADVANCE_BATCH() \ + assert(__map == brw->batch.map_next); \ + intel_batchbuffer_advance(brw); \ +} while (0) #ifdef __cplusplus }