diff options
Diffstat (limited to 'src/mesa/drivers/dri/i965')
31 files changed, 1441 insertions, 324 deletions
diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index 00a42111da..128afb5686 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -43,6 +43,7 @@ DRIVER_SOURCES = \ brw_clip_util.c \ brw_context.c \ brw_curbe.c \ + brw_disasm.c \ brw_draw.c \ brw_draw_upload.c \ brw_eu.c \ diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c index 54d30a3f42..20a927cf38 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.c +++ b/src/mesa/drivers/dri/i965/brw_clip.c @@ -166,7 +166,8 @@ static void upload_clip_prog(struct brw_context *brw) /* _NEW_POLYGON */ if (key.primitive == GL_TRIANGLES) { - if (ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK) + if (ctx->Polygon.CullFlag && + ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK) key.clip_mode = BRW_CLIPMODE_REJECT_ALL; else { GLuint fill_front = CLIP_CULL; diff --git a/src/mesa/drivers/dri/i965/brw_clip.h b/src/mesa/drivers/dri/i965/brw_clip.h index 12e8548df1..957df441ab 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.h +++ b/src/mesa/drivers/dri/i965/brw_clip.h @@ -100,6 +100,8 @@ struct brw_clip_compile { struct brw_reg fixed_planes; struct brw_reg plane_equation; + + struct brw_reg ff_sync; } reg; /* 3 different ways of expressing vertex size: @@ -173,4 +175,5 @@ struct brw_reg get_tmp( struct brw_clip_compile *c ); void brw_clip_project_position(struct brw_clip_compile *c, struct brw_reg pos ); void brw_clip_ff_sync(struct brw_clip_compile *c); +void brw_clip_init_ff_sync(struct brw_clip_compile *c); #endif diff --git a/src/mesa/drivers/dri/i965/brw_clip_line.c b/src/mesa/drivers/dri/i965/brw_clip_line.c index 9abd0642aa..048ca620fa 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_line.c +++ b/src/mesa/drivers/dri/i965/brw_clip_line.c @@ -85,6 +85,10 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) i++; } + if (c->need_ff_sync) { + c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); + i++; + } c->first_tmp = i; c->last_tmp = i; @@ -246,8 +250,6 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1); brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0)); - if (c->need_ff_sync) - brw_clip_ff_sync(c); not_culled = brw_IF(p, BRW_EXECUTE_1); { brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, GL_FALSE); @@ -265,6 +267,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) void brw_emit_line_clip( struct brw_clip_compile *c ) { brw_clip_line_alloc_regs(c); + brw_clip_init_ff_sync(c); if (c->key.do_flat_shading) brw_clip_copy_colors(c, 0, 1); diff --git a/src/mesa/drivers/dri/i965/brw_clip_point.c b/src/mesa/drivers/dri/i965/brw_clip_point.c index 9738299168..8458f61c5a 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_point.c +++ b/src/mesa/drivers/dri/i965/brw_clip_point.c @@ -50,7 +50,7 @@ void brw_emit_point_clip( struct brw_clip_compile *c ) /* Send an empty message to kill the thread: */ brw_clip_tri_alloc_regs(c, 0); - if (c->need_ff_sync) - brw_clip_ff_sync(c); + brw_clip_init_ff_sync(c); + brw_clip_kill_thread(c); } diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c b/src/mesa/drivers/dri/i965/brw_clip_tri.c index 4c2d655fb1..0efd77225e 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_tri.c +++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c @@ -119,6 +119,11 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, i++; } + if (c->need_ff_sync) { + c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); + i++; + } + c->first_tmp = i; c->last_tmp = i; @@ -563,6 +568,7 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); brw_clip_tri_init_vertices(c); brw_clip_init_clipmask(c); + brw_clip_init_ff_sync(c); /* if -ve rhw workaround bit is set, do cliptest */ @@ -589,8 +595,6 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) else maybe_do_clip_tri(c); - if (c->need_ff_sync) - brw_clip_ff_sync(c); brw_clip_tri_emit_polygon(c); /* Send an empty message to kill the thread: diff --git a/src/mesa/drivers/dri/i965/brw_clip_unfilled.c b/src/mesa/drivers/dri/i965/brw_clip_unfilled.c index 26950383c1..ad1bfa435f 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_unfilled.c +++ b/src/mesa/drivers/dri/i965/brw_clip_unfilled.c @@ -453,6 +453,7 @@ void brw_emit_unfilled_clip( struct brw_clip_compile *c ) brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); brw_clip_tri_init_vertices(c); + brw_clip_init_ff_sync(c); assert(c->offset[VERT_RESULT_EDGE]); @@ -496,8 +497,6 @@ void brw_emit_unfilled_clip( struct brw_clip_compile *c ) } brw_ENDIF(p, do_clip); - if (c->need_ff_sync) - brw_clip_ff_sync(c); emit_unfilled_primitives(c); brw_clip_kill_thread(c); } diff --git a/src/mesa/drivers/dri/i965/brw_clip_util.c b/src/mesa/drivers/dri/i965/brw_clip_util.c index e09efc07ed..5a73abdfee 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_util.c +++ b/src/mesa/drivers/dri/i965/brw_clip_util.c @@ -213,6 +213,8 @@ void brw_clip_emit_vue(struct brw_clip_compile *c, struct brw_compile *p = &c->func; GLuint start = c->last_mrf; + brw_clip_ff_sync(c); + assert(!(allocate && eot)); /* Cycle through mrf regs - probably futile as we have to wait for @@ -263,6 +265,7 @@ void brw_clip_kill_thread(struct brw_clip_compile *c) { struct brw_compile *p = &c->func; + brw_clip_ff_sync(c); /* Send an empty message to kill the thread and release any * allocated urb entry: */ @@ -356,17 +359,38 @@ void brw_clip_init_clipmask( struct brw_clip_compile *c ) void brw_clip_ff_sync(struct brw_clip_compile *c) { + if (c->need_ff_sync) { + struct brw_compile *p = &c->func; + struct brw_instruction *need_ff_sync; + + brw_set_conditionalmod(p, BRW_CONDITIONAL_Z); + brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1)); + need_ff_sync = brw_IF(p, BRW_EXECUTE_1); + { + brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1)); + brw_ff_sync(p, + c->reg.R0, + 0, + c->reg.R0, + 1, + 1, /* used */ + 1, /* msg length */ + 1, /* response length */ + 0, /* eot */ + 1, /* write compelete */ + 0, /* urb offset */ + BRW_URB_SWIZZLE_NONE); + } + brw_ENDIF(p, need_ff_sync); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } +} + +void brw_clip_init_ff_sync(struct brw_clip_compile *c) +{ + if (c->need_ff_sync) { struct brw_compile *p = &c->func; - brw_ff_sync(p, - c->reg.R0, - 0, - c->reg.R0, - 1, - 1, /* used */ - 1, /* msg length */ - 1, /* response length */ - 0, /* eot */ - 1, /* write compelete */ - 0, /* urb offset */ - BRW_URB_SWIZZLE_NONE); + + brw_MOV(p, c->reg.ff_sync, brw_imm_ud(0)); + } } diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index 57ddf75413..847c44ed83 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -143,6 +143,7 @@ struct brw_context; #define BRW_NEW_DEPTH_BUFFER 0x20000 #define BRW_NEW_NR_WM_SURFACES 0x40000 #define BRW_NEW_NR_VS_SURFACES 0x80000 +#define BRW_NEW_INDEX_BUFFER 0x100000 struct brw_state_flags { /** State update flags signalled by mesa internals */ @@ -438,9 +439,13 @@ struct brw_query_object { unsigned int count; }; + +/** + * brw_context is derived from intel_context. + */ struct brw_context { - struct intel_context intel; + struct intel_context intel; /**< base class, must be first field */ GLuint primitive; GLboolean emit_state_always; @@ -475,6 +480,9 @@ struct brw_context struct { struct brw_vertex_element inputs[VERT_ATTRIB_MAX]; + struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; + GLuint nr_enabled; + #define BRW_NR_UPLOAD_BUFS 17 #define BRW_UPLOAD_INIT_SIZE (128*1024) @@ -498,8 +506,15 @@ struct brw_context */ const struct _mesa_index_buffer *ib; + /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */ dri_bo *bo; unsigned int offset; + unsigned int size; + /* Offset to index buffer index to use in CMD_3D_PRIM so that we can + * avoid re-uploading the IB packet over and over if we're actually + * referencing the same index buffer. + */ + unsigned int start_vertex_offset; } ib; /* Active vertex program: @@ -706,6 +721,8 @@ void brw_upload_urb_fence(struct brw_context *brw); */ void brw_upload_cs_urb_state(struct brw_context *brw); +/* brw_disasm.c */ +int brw_disasm (FILE *file, struct brw_instruction *inst); /*====================================================================== * Inline conversion functions. These are better-typed than the diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h index d166250b4f..78d457ad2b 100644 --- a/src/mesa/drivers/dri/i965/brw_defines.h +++ b/src/mesa/drivers/dri/i965/brw_defines.h @@ -471,8 +471,9 @@ #define BRW_CONDITIONAL_GE 4 #define BRW_CONDITIONAL_L 5 #define BRW_CONDITIONAL_LE 6 -#define BRW_CONDITIONAL_C 7 +#define BRW_CONDITIONAL_R 7 #define BRW_CONDITIONAL_O 8 +#define BRW_CONDITIONAL_U 9 #define BRW_DEBUG_NONE 0 #define BRW_DEBUG_BREAKPOINT 1 @@ -512,6 +513,7 @@ #define BRW_OPCODE_RSL 11 #define BRW_OPCODE_ASR 12 #define BRW_OPCODE_CMP 16 +#define BRW_OPCODE_CMPN 17 #define BRW_OPCODE_JMPI 32 #define BRW_OPCODE_IF 34 #define BRW_OPCODE_IFF 35 diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c new file mode 100644 index 0000000000..9fef230507 --- /dev/null +++ b/src/mesa/drivers/dri/i965/brw_disasm.c @@ -0,0 +1,903 @@ +/* + * Copyright © 2008 Keith Packard + * + * Permission to use, copy, modify, distribute, and sell this software and its + * documentation for any purpose is hereby granted without fee, provided that + * the above copyright notice appear in all copies and that both that copyright + * notice and this permission notice appear in supporting documentation, and + * that the name of the copyright holders not be used in advertising or + * publicity pertaining to distribution of the software without specific, + * written prior permission. The copyright holders make no representations + * about the suitability of this software for any purpose. It is provided "as + * is" without express or implied warranty. + * + * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, + * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO + * EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY SPECIAL, INDIRECT OR + * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, + * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER + * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE + * OF THIS SOFTWARE. + */ + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <getopt.h> +#include <unistd.h> +#include <stdarg.h> + +#include "main/mtypes.h" + +#include "brw_context.h" +#include "brw_defines.h" + +struct { + char *name; + int nsrc; + int ndst; +} opcode[128] = { + [BRW_OPCODE_MOV] = { .name = "mov", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_FRC] = { .name = "frc", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDU] = { .name = "rndu", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDD] = { .name = "rndd", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDE] = { .name = "rnde", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDZ] = { .name = "rndz", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_NOT] = { .name = "not", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_LZD] = { .name = "lzd", .nsrc = 1, .ndst = 1 }, + + [BRW_OPCODE_MUL] = { .name = "mul", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_MAC] = { .name = "mac", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_MACH] = { .name = "mach", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_LINE] = { .name = "line", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SAD2] = { .name = "sad2", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SADA2] = { .name = "sada2", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DP4] = { .name = "dp4", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DPH] = { .name = "dph", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DP3] = { .name = "dp3", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DP2] = { .name = "dp2", .nsrc = 2, .ndst = 1 }, + + [BRW_OPCODE_AVG] = { .name = "avg", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_ADD] = { .name = "add", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SEL] = { .name = "sel", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_AND] = { .name = "and", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_OR] = { .name = "or", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_XOR] = { .name = "xor", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SHR] = { .name = "shr", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SHL] = { .name = "shl", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_ASR] = { .name = "asr", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_CMP] = { .name = "cmp", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_CMPN] = { .name = "cmpn", .nsrc = 2, .ndst = 1 }, + + [BRW_OPCODE_SEND] = { .name = "send", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_NOP] = { .name = "nop", .nsrc = 0, .ndst = 0 }, + [BRW_OPCODE_JMPI] = { .name = "jmpi", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_IF] = { .name = "if", .nsrc = 2, .ndst = 0 }, + [BRW_OPCODE_IFF] = { .name = "iff", .nsrc = 1, .ndst = 01 }, + [BRW_OPCODE_WHILE] = { .name = "while", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_ELSE] = { .name = "else", .nsrc = 2, .ndst = 0 }, + [BRW_OPCODE_BREAK] = { .name = "break", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_CONTINUE] = { .name = "cont", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_HALT] = { .name = "halt", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_MSAVE] = { .name = "msave", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_PUSH] = { .name = "push", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_MRESTORE] = { .name = "mrest", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_POP] = { .name = "pop", .nsrc = 2, .ndst = 0 }, + [BRW_OPCODE_WAIT] = { .name = "wait", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_DO] = { .name = "do", .nsrc = 0, .ndst = 0 }, + [BRW_OPCODE_ENDIF] = { .name = "endif", .nsrc = 2, .ndst = 0 }, +}; + +char *conditional_modifier[16] = { + [BRW_CONDITIONAL_NONE] = "", + [BRW_CONDITIONAL_Z] = ".e", + [BRW_CONDITIONAL_NZ] = ".ne", + [BRW_CONDITIONAL_G] = ".g", + [BRW_CONDITIONAL_GE] = ".ge", + [BRW_CONDITIONAL_L] = ".l", + [BRW_CONDITIONAL_LE] = ".le", + [BRW_CONDITIONAL_R] = ".r", + [BRW_CONDITIONAL_O] = ".o", + [BRW_CONDITIONAL_U] = ".u", +}; + +char *negate[2] = { + [0] = "", + [1] = "-", +}; + +char *_abs[2] = { + [0] = "", + [1] = "(abs)", +}; + +char *vert_stride[16] = { + [0] = "0", + [1] = "1", + [2] = "2", + [3] = "4", + [4] = "8", + [5] = "16", + [6] = "32", + [15] = "VxH", +}; + +char *width[8] = { + [0] = "1", + [1] = "2", + [2] = "4", + [3] = "8", + [4] = "16", +}; + +char *horiz_stride[4] = { + [0] = "0", + [1] = "1", + [2] = "2", + [3] = "4" +}; + +char *chan_sel[4] = { + [0] = "x", + [1] = "y", + [2] = "z", + [3] = "w", +}; + +char *dest_condmod[16] = { +}; + +char *debug_ctrl[2] = { + [0] = "", + [1] = ".breakpoint" +}; + +char *saturate[2] = { + [0] = "", + [1] = ".sat" +}; + +char *exec_size[8] = { + [0] = "1", + [1] = "2", + [2] = "4", + [3] = "8", + [4] = "16", + [5] = "32" +}; + +char *pred_inv[2] = { + [0] = "+", + [1] = "-" +}; + +char *pred_ctrl_align16[16] = { + [1] = "", + [2] = ".x", + [3] = ".y", + [4] = ".z", + [5] = ".w", + [6] = ".any4h", + [7] = ".all4h", +}; + +char *pred_ctrl_align1[16] = { + [1] = "", + [2] = ".anyv", + [3] = ".allv", + [4] = ".any2h", + [5] = ".all2h", + [6] = ".any4h", + [7] = ".all4h", + [8] = ".any8h", + [9] = ".all8h", + [10] = ".any16h", + [11] = ".all16h", +}; + +char *thread_ctrl[4] = { + [0] = "", + [2] = "switch" +}; + +char *compr_ctrl[4] = { + [0] = "", + [1] = "sechalf", + [2] = "compr", +}; + +char *dep_ctrl[4] = { + [0] = "", + [1] = "NoDDClr", + [2] = "NoDDChk", + [3] = "NoDDClr,NoDDChk", +}; + +char *mask_ctrl[4] = { + [0] = "", + [1] = "nomask", +}; + +char *access_mode[2] = { + [0] = "align1", + [1] = "align16", +}; + +char *reg_encoding[8] = { + [0] = "UD", + [1] = "D", + [2] = "UW", + [3] = "W", + [4] = "UB", + [5] = "B", + [7] = "F" +}; + +char *imm_encoding[8] = { + [0] = "UD", + [1] = "D", + [2] = "UW", + [3] = "W", + [5] = "VF", + [5] = "V", + [7] = "F" +}; + +char *reg_file[4] = { + [0] = "A", + [1] = "g", + [2] = "m", + [3] = "imm", +}; + +char *writemask[16] = { + [0x0] = ".", + [0x1] = ".x", + [0x2] = ".y", + [0x3] = ".xy", + [0x4] = ".z", + [0x5] = ".xz", + [0x6] = ".yz", + [0x7] = ".xyz", + [0x8] = ".w", + [0x9] = ".xw", + [0xa] = ".yw", + [0xb] = ".xyw", + [0xc] = ".zw", + [0xd] = ".xzw", + [0xe] = ".yzw", + [0xf] = "", +}; + +char *end_of_thread[2] = { + [0] = "", + [1] = "EOT" +}; + +char *target_function[16] = { + [BRW_MESSAGE_TARGET_NULL] = "null", + [BRW_MESSAGE_TARGET_MATH] = "math", + [BRW_MESSAGE_TARGET_SAMPLER] = "sampler", + [BRW_MESSAGE_TARGET_GATEWAY] = "gateway", + [BRW_MESSAGE_TARGET_DATAPORT_READ] = "read", + [BRW_MESSAGE_TARGET_DATAPORT_WRITE] = "write", + [BRW_MESSAGE_TARGET_URB] = "urb", + [BRW_MESSAGE_TARGET_THREAD_SPAWNER] = "thread_spawner" +}; + +char *math_function[16] = { + [BRW_MATH_FUNCTION_INV] = "inv", + [BRW_MATH_FUNCTION_LOG] = "log", + [BRW_MATH_FUNCTION_EXP] = "exp", + [BRW_MATH_FUNCTION_SQRT] = "sqrt", + [BRW_MATH_FUNCTION_RSQ] = "rsq", + [BRW_MATH_FUNCTION_SIN] = "sin", + [BRW_MATH_FUNCTION_COS] = "cos", + [BRW_MATH_FUNCTION_SINCOS] = "sincos", + [BRW_MATH_FUNCTION_TAN] = "tan", + [BRW_MATH_FUNCTION_POW] = "pow", + [BRW_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER] = "intdivmod", + [BRW_MATH_FUNCTION_INT_DIV_QUOTIENT] = "intmod", + [BRW_MATH_FUNCTION_INT_DIV_REMAINDER] = "intdiv", +}; + +char *math_saturate[2] = { + [0] = "", + [1] = "sat" +}; + +char *math_signed[2] = { + [0] = "", + [1] = "signed" +}; + +char *math_scalar[2] = { + [0] = "", + [1] = "scalar" +}; + +char *math_precision[2] = { + [0] = "", + [1] = "partial_precision" +}; + +char *urb_swizzle[4] = { + [BRW_URB_SWIZZLE_NONE] = "", + [BRW_URB_SWIZZLE_INTERLEAVE] = "interleave", + [BRW_URB_SWIZZLE_TRANSPOSE] = "transpose", +}; + +char *urb_allocate[2] = { + [0] = "", + [1] = "allocate" +}; + +char *urb_used[2] = { + [0] = "", + [1] = "used" +}; + +char *urb_complete[2] = { + [0] = "", + [1] = "complete" +}; + +char *sampler_target_format[4] = { + [0] = "F", + [2] = "UD", + [3] = "D" +}; + + +static int column; + +static int string (FILE *file, char *string) +{ + fputs (string, file); + column += strlen (string); + return 0; +} + +static int format (FILE *f, char *format, ...) +{ + char buf[1024]; + va_list args; + va_start (args, format); + + vsnprintf (buf, sizeof (buf) - 1, format, args); + string (f, buf); + return 0; +} + +static int newline (FILE *f) +{ + putc ('\n', f); + column = 0; + return 0; +} + +static int pad (FILE *f, int c) +{ + do + string (f, " "); + while (column < c); + return 0; +} + +static int control (FILE *file, char *name, char *ctrl[], GLuint id, int *space) +{ + if (!ctrl[id]) { + fprintf (file, "*** invalid %s value %d ", + name, id); + return 1; + } + if (ctrl[id][0]) + { + if (space && *space) + string (file, " "); + string (file, ctrl[id]); + if (space) + *space = 1; + } + return 0; +} + +static int print_opcode (FILE *file, int id) +{ + if (!opcode[id].name) { + format (file, "*** invalid opcode value %d ", id); + return 1; + } + string (file, opcode[id].name); + return 0; +} + +static int reg (FILE *file, GLuint _reg_file, GLuint _reg_nr) +{ + int err = 0; + if (_reg_file == BRW_ARCHITECTURE_REGISTER_FILE) { + switch (_reg_nr & 0xf0) { + case BRW_ARF_NULL: + string (file, "null"); + return -1; + case BRW_ARF_ADDRESS: + format (file, "a%d", _reg_nr & 0x0f); + break; + case BRW_ARF_ACCUMULATOR: + format (file, "acc%d", _reg_nr & 0x0f); + break; + case BRW_ARF_MASK: + format (file, "mask%d", _reg_nr & 0x0f); + break; + case BRW_ARF_MASK_STACK: + format (file, "msd%d", _reg_nr & 0x0f); + break; + case BRW_ARF_STATE: + format (file, "sr%d", _reg_nr & 0x0f); + break; + case BRW_ARF_CONTROL: + format (file, "cr%d", _reg_nr & 0x0f); + break; + case BRW_ARF_NOTIFICATION_COUNT: + format (file, "n%d", _reg_nr & 0x0f); + break; + case BRW_ARF_IP: + string (file, "ip"); + return -1; + break; + default: + format (file, "ARF%d", _reg_nr); + break; + } + } else { + err |= control (file, "src reg file", reg_file, _reg_file, NULL); + format (file, "%d", _reg_nr); + } + return err; +} + +static int dest (FILE *file, struct brw_instruction *inst) +{ + int err = 0; + + if (inst->header.access_mode == BRW_ALIGN_1) + { + if (inst->bits1.da1.dest_address_mode == BRW_ADDRESS_DIRECT) + { + err |= reg (file, inst->bits1.da1.dest_reg_file, inst->bits1.da1.dest_reg_nr); + if (err == -1) + return 0; + if (inst->bits1.da1.dest_subreg_nr) + format (file, ".%d", inst->bits1.da1.dest_subreg_nr); + format (file, "<%d>", inst->bits1.da1.dest_horiz_stride); + err |= control (file, "dest reg encoding", reg_encoding, inst->bits1.da1.dest_reg_type, NULL); + } + else + { + string (file, "g[a0"); + if (inst->bits1.ia1.dest_subreg_nr) + format (file, ".%d", inst->bits1.ia1.dest_subreg_nr); + if (inst->bits1.ia1.dest_indirect_offset) + format (file, " %d", inst->bits1.ia1.dest_indirect_offset); + string (file, "]"); + format (file, "<%d>", inst->bits1.ia1.dest_horiz_stride); + err |= control (file, "dest reg encoding", reg_encoding, inst->bits1.ia1.dest_reg_type, NULL); + } + } + else + { + if (inst->bits1.da16.dest_address_mode == BRW_ADDRESS_DIRECT) + { + err |= reg (file, inst->bits1.da16.dest_reg_file, inst->bits1.da16.dest_reg_nr); + if (err == -1) + return 0; + if (inst->bits1.da16.dest_subreg_nr) + format (file, ".%d", inst->bits1.da16.dest_subreg_nr); + string (file, "<1>"); + err |= control (file, "writemask", writemask, inst->bits1.da16.dest_writemask, NULL); + err |= control (file, "dest reg encoding", reg_encoding, inst->bits1.da16.dest_reg_type, NULL); + } + else + { + err = 1; + string (file, "Indirect align16 address mode not supported"); + } + } + + return 0; +} + +static int src_align1_region (FILE *file, + GLuint _vert_stride, GLuint _width, GLuint _horiz_stride) +{ + int err = 0; + string (file, "<"); + err |= control (file, "vert stride", vert_stride, _vert_stride, NULL); + string (file, ","); + err |= control (file, "width", width, _width, NULL); + string (file, ","); + err |= control (file, "horiz_stride", horiz_stride, _horiz_stride, NULL); + string (file, ">"); + return err; +} + +static int src_da1 (FILE *file, GLuint type, GLuint _reg_file, + GLuint _vert_stride, GLuint _width, GLuint _horiz_stride, + GLuint reg_num, GLuint sub_reg_num, GLuint __abs, GLuint _negate) +{ + int err = 0; + err |= control (file, "negate", negate, _negate, NULL); + err |= control (file, "abs", _abs, __abs, NULL); + + err |= reg (file, _reg_file, reg_num); + if (err == -1) + return 0; + if (sub_reg_num) + format (file, ".%d", sub_reg_num); + src_align1_region (file, _vert_stride, _width, _horiz_stride); + err |= control (file, "src reg encoding", reg_encoding, type, NULL); + return err; +} + +static int src_ia1 (FILE *file, + GLuint type, + GLuint _reg_file, + GLint _addr_imm, + GLuint _addr_subreg_nr, + GLuint _negate, + GLuint __abs, + GLuint _addr_mode, + GLuint _horiz_stride, + GLuint _width, + GLuint _vert_stride) +{ + int err = 0; + err |= control (file, "negate", negate, _negate, NULL); + err |= control (file, "abs", _abs, __abs, NULL); + + string (file, "g[a0"); + if (_addr_subreg_nr) + format (file, ".%d", _addr_subreg_nr); + if (_addr_imm) + format (file, " %d", _addr_imm); + string (file, "]"); + src_align1_region (file, _vert_stride, _width, _horiz_stride); + err |= control (file, "src reg encoding", reg_encoding, type, NULL); + return err; +} + +static int src_da16 (FILE *file, + GLuint _reg_type, + GLuint _reg_file, + GLuint _vert_stride, + GLuint _reg_nr, + GLuint _subreg_nr, + GLuint __abs, + GLuint _negate, + GLuint swz_x, + GLuint swz_y, + GLuint swz_z, + GLuint swz_w) +{ + int err = 0; + err |= control (file, "negate", negate, _negate, NULL); + err |= control (file, "abs", _abs, __abs, NULL); + + err |= reg (file, _reg_file, _reg_nr); + if (err == -1) + return 0; + if (_subreg_nr) + format (file, ".%d", _subreg_nr); + string (file, "<"); + err |= control (file, "vert stride", vert_stride, _vert_stride, NULL); + string (file, ",1,1>"); + err |= control (file, "src da16 reg type", reg_encoding, _reg_type, NULL); + /* + * Three kinds of swizzle display: + * identity - nothing printed + * 1->all - print the single channel + * 1->1 - print the mapping + */ + if (swz_x == BRW_CHANNEL_X && + swz_y == BRW_CHANNEL_Y && + swz_z == BRW_CHANNEL_Z && + swz_w == BRW_CHANNEL_W) + { + ; + } + else if (swz_x == swz_y && swz_x == swz_z && swz_x == swz_w) + { + string (file, "."); + err |= control (file, "channel select", chan_sel, swz_x, NULL); + } + else + { + string (file, "."); + err |= control (file, "channel select", chan_sel, swz_x, NULL); + err |= control (file, "channel select", chan_sel, swz_y, NULL); + err |= control (file, "channel select", chan_sel, swz_z, NULL); + err |= control (file, "channel select", chan_sel, swz_w, NULL); + } + return err; +} + + +static int imm (FILE *file, GLuint type, struct brw_instruction *inst) { + switch (type) { + case BRW_REGISTER_TYPE_UD: + format (file, "0x%08xUD", inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_D: + format (file, "%dD", inst->bits3.d); + break; + case BRW_REGISTER_TYPE_UW: + format (file, "0x%04xUW", (uint16_t) inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_W: + format (file, "%dW", (int16_t) inst->bits3.d); + break; + case BRW_REGISTER_TYPE_UB: + format (file, "0x%02xUB", (int8_t) inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_VF: + format (file, "Vector Float"); + break; + case BRW_REGISTER_TYPE_V: + format (file, "0x%08xV", inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_F: + format (file, "%-gF", inst->bits3.f); + } + return 0; +} + +static int src0 (FILE *file, struct brw_instruction *inst) +{ + if (inst->bits1.da1.src0_reg_file == BRW_IMMEDIATE_VALUE) + return imm (file, inst->bits1.da1.src0_reg_type, + inst); + else if (inst->header.access_mode == BRW_ALIGN_1) + { + if (inst->bits2.da1.src0_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da1 (file, + inst->bits1.da1.src0_reg_type, + inst->bits1.da1.src0_reg_file, + inst->bits2.da1.src0_vert_stride, + inst->bits2.da1.src0_width, + inst->bits2.da1.src0_horiz_stride, + inst->bits2.da1.src0_reg_nr, + inst->bits2.da1.src0_subreg_nr, + inst->bits2.da1.src0_abs, + inst->bits2.da1.src0_negate); + } + else + { + return src_ia1 (file, + inst->bits1.ia1.src0_reg_type, + inst->bits1.ia1.src0_reg_file, + inst->bits2.ia1.src0_indirect_offset, + inst->bits2.ia1.src0_subreg_nr, + inst->bits2.ia1.src0_negate, + inst->bits2.ia1.src0_abs, + inst->bits2.ia1.src0_address_mode, + inst->bits2.ia1.src0_horiz_stride, + inst->bits2.ia1.src0_width, + inst->bits2.ia1.src0_vert_stride); + } + } + else + { + if (inst->bits2.da16.src0_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da16 (file, + inst->bits1.da16.src0_reg_type, + inst->bits1.da16.src0_reg_file, + inst->bits2.da16.src0_vert_stride, + inst->bits2.da16.src0_reg_nr, + inst->bits2.da16.src0_subreg_nr, + inst->bits2.da16.src0_abs, + inst->bits2.da16.src0_negate, + inst->bits2.da16.src0_swz_x, + inst->bits2.da16.src0_swz_y, + inst->bits2.da16.src0_swz_z, + inst->bits2.da16.src0_swz_w); + } + else + { + string (file, "Indirect align16 address mode not supported"); + return 1; + } + } +} + +static int src1 (FILE *file, struct brw_instruction *inst) +{ + if (inst->bits1.da1.src1_reg_file == BRW_IMMEDIATE_VALUE) + return imm (file, inst->bits1.da1.src1_reg_type, + inst); + else if (inst->header.access_mode == BRW_ALIGN_1) + { + if (inst->bits3.da1.src1_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da1 (file, + inst->bits1.da1.src1_reg_type, + inst->bits1.da1.src1_reg_file, + inst->bits3.da1.src1_vert_stride, + inst->bits3.da1.src1_width, + inst->bits3.da1.src1_horiz_stride, + inst->bits3.da1.src1_reg_nr, + inst->bits3.da1.src1_subreg_nr, + inst->bits3.da1.src1_abs, + inst->bits3.da1.src1_negate); + } + else + { + return src_ia1 (file, + inst->bits1.ia1.src1_reg_type, + inst->bits1.ia1.src1_reg_file, + inst->bits3.ia1.src1_indirect_offset, + inst->bits3.ia1.src1_subreg_nr, + inst->bits3.ia1.src1_negate, + inst->bits3.ia1.src1_abs, + inst->bits3.ia1.src1_address_mode, + inst->bits3.ia1.src1_horiz_stride, + inst->bits3.ia1.src1_width, + inst->bits3.ia1.src1_vert_stride); + } + } + else + { + if (inst->bits3.da16.src1_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da16 (file, + inst->bits1.da16.src1_reg_type, + inst->bits1.da16.src1_reg_file, + inst->bits3.da16.src1_vert_stride, + inst->bits3.da16.src1_reg_nr, + inst->bits3.da16.src1_subreg_nr, + inst->bits3.da16.src1_abs, + inst->bits3.da16.src1_negate, + inst->bits3.da16.src1_swz_x, + inst->bits3.da16.src1_swz_y, + inst->bits3.da16.src1_swz_z, + inst->bits3.da16.src1_swz_w); + } + else + { + string (file, "Indirect align16 address mode not supported"); + return 1; + } + } +} + +int brw_disasm (FILE *file, struct brw_instruction *inst) +{ + int err = 0; + int space = 0; + + if (inst->header.predicate_control) { + string (file, "("); + err |= control (file, "predicate inverse", pred_inv, inst->header.predicate_inverse, NULL); + string (file, "f0"); + if (inst->bits2.da1.flag_reg_nr) + format (file, ".%d", inst->bits2.da1.flag_reg_nr); + if (inst->header.access_mode == BRW_ALIGN_1) + err |= control (file, "predicate control align1", pred_ctrl_align1, + inst->header.predicate_control, NULL); + else + err |= control (file, "predicate control align16", pred_ctrl_align16, + inst->header.predicate_control, NULL); + string (file, ") "); + } + + err |= print_opcode (file, inst->header.opcode); + err |= control (file, "saturate", saturate, inst->header.saturate, NULL); + err |= control (file, "debug control", debug_ctrl, inst->header.debug_control, NULL); + + if (inst->header.opcode != BRW_OPCODE_SEND) + err |= control (file, "conditional modifier", conditional_modifier, + inst->header.destreg__conditionalmod, NULL); + + if (inst->header.opcode != BRW_OPCODE_NOP) { + string (file, "("); + err |= control (file, "execution size", exec_size, inst->header.execution_size, NULL); + string (file, ")"); + } + + if (inst->header.opcode == BRW_OPCODE_SEND) + format (file, " %d", inst->header.destreg__conditionalmod); + + if (opcode[inst->header.opcode].ndst > 0) { + pad (file, 16); + err |= dest (file, inst); + } + if (opcode[inst->header.opcode].nsrc > 0) { + pad (file, 32); + err |= src0 (file, inst); + } + if (opcode[inst->header.opcode].nsrc > 1) { + pad (file, 48); + err |= src1 (file, inst); + } + + if (inst->header.opcode == BRW_OPCODE_SEND) { + newline (file); + pad (file, 16); + space = 0; + err |= control (file, "target function", target_function, + inst->bits3.generic.msg_target, &space); + switch (inst->bits3.generic.msg_target) { + case BRW_MESSAGE_TARGET_MATH: + err |= control (file, "math function", math_function, + inst->bits3.math.function, &space); + err |= control (file, "math saturate", math_saturate, + inst->bits3.math.saturate, &space); + err |= control (file, "math signed", math_signed, + inst->bits3.math.int_type, &space); + err |= control (file, "math scalar", math_scalar, + inst->bits3.math.data_type, &space); + err |= control (file, "math precision", math_precision, + inst->bits3.math.precision, &space); + break; + case BRW_MESSAGE_TARGET_SAMPLER: + format (file, " (%d, %d, ", + inst->bits3.sampler.binding_table_index, + inst->bits3.sampler.sampler); + err |= control (file, "sampler target format", sampler_target_format, + inst->bits3.sampler.return_format, NULL); + string (file, ")"); + break; + case BRW_MESSAGE_TARGET_DATAPORT_WRITE: + format (file, " (%d, %d, %d, %d)", + inst->bits3.dp_write.binding_table_index, + (inst->bits3.dp_write.pixel_scoreboard_clear << 3) | + inst->bits3.dp_write.msg_control, + inst->bits3.dp_write.msg_type, + inst->bits3.dp_write.send_commit_msg); + break; + case BRW_MESSAGE_TARGET_URB: + format (file, " %d", inst->bits3.urb.offset); + space = 1; + err |= control (file, "urb swizzle", urb_swizzle, + inst->bits3.urb.swizzle_control, &space); + err |= control (file, "urb allocate", urb_allocate, + inst->bits3.urb.allocate, &space); + err |= control (file, "urb used", urb_used, + inst->bits3.urb.used, &space); + err |= control (file, "urb complete", urb_complete, + inst->bits3.urb.complete, &space); + break; + case BRW_MESSAGE_TARGET_THREAD_SPAWNER: + break; + default: + format (file, "unsupported target %d", inst->bits3.generic.msg_target); + break; + } + if (space) + string (file, " "); + format (file, "mlen %d", + inst->bits3.generic.msg_length); + format (file, " rlen %d", + inst->bits3.generic.response_length); + } + pad (file, 64); + if (inst->header.opcode != BRW_OPCODE_NOP) { + string (file, "{"); + space = 1; + err |= control(file, "access mode", access_mode, inst->header.access_mode, &space); + err |= control (file, "mask control", mask_ctrl, inst->header.mask_control, &space); + err |= control (file, "dependency control", dep_ctrl, inst->header.dependency_control, &space); + err |= control (file, "compression control", compr_ctrl, inst->header.compression_control, &space); + err |= control (file, "thread control", thread_ctrl, inst->header.thread_control, &space); + if (inst->header.opcode == BRW_OPCODE_SEND) + err |= control (file, "end of thread", end_of_thread, + inst->bits3.generic.end_of_thread, &space); + if (space) + string (file, " "); + string (file, "}"); + } + string (file, ";"); + newline (file); + return err; +} diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c index 5152c3f3a5..682094ff13 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.c +++ b/src/mesa/drivers/dri/i965/brw_draw.c @@ -141,6 +141,8 @@ static void brw_emit_prim(struct brw_context *brw, prim_packet.verts_per_instance = trim(prim->mode, prim->count); prim_packet.start_vert_location = prim->start; + if (prim->indexed) + prim_packet.start_vert_location += brw->ib.start_vertex_offset; prim_packet.instance_count = 1; prim_packet.start_instance_location = 0; prim_packet.base_vert_location = 0; @@ -422,54 +424,31 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, return retval; } -static GLboolean brw_need_rebase( GLcontext *ctx, - const struct gl_client_array *arrays[], - const struct _mesa_index_buffer *ib, - GLuint min_index ) -{ - if (min_index == 0) - return GL_FALSE; - - if (ib) { - if (!vbo_all_varyings_in_vbos(arrays)) - return GL_TRUE; - else - return GL_FALSE; - } - else { - /* Hmm. This isn't quite what I wanted. BRW can actually - * handle the mixed case well enough that we shouldn't need to - * rebase. However, it's probably not very common, nor hugely - * expensive to do it this way: - */ - if (!vbo_all_varyings_in_vbos(arrays)) - return GL_TRUE; - else - return GL_FALSE; - } -} - - void brw_draw_prims( GLcontext *ctx, const struct gl_client_array *arrays[], const struct _mesa_prim *prim, GLuint nr_prims, const struct _mesa_index_buffer *ib, + GLboolean index_bounds_valid, GLuint min_index, GLuint max_index ) { GLboolean retval; - /* Decide if we want to rebase. If so we end up recursing once - * only into this function. - */ - if (brw_need_rebase( ctx, arrays, ib, min_index )) { - vbo_rebase_prims( ctx, arrays, - prim, nr_prims, - ib, min_index, max_index, - brw_draw_prims ); - - return; + if (!vbo_all_varyings_in_vbos(arrays)) { + if (!index_bounds_valid) + vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index); + + /* Decide if we want to rebase. If so we end up recursing once + * only into this function. + */ + if (min_index != 0) { + vbo_rebase_prims(ctx, arrays, + prim, nr_prims, + ib, min_index, max_index, + brw_draw_prims ); + return; + } } /* Make a first attempt at drawing: diff --git a/src/mesa/drivers/dri/i965/brw_draw.h b/src/mesa/drivers/dri/i965/brw_draw.h index 9aebbdb1b8..2a14db217f 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.h +++ b/src/mesa/drivers/dri/i965/brw_draw.h @@ -39,6 +39,7 @@ void brw_draw_prims( GLcontext *ctx, const struct _mesa_prim *prims, GLuint nr_prims, const struct _mesa_index_buffer *ib, + GLboolean index_bounds_valid, GLuint min_index, GLuint max_index ); diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index c29f1dd5c0..d49fb0fd95 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -350,9 +350,6 @@ static void brw_prepare_vertices(struct brw_context *brw) unsigned int min_index = brw->vb.min_index; unsigned int max_index = brw->vb.max_index; - struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; - GLuint nr_enabled = 0; - struct brw_vertex_element *upload[VERT_ATTRIB_MAX]; GLuint nr_uploads = 0; @@ -362,12 +359,13 @@ static void brw_prepare_vertices(struct brw_context *brw) _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); /* Accumulate the list of enabled arrays. */ + brw->vb.nr_enabled = 0; while (vs_inputs) { GLuint i = _mesa_ffsll(vs_inputs) - 1; struct brw_vertex_element *input = &brw->vb.inputs[i]; vs_inputs &= ~(1 << i); - enabled[nr_enabled++] = input; + brw->vb.enabled[brw->vb.nr_enabled++] = input; } /* XXX: In the rare cases where this happens we fallback all @@ -376,16 +374,15 @@ static void brw_prepare_vertices(struct brw_context *brw) * cases with > 17 vertex attributes enabled, so it probably * isn't an issue at this point. */ - if (nr_enabled >= BRW_VEP_MAX) { + if (brw->vb.nr_enabled >= BRW_VEP_MAX) { intel->Fallback = 1; return; } - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; input->element_size = get_size(input->glarray->Type) * input->glarray->Size; - input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1; if (input->glarray->BufferObj->Name != 0) { struct intel_buffer_object *intel_buffer = @@ -398,7 +395,23 @@ static void brw_prepare_vertices(struct brw_context *brw) dri_bo_reference(input->bo); input->offset = (unsigned long)input->glarray->Ptr; input->stride = input->glarray->StrideB; + input->count = input->glarray->_MaxElement; + + /* This is a common place to reach if the user mistakenly supplies + * a pointer in place of a VBO offset. If we just let it go through, + * we may end up dereferencing a pointer beyond the bounds of the + * GTT. We would hope that the VBO's max_index would save us, but + * Mesa appears to hand us min/max values not clipped to the + * array object's _MaxElement, and _MaxElement frequently appears + * to be wrong anyway. + * + * The VBO spec allows application termination in this case, and it's + * probably a service to the poor programmer to do so rather than + * trying to just not render. + */ + assert(input->offset < input->bo->size); } else { + input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1; if (input->bo != NULL) { /* Already-uploaded vertex data is present from a previous * prepare_vertices, but we had to re-validate state due to @@ -466,8 +479,8 @@ static void brw_prepare_vertices(struct brw_context *brw) brw_prepare_query_begin(brw); - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; brw_add_validated_bo(brw, input->bo); } @@ -477,34 +490,44 @@ static void brw_emit_vertices(struct brw_context *brw) { GLcontext *ctx = &brw->intel.ctx; struct intel_context *intel = intel_context(ctx); - GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; - struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; GLuint i; - GLuint nr_enabled = 0; - /* Accumulate the list of enabled arrays. */ - while (vs_inputs) { - i = _mesa_ffsll(vs_inputs) - 1; - struct brw_vertex_element *input = &brw->vb.inputs[i]; + brw_emit_query_begin(brw); - vs_inputs &= ~(1 << i); - enabled[nr_enabled++] = input; + /* If the VS doesn't read any inputs (calculating vertex position from + * a state variable for some reason, for example), emit a single pad + * VERTEX_ELEMENT struct and bail. + * + * The stale VB state stays in place, but they don't do anything unless + * a VE loads from them. + */ + if (brw->vb.nr_enabled == 0) { + BEGIN_BATCH(3, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1); + OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) | + BRW_VE0_VALID | + (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) | + (0 << BRW_VE0_SRC_OFFSET_SHIFT)); + OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) | + (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) | + (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) | + (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT)); + ADVANCE_BATCH(); + return; } - brw_emit_query_begin(brw); - /* Now emit VB and VEP state packets. * * This still defines a hardware VB for each input, even if they * are interleaved or from the same VBO. TBD if this makes a * performance difference. */ - BEGIN_BATCH(1 + nr_enabled * 4, IGNORE_CLIPRECTS); + BEGIN_BATCH(1 + brw->vb.nr_enabled * 4, IGNORE_CLIPRECTS); OUT_BATCH((CMD_VERTEX_BUFFER << 16) | - ((1 + nr_enabled * 4) - 2)); + ((1 + brw->vb.nr_enabled * 4) - 2)); - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) | BRW_VB0_ACCESS_VERTEXDATA | @@ -524,15 +547,15 @@ static void brw_emit_vertices(struct brw_context *brw) input->offset + input->element_size); } } else - OUT_BATCH(brw->vb.max_index); + OUT_BATCH(input->stride ? input->count : 0); OUT_BATCH(0); /* Instance data step rate */ } ADVANCE_BATCH(); - BEGIN_BATCH(1 + nr_enabled * 2, IGNORE_CLIPRECTS); - OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + nr_enabled * 2) - 2)); - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + BEGIN_BATCH(1 + brw->vb.nr_enabled * 2, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2)); + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; uint32_t format = get_surface_type(input->glarray->Type, input->glarray->Size, input->glarray->Format, @@ -589,17 +612,20 @@ static void brw_prepare_indices(struct brw_context *brw) dri_bo *bo = NULL; struct gl_buffer_object *bufferobj; GLuint offset; + GLuint ib_type_size; if (index_buffer == NULL) return; - ib_size = get_size(index_buffer->type) * index_buffer->count; + ib_type_size = get_size(index_buffer->type); + ib_size = ib_type_size * index_buffer->count; bufferobj = index_buffer->obj;; /* Turn into a proper VBO: */ if (!bufferobj->Name) { - + brw->ib.start_vertex_offset = 0; + /* Get new bufferobj, offset: */ get_space(brw, ib_size, &bo, &offset); @@ -615,6 +641,7 @@ static void brw_prepare_indices(struct brw_context *brw) } } else { offset = (GLuint) (unsigned long) index_buffer->ptr; + brw->ib.start_vertex_offset = 0; /* If the index buffer isn't aligned to its element size, we have to * rebase it into a temporary. @@ -635,39 +662,62 @@ static void brw_prepare_indices(struct brw_context *brw) bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj), INTEL_READ); dri_bo_reference(bo); + + /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading + * the index buffer state when we're just moving the start index + * of our drawing. + */ + brw->ib.start_vertex_offset = offset / ib_type_size; + offset = 0; + ib_size = bo->size; } } - dri_bo_unreference(brw->ib.bo); - brw->ib.bo = bo; - brw->ib.offset = offset; + if (brw->ib.bo != bo || + brw->ib.offset != offset || + brw->ib.size != ib_size) + { + drm_intel_bo_unreference(brw->ib.bo); + brw->ib.bo = bo; + brw->ib.offset = offset; + brw->ib.size = ib_size; + + brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER; + } else { + drm_intel_bo_unreference(bo); + } brw_add_validated_bo(brw, brw->ib.bo); } -static void brw_emit_indices(struct brw_context *brw) +const struct brw_tracked_state brw_indices = { + .dirty = { + .mesa = 0, + .brw = BRW_NEW_INDICES, + .cache = 0, + }, + .prepare = brw_prepare_indices, +}; + +static void brw_emit_index_buffer(struct brw_context *brw) { struct intel_context *intel = &brw->intel; const struct _mesa_index_buffer *index_buffer = brw->ib.ib; - GLuint ib_size; if (index_buffer == NULL) return; - ib_size = get_size(index_buffer->type) * index_buffer->count - 1; - /* Emit the indexbuffer packet: */ { struct brw_indexbuffer ib; memset(&ib, 0, sizeof(ib)); - + ib.header.bits.opcode = CMD_INDEX_BUFFER; ib.header.bits.length = sizeof(ib)/4 - 2; ib.header.bits.index_format = get_index_type(index_buffer->type); ib.header.bits.cut_index_enable = 0; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); OUT_BATCH( ib.header.dword ); @@ -676,18 +726,17 @@ static void brw_emit_indices(struct brw_context *brw) brw->ib.offset); OUT_RELOC(brw->ib.bo, I915_GEM_DOMAIN_VERTEX, 0, - brw->ib.offset + ib_size); + brw->ib.offset + brw->ib.size); OUT_BATCH( 0 ); ADVANCE_BATCH(); } } -const struct brw_tracked_state brw_indices = { +const struct brw_tracked_state brw_index_buffer = { .dirty = { .mesa = 0, - .brw = BRW_NEW_BATCH | BRW_NEW_INDICES, + .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER, .cache = 0, }, - .prepare = brw_prepare_indices, - .emit = brw_emit_indices, + .emit = brw_emit_index_buffer, }; diff --git a/src/mesa/drivers/dri/i965/brw_eu.c b/src/mesa/drivers/dri/i965/brw_eu.c index c53efba599..1df561386e 100644 --- a/src/mesa/drivers/dri/i965/brw_eu.c +++ b/src/mesa/drivers/dri/i965/brw_eu.c @@ -62,7 +62,7 @@ void brw_set_predicate_control( struct brw_compile *p, GLuint pc ) void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional ) { - p->current->header.destreg__conditonalmod = conditional; + p->current->header.destreg__conditionalmod = conditional; } void brw_set_access_mode( struct brw_compile *p, GLuint access_mode ) diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index 2412014248..241cdc33f8 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -481,8 +481,8 @@ static struct brw_instruction *next_insn( struct brw_compile *p, /* Reset this one-shot flag: */ - if (p->current->header.destreg__conditonalmod) { - p->current->header.destreg__conditonalmod = 0; + if (p->current->header.destreg__conditionalmod) { + p->current->header.destreg__conditionalmod = 0; p->current->header.predicate_control = BRW_PREDICATE_NORMAL; } @@ -679,7 +679,7 @@ struct brw_instruction *brw_ELSE(struct brw_compile *p, assert(if_insn->header.opcode == BRW_OPCODE_IF); if_insn->bits3.if_else.jump_count = br * (insn - if_insn); - if_insn->bits3.if_else.pop_count = 1; + if_insn->bits3.if_else.pop_count = 0; if_insn->bits3.if_else.pad0 = 0; } @@ -871,7 +871,7 @@ void brw_CMP(struct brw_compile *p, { struct brw_instruction *insn = next_insn(p, BRW_OPCODE_CMP); - insn->header.destreg__conditonalmod = conditional; + insn->header.destreg__conditionalmod = conditional; brw_set_dest(insn, dest); brw_set_src0(insn, src0); brw_set_src1(insn, src1); @@ -915,7 +915,7 @@ void brw_math( struct brw_compile *p, * instructions. */ insn->header.predicate_control = 0; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src); @@ -952,7 +952,7 @@ void brw_math_16( struct brw_compile *p, brw_set_compression_control(p, BRW_COMPRESSION_NONE); insn = next_insn(p, BRW_OPCODE_SEND); - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src); @@ -969,7 +969,7 @@ void brw_math_16( struct brw_compile *p, */ insn = next_insn(p, BRW_OPCODE_SEND); insn->header.compression_control = BRW_COMPRESSION_2NDHALF; - insn->header.destreg__conditonalmod = msg_reg_nr+1; + insn->header.destreg__conditionalmod = msg_reg_nr+1; brw_set_dest(insn, offset(dest,1)); brw_set_src0(insn, src); @@ -1016,7 +1016,7 @@ void brw_dp_WRITE_16( struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src); @@ -1062,7 +1062,7 @@ void brw_dp_READ_16( struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); /* UW? */ brw_set_src0(insn, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW)); @@ -1116,7 +1116,7 @@ void brw_dp_READ_4( struct brw_compile *p, insn->header.predicate_control = BRW_PREDICATE_NONE; insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; insn->header.mask_control = BRW_MASK_DISABLE; /* cast dest to a uword[8] vector */ @@ -1190,7 +1190,7 @@ void brw_dp_READ_4_vs(struct brw_compile *p, insn->header.predicate_control = BRW_PREDICATE_NONE; insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; insn->header.mask_control = BRW_MASK_DISABLE; /*insn->header.access_mode = BRW_ALIGN_16;*/ @@ -1224,7 +1224,7 @@ void brw_fb_WRITE(struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src0); @@ -1322,7 +1322,7 @@ void brw_SAMPLE(struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src0); @@ -1375,7 +1375,7 @@ void brw_urb_WRITE(struct brw_compile *p, brw_set_src0(insn, src0); brw_set_src1(insn, brw_imm_d(0)); - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_urb_message(p->brw, insn, @@ -1410,7 +1410,7 @@ void brw_ff_sync(struct brw_compile *p, brw_set_src0(insn, src0); brw_set_src1(insn, brw_imm_d(0)); - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_ff_sync_message(p->brw, insn, diff --git a/src/mesa/drivers/dri/i965/brw_gs_emit.c b/src/mesa/drivers/dri/i965/brw_gs_emit.c index 980eac7646..a9b2aa2eac 100644 --- a/src/mesa/drivers/dri/i965/brw_gs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_gs_emit.c @@ -101,7 +101,7 @@ static void brw_gs_emit_vue(struct brw_gs_compile *c, BRW_URB_SWIZZLE_NONE); } -void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim) +static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim) { struct brw_compile *p = &c->func; brw_MOV(p, get_element_ud(c->reg.R0, 1), brw_imm_ud(num_prim)); diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c index b5f6371c82..bc0f076073 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_state.c +++ b/src/mesa/drivers/dri/i965/brw_sf_state.c @@ -113,7 +113,7 @@ struct brw_sf_unit_key { unsigned int nr_urb_entries, urb_size, sfsize; - GLenum front_face, cull_face; + GLenum front_face, cull_face, provoking_vertex; unsigned scissor:1; unsigned line_smooth:1; unsigned point_sprite:1; @@ -153,6 +153,9 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key) key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize); key->point_attenuated = ctx->Point._Attenuated; + /* _NEW_LIGHT */ + key->provoking_vertex = ctx->Light.ProvokingVertex; + key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0; } @@ -284,9 +287,15 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons: */ - sf.sf7.trifan_pv = 2; - sf.sf7.linestrip_pv = 1; - sf.sf7.tristrip_pv = 2; + if (key->provoking_vertex == GL_LAST_VERTEX_CONVENTION) { + sf.sf7.trifan_pv = 2; + sf.sf7.linestrip_pv = 1; + sf.sf7.tristrip_pv = 2; + } else { + sf.sf7.trifan_pv = 1; + sf.sf7.linestrip_pv = 0; + sf.sf7.tristrip_pv = 0; + } sf.sf7.line_last_pixel_enable = 0; /* Set bias for OpenGL rasterization rules: @@ -300,6 +309,9 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, &sf, sizeof(sf), NULL, NULL); + /* STATE_PREFETCH command description describes this state as being + * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain. + */ /* Emit SF program relocation */ dri_bo_emit_reloc(bo, I915_GEM_DOMAIN_INSTRUCTION, 0, @@ -340,6 +352,7 @@ static void upload_sf_unit( struct brw_context *brw ) const struct brw_tracked_state brw_sf_unit = { .dirty = { .mesa = (_NEW_POLYGON | + _NEW_LIGHT | _NEW_LINE | _NEW_POINT | _NEW_SCISSOR | diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index bf9f6cae55..78572356a3 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -92,6 +92,7 @@ const struct brw_tracked_state brw_clear_batch_cache; const struct brw_tracked_state brw_drawing_rect; const struct brw_tracked_state brw_indices; const struct brw_tracked_state brw_vertices; +const struct brw_tracked_state brw_index_buffer; /** * Use same key for WM and VS surfaces. diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index 38d9dd8991..95d42d2dcc 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -94,6 +94,7 @@ const struct brw_tracked_state *atoms[] = &brw_drawing_rect, &brw_indices, + &brw_index_buffer, &brw_vertices, &brw_constant_buffer @@ -208,6 +209,7 @@ static struct dirty_bit_map brw_bits[] = { DEFINE_BIT(BRW_NEW_PSP), DEFINE_BIT(BRW_NEW_FENCE), DEFINE_BIT(BRW_NEW_INDICES), + DEFINE_BIT(BRW_NEW_INDEX_BUFFER), DEFINE_BIT(BRW_NEW_VERTICES), DEFINE_BIT(BRW_NEW_BATCH), DEFINE_BIT(BRW_NEW_DEPTH_BUFFER), diff --git a/src/mesa/drivers/dri/i965/brw_structs.h b/src/mesa/drivers/dri/i965/brw_structs.h index 8ba7eb27b3..a6de09207b 100644 --- a/src/mesa/drivers/dri/i965/brw_structs.h +++ b/src/mesa/drivers/dri/i965/brw_structs.h @@ -1200,7 +1200,7 @@ struct brw_instruction GLuint predicate_control:4; GLuint predicate_inverse:1; GLuint execution_size:3; - GLuint destreg__conditonalmod:4; /* destreg - send, conditionalmod - others */ + GLuint destreg__conditionalmod:4; /* destreg - send, conditionalmod - others */ GLuint pad0:2; GLuint debug_control:1; GLuint saturate:1; @@ -1228,7 +1228,9 @@ struct brw_instruction GLuint dest_reg_type:3; GLuint src0_reg_file:2; GLuint src0_reg_type:3; - GLuint pad:6; + GLuint src1_reg_file:2; /* 0x00000c00 */ + GLuint src1_reg_type:3; /* 0x00007000 */ + GLuint pad:1; GLint dest_indirect_offset:10; /* offset against the deref'd address reg */ GLuint dest_subreg_nr:3; /* subnr for the address reg a0.x */ GLuint dest_horiz_stride:2; @@ -1243,7 +1245,7 @@ struct brw_instruction GLuint src0_reg_type:3; GLuint src1_reg_file:2; GLuint src1_reg_type:3; - GLuint pad0:1; + GLuint pad:1; GLuint dest_writemask:4; GLuint dest_subreg_nr:1; GLuint dest_reg_nr:8; @@ -1348,7 +1350,7 @@ struct brw_instruction GLuint src1_reg_nr:8; GLuint src1_abs:1; GLuint src1_negate:1; - GLuint pad:1; + GLuint src1_address_mode:1; GLuint src1_horiz_stride:2; GLuint src1_width:3; GLuint src1_vert_stride:4; @@ -1363,7 +1365,7 @@ struct brw_instruction GLuint src1_reg_nr:8; GLuint src1_abs:1; GLuint src1_negate:1; - GLuint pad0:1; + GLuint src1_address_mode:1; GLuint src1_swz_z:2; GLuint src1_swz_w:2; GLuint pad1:1; @@ -1377,7 +1379,7 @@ struct brw_instruction GLuint src1_subreg_nr:3; GLuint src1_abs:1; GLuint src1_negate:1; - GLuint pad0:1; + GLuint src1_address_mode:1; GLuint src1_horiz_stride:2; GLuint src1_width:3; GLuint src1_vert_stride:4; @@ -1565,6 +1567,7 @@ struct brw_instruction GLint d; GLuint ud; + float f; } bits3; }; diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c index 7f9b253534..1d2e953eb1 100644 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c @@ -65,11 +65,6 @@ GLboolean brw_miptree_layout(struct intel_context *intel, if (mt->compressed) { mt->pitch = ALIGN(mt->width0, align_w); - qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * mt->pitch * mt->cpp; - mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * 6; - } else { - qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * mt->pitch * mt->cpp; - mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * 6; } if (mt->first_level != mt->last_level) { @@ -90,6 +85,14 @@ GLboolean brw_miptree_layout(struct intel_context *intel, mt->pitch = intel_miptree_pitch_align(intel, mt, tiling, mt->pitch); + if (mt->compressed) { + qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * mt->pitch * mt->cpp; + mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * 6; + } else { + qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * mt->pitch * mt->cpp; + mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * 6; + } + for (level = mt->first_level; level <= mt->last_level; level++) { GLuint img_height; GLuint nr_images = 6; diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c index 514f15d5e3..83167b9258 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c @@ -68,6 +68,7 @@ static void release_tmps( struct brw_vs_compile *c ) static void brw_vs_alloc_regs( struct brw_vs_compile *c ) { GLuint i, reg = 0, mrf; + int attributes_in_vue; /* Determine whether to use a real constant buffer or use a block * of GRF registers for constants. The later is faster but only @@ -128,6 +129,11 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) reg++; } } + /* If there are no inputs, we'll still be reading one attribute's worth + * because it's required -- see urb_read_length setting. + */ + if (c->nr_inputs == 0) + reg++; /* Allocate outputs. The non-position outputs go straight into message regs. */ @@ -220,11 +226,22 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) * vertex urb, so is half the amount: */ c->prog_data.urb_read_length = (c->nr_inputs + 1) / 2; + /* Setting this field to 0 leads to undefined behavior according to the + * the VS_STATE docs. Our VUEs will always have at least one attribute + * sitting in them, even if it's padding. + */ + if (c->prog_data.urb_read_length == 0) + c->prog_data.urb_read_length = 1; + + /* The VS VUEs are shared by VF (outputting our inputs) and VS, so size + * them to fit the biggest thing they need to. + */ + attributes_in_vue = MAX2(c->nr_outputs, c->nr_inputs); if (BRW_IS_IGDNG(c->func.brw)) - c->prog_data.urb_entry_size = (c->nr_outputs + 6 + 3) / 4; + c->prog_data.urb_entry_size = (attributes_in_vue + 6 + 3) / 4; else - c->prog_data.urb_entry_size = (c->nr_outputs + 2 + 3) / 4; + c->prog_data.urb_entry_size = (attributes_in_vue + 2 + 3) / 4; c->prog_data.total_grf = reg; @@ -1245,9 +1262,30 @@ post_vs_emit( struct brw_vs_compile *c, /* patch up the END code to jump past subroutines, etc */ offset = last_inst - end_inst; - brw_set_src1(end_inst, brw_imm_d(offset * 16)); + if (offset > 1) { + brw_set_src1(end_inst, brw_imm_d(offset * 16)); + } else { + end_inst->header.opcode = BRW_OPCODE_NOP; + } } +static uint32_t +get_predicate(uint32_t swizzle) +{ + switch (swizzle) { + case SWIZZLE_XXXX: + return BRW_PREDICATE_ALIGN16_REPLICATE_X; + case SWIZZLE_YYYY: + return BRW_PREDICATE_ALIGN16_REPLICATE_Y; + case SWIZZLE_ZZZZ: + return BRW_PREDICATE_ALIGN16_REPLICATE_Z; + case SWIZZLE_WWWW: + return BRW_PREDICATE_ALIGN16_REPLICATE_W; + default: + _mesa_problem(NULL, "Unexpected predicate: 0x%08x\n", swizzle); + return BRW_PREDICATE_NORMAL; + } +} /* Emit the vertex program instructions here. */ @@ -1266,7 +1304,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) GLuint file; if (INTEL_DEBUG & DEBUG_VS) { - _mesa_printf("vs-emit:\n"); + _mesa_printf("vs-mesa:\n"); _mesa_print_program(&c->vp->program.Base); _mesa_printf("\n"); } @@ -1453,7 +1491,10 @@ void brw_vs_emit(struct brw_vs_compile *c ) break; case OPCODE_IF: assert(if_depth < MAX_IF_DEPTH); - if_inst[if_depth++] = brw_IF(p, BRW_EXECUTE_8); + if_inst[if_depth] = brw_IF(p, BRW_EXECUTE_8); + if_inst[if_depth]->header.predicate_control = + get_predicate(inst->DstReg.CondSwizzle); + if_depth++; break; case OPCODE_ELSE: if_inst[if_depth-1] = brw_ELSE(p, if_inst[if_depth-1]); @@ -1541,6 +1582,19 @@ void brw_vs_emit(struct brw_vs_compile *c ) "unknown"); } + /* Set the predication update on the last instruction of the native + * instruction sequence. + * + * This would be problematic if it was set on a math instruction, + * but that shouldn't be the case with the current GLSL compiler. + */ + if (inst->CondUpdate) { + struct brw_instruction *hw_insn = &p->store[p->nr_insn - 1]; + + assert(hw_insn->header.destreg__conditionalmod == 0); + hw_insn->header.destreg__conditionalmod = BRW_CONDITIONAL_NZ; + } + if ((inst->DstReg.File == PROGRAM_OUTPUT) && (inst->DstReg.Index != VERT_RESULT_HPOS) && c->output_regs[inst->DstReg.Index].used_in_src) { @@ -1578,4 +1632,13 @@ void brw_vs_emit(struct brw_vs_compile *c ) emit_vertex_write(c); post_vs_emit(c, end_inst, last_inst); + + if (INTEL_DEBUG & DEBUG_VS) { + int i; + + _mesa_printf("vs-native:\n"); + for (i = 0; i < p->nr_insn; i++) + brw_disasm(stderr, &p->store[i]); + _mesa_printf("\n"); + } } diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 14e05be4f6..2292de94c4 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -41,13 +41,13 @@ GLuint brw_wm_nr_args( GLuint opcode ) { switch (opcode) { case WM_FRONTFACING: - return 0; case WM_PIXELXY: + return 0; case WM_CINTERP: case WM_WPOSXY: + case WM_DELTAXY: return 1; case WM_LINTERP: - case WM_DELTAXY: case WM_PIXELW: return 2; case WM_FB_WRITE: @@ -171,9 +171,11 @@ static void do_wm_prog( struct brw_context *brw, * differently from "simple" shaders. */ if (fp->isGLSL) { + c->dispatch_width = 8; brw_wm_glsl_emit(brw, c); } else { + c->dispatch_width = 16; brw_wm_non_glsl_emit(brw, c); } @@ -202,6 +204,7 @@ static void brw_wm_populate_key( struct brw_context *brw, /* BRW_NEW_FRAGMENT_PROGRAM */ const struct brw_fragment_program *fp = (struct brw_fragment_program *)brw->fragment_program; + GLboolean uses_depth = (fp->program.Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) != 0; GLuint lookup = 0; GLuint line_aa; GLuint i; @@ -263,6 +266,7 @@ static void brw_wm_populate_key( struct brw_context *brw, brw_wm_lookup_iz(line_aa, lookup, + uses_depth, key); diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index ba497432c6..ae98b5492d 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -260,6 +260,7 @@ struct brw_wm_compile { GLuint tmp_index; GLuint tmp_max; GLuint subroutines[BRW_WM_MAX_SUBROUTINE]; + GLuint dispatch_width; /** we may need up to 3 constants per instruction (if use_const_buffer) */ struct { @@ -292,6 +293,7 @@ void brw_wm_print_program( struct brw_wm_compile *c, void brw_wm_lookup_iz( GLuint line_aa, GLuint lookup, + GLboolean ps_uses_depth, struct brw_wm_prog_key *key ); GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp); diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index 9f82916c02..b3cf524c63 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -65,8 +65,7 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) static void emit_pixel_xy(struct brw_compile *p, const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) + GLuint mask) { struct brw_reg r1 = brw_vec1_grf(1, 0); struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); @@ -98,8 +97,7 @@ static void emit_pixel_xy(struct brw_compile *p, static void emit_delta_xy(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1) + const struct brw_reg *arg0) { struct brw_reg r1 = brw_vec1_grf(1, 0); @@ -545,16 +543,18 @@ static void emit_dp3( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert((mask & WRITEMASK_XYZW) == WRITEMASK_X); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); brw_set_saturate(p, (mask & SATURATE) ? 1 : 0); - brw_MAC(p, dst[0], arg0[2], arg1[2]); + brw_MAC(p, dst[dst_chan], arg0[2], arg1[2]); brw_set_saturate(p, 0); } @@ -565,17 +565,19 @@ static void emit_dp4( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert((mask & WRITEMASK_XYZW) == WRITEMASK_X); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); brw_MAC(p, brw_null_reg(), arg0[2], arg1[2]); brw_set_saturate(p, (mask & SATURATE) ? 1 : 0); - brw_MAC(p, dst[0], arg0[3], arg1[3]); + brw_MAC(p, dst[dst_chan], arg0[3], arg1[3]); brw_set_saturate(p, 0); } @@ -632,18 +634,19 @@ static void emit_math1( struct brw_compile *p, GLuint mask, const struct brw_reg *arg0 ) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - //assert((mask & WRITEMASK_XYZW) == WRITEMASK_X || - // function == BRW_MATH_FUNCTION_SINCOS); - + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + brw_MOV(p, brw_message_reg(2), arg0[0]); /* Send two messages to perform all 16 operations: */ brw_math_16(p, - dst[0], + dst[dst_chan], function, (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 2, @@ -659,10 +662,12 @@ static void emit_math2( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert((mask & WRITEMASK_XYZW) == WRITEMASK_X); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); brw_push_insn_state(p); @@ -681,7 +686,7 @@ static void emit_math2( struct brw_compile *p, */ brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_math(p, - dst[0], + dst[dst_chan], function, (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 2, @@ -691,7 +696,7 @@ static void emit_math2( struct brw_compile *p, brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); brw_math(p, - offset(dst[0],1), + offset(dst[dst_chan],1), function, (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 4, @@ -1194,11 +1199,11 @@ void brw_wm_emit( struct brw_wm_compile *c ) /* Generated instructions for calculating triangle interpolants: */ case WM_PIXELXY: - emit_pixel_xy(p, dst, dst_flags, args[0]); + emit_pixel_xy(p, dst, dst_flags); break; case WM_DELTAXY: - emit_delta_xy(p, dst, dst_flags, args[0], args[1]); + emit_delta_xy(p, dst, dst_flags, args[0]); break; case WM_WPOSXY: @@ -1385,4 +1390,13 @@ void brw_wm_emit( struct brw_wm_compile *c ) inst->dst[i]->hw_reg, inst->dst[i]->spill_slot); } + + if (INTEL_DEBUG & DEBUG_WM) { + int i; + + _mesa_printf("wm-native:\n"); + for (i = 0; i < p->nr_insn; i++) + brw_disasm(stderr, &p->store[i]); + _mesa_printf("\n"); + } } diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c index b9e8dd2e96..5dc076a825 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_fp.c +++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c @@ -226,9 +226,42 @@ static struct prog_instruction * emit_op(struct brw_wm_compile *c, 0, 0, 0, /* tex unit, target, shadow */ src0, src1, src2); } - +/* Many Mesa opcodes produce the same value across all the result channels. + * We'd rather not have to support that splatting in the opcode implementations, + * and brw_wm_pass*.c wants to optimize them out by shuffling references around + * anyway. We can easily get both by emitting the opcode to one channel, and + * then MOVing it to the others, which brw_wm_pass*.c already understands. + */ +static struct prog_instruction *emit_scalar_insn(struct brw_wm_compile *c, + const struct prog_instruction *inst0) +{ + struct prog_instruction *inst; + unsigned int dst_chan; + unsigned int other_channel_mask; + + if (inst0->DstReg.WriteMask == 0) + return NULL; + + dst_chan = _mesa_ffs(inst0->DstReg.WriteMask) - 1; + inst = get_fp_inst(c); + *inst = *inst0; + inst->DstReg.WriteMask = 1 << dst_chan; + + other_channel_mask = inst0->DstReg.WriteMask & ~(1 << dst_chan); + if (other_channel_mask != 0) { + inst = emit_op(c, + OPCODE_MOV, + dst_mask(inst0->DstReg, other_channel_mask), + 0, + src_swizzle1(src_reg_from_dst(inst0->DstReg), dst_chan), + src_undef(), + src_undef()); + } + return inst; +} + /*********************************************************************** * Special instructions for interpolation and other tasks @@ -376,14 +409,6 @@ static void emit_interp( struct brw_wm_compile *c, } break; case FRAG_ATTRIB_FOGC: - /* The FOGC input is really special. When a program uses glFogFragCoord, - * the results returned are supposed to be (f,0,0,1). But for Mesa GLSL, - * the glFrontFacing and glPointCoord values are also stashed in FOGC. - * So, write the interpolated fog value to X, then either 0, 1, or the - * stashed values to Y, Z, W. Note that this means that - * glFogFragCoord.yzw can be wrong in those cases! - */ - /* Interpolate the fog coordinate */ emit_op(c, WM_PINTERP, @@ -393,26 +418,40 @@ static void emit_interp( struct brw_wm_compile *c, deltas, get_pixel_w(c)); - /* Move the front facing value into FOGC.y if it's needed. */ - if (c->fp->program.UsesFrontFacing) { - emit_op(c, - WM_FRONTFACING, - dst_mask(dst, WRITEMASK_Y), - 0, - src_undef(), - src_undef(), - src_undef()); - } else { - emit_op(c, - OPCODE_MOV, - dst_mask(dst, WRITEMASK_Y), - 0, - src_swizzle1(interp, SWIZZLE_ZERO), - src_undef(), - src_undef()); - } + emit_op(c, + OPCODE_MOV, + dst_mask(dst, WRITEMASK_YZW), + 0, + src_swizzle(interp, + SWIZZLE_ZERO, + SWIZZLE_ZERO, + SWIZZLE_ZERO, + SWIZZLE_ONE), + src_undef(), + src_undef()); + break; + + case FRAG_ATTRIB_FACE: + /* XXX review/test this case */ + emit_op(c, + WM_FRONTFACING, + dst_mask(dst, WRITEMASK_X), + 0, + src_undef(), + src_undef(), + src_undef()); + break; + + case FRAG_ATTRIB_PNTC: + /* XXX review/test this case */ + emit_op(c, + WM_PINTERP, + dst_mask(dst, WRITEMASK_XY), + 0, + interp, + deltas, + get_pixel_w(c)); - /* Should do the PointCoord thing here. */ emit_op(c, OPCODE_MOV, dst_mask(dst, WRITEMASK_ZW), @@ -425,6 +464,7 @@ static void emit_interp( struct brw_wm_compile *c, src_undef(), src_undef()); break; + default: emit_op(c, WM_PINTERP, @@ -683,7 +723,7 @@ static void precalc_tex( struct brw_wm_compile *c, /* tmp0 = 1 / tmp1 */ emit_op(c, OPCODE_RCP, - tmp0, + dst_mask(tmp0, WRITEMASK_X), 0, tmp1src, src_undef(), @@ -694,7 +734,7 @@ static void precalc_tex( struct brw_wm_compile *c, tmpcoord, 0, src0, - tmp0src, + src_swizzle1(tmp0src, SWIZZLE_X), src_undef()); release_temp(c, tmp0); @@ -717,7 +757,11 @@ static void precalc_tex( struct brw_wm_compile *c, tmpcoord, 0, inst->SrcReg[0], - scale, + src_swizzle(scale, + SWIZZLE_X, + SWIZZLE_Y, + SWIZZLE_ONE, + SWIZZLE_ONE), src_undef()); coord = src_reg_from_dst(tmpcoord); @@ -1134,9 +1178,11 @@ void brw_wm_pass_fp( struct brw_wm_compile *c ) break; case OPCODE_PRINT: break; - default: - emit_insn(c, inst); + if (brw_wm_is_scalar_result(inst->Opcode)) + emit_scalar_insn(c, inst); + else + emit_insn(c, inst); break; } } diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 19f777fe32..a5b18ec7d7 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -10,6 +10,9 @@ enum _subroutine { SUB_NOISE1, SUB_NOISE2, SUB_NOISE3, SUB_NOISE4 }; +static struct brw_reg get_dst_reg(struct brw_wm_compile *c, + const struct prog_instruction *inst, + GLuint component); /** * Determine if the given fragment program uses GLSL features such @@ -22,6 +25,7 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) for (i = 0; i < fp->Base.NumInstructions; i++) { const struct prog_instruction *inst = &fp->Base.Instructions[i]; switch (inst->Opcode) { + case OPCODE_ARL: case OPCODE_IF: case OPCODE_ENDIF: case OPCODE_CAL: @@ -130,19 +134,6 @@ static void set_reg(struct brw_wm_compile *c, int file, int index, c->wm_regs[file][index][component].inited = GL_TRUE; } -/** - * Examine instruction's write mask to find index of first component - * enabled for writing. - */ -static int get_scalar_dst_index(const struct prog_instruction *inst) -{ - int i; - for (i = 0; i < 4; i++) - if (inst->DstReg.WriteMask & (1<<i)) - break; - return i; -} - static struct brw_reg alloc_tmp(struct brw_wm_compile *c) { struct brw_reg reg; @@ -402,6 +393,27 @@ static void prealloc_reg(struct brw_wm_compile *c) prealloc_grf(c, 126); prealloc_grf(c, 127); + for (i = 0; i < c->nr_fp_insns; i++) { + const struct prog_instruction *inst = &c->prog_instructions[i]; + struct brw_reg dst[4]; + + switch (inst->Opcode) { + case OPCODE_TEX: + case OPCODE_TXB: + /* Allocate the channels of texture results contiguously, + * since they are written out that way by the sampler unit. + */ + for (j = 0; j < 4; j++) { + dst[j] = get_dst_reg(c, inst, j); + if (j != 0) + assert(dst[j].nr == dst[j - 1].nr + 1); + } + break; + default: + break; + } + } + /* An instruction may reference up to three constants. * They'll be found in these registers. * XXX alloc these on demand! @@ -639,23 +651,6 @@ static void invoke_subroutine( struct brw_wm_compile *c, } } -static void emit_abs( struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - int i; - struct brw_compile *p = &c->func; - brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF); - for (i = 0; i < 4; i++) { - if (inst->DstReg.WriteMask & (1<<i)) { - struct brw_reg src, dst; - dst = get_dst_reg(c, inst, i); - src = get_src_reg(c, inst, 0, i); - brw_MOV(p, dst, brw_abs(src)); - } - } - brw_set_saturate(p, 0); -} - static void emit_trunc( struct brw_wm_compile *c, const struct prog_instruction *inst) { @@ -1031,12 +1026,20 @@ static void emit_dp3(struct brw_wm_compile *c, struct brw_reg src0[3], src1[3], dst; int i; struct brw_compile *p = &c->func; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + for (i = 0; i < 3; i++) { src0[i] = get_src_reg(c, inst, 0, i); src1[i] = get_src_reg_imm(c, inst, 1, i); } - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + dst = get_dst_reg(c, inst, dst_chan); brw_MUL(p, brw_null_reg(), src0[0], src1[0]); brw_MAC(p, brw_null_reg(), src0[1], src1[1]); brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); @@ -1050,11 +1053,19 @@ static void emit_dp4(struct brw_wm_compile *c, struct brw_reg src0[4], src1[4], dst; int i; struct brw_compile *p = &c->func; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + for (i = 0; i < 4; i++) { src0[i] = get_src_reg(c, inst, 0, i); src1[i] = get_src_reg_imm(c, inst, 1, i); } - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + dst = get_dst_reg(c, inst, dst_chan); brw_MUL(p, brw_null_reg(), src0[0], src1[0]); brw_MAC(p, brw_null_reg(), src0[1], src1[1]); brw_MAC(p, brw_null_reg(), src0[2], src1[2]); @@ -1069,11 +1080,19 @@ static void emit_dph(struct brw_wm_compile *c, struct brw_reg src0[4], src1[4], dst; int i; struct brw_compile *p = &c->func; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + for (i = 0; i < 4; i++) { src0[i] = get_src_reg(c, inst, 0, i); src1[i] = get_src_reg_imm(c, inst, 1, i); } - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + dst = get_dst_reg(c, inst, dst_chan); brw_MUL(p, brw_null_reg(), src0[0], src1[0]); brw_MAC(p, brw_null_reg(), src0[1], src1[1]); brw_MAC(p, dst, src0[2], src1[2]); @@ -1091,37 +1110,28 @@ static void emit_math1(struct brw_wm_compile *c, const struct prog_instruction *inst, GLuint func) { struct brw_compile *p = &c->func; - struct brw_reg src0, dst, tmp; - const int mark = mark_tmps( c ); - int i; + struct brw_reg src0, dst; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; - tmp = alloc_tmp(c); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); /* Get first component of source register */ + dst = get_dst_reg(c, inst, dst_chan); src0 = get_src_reg(c, inst, 0, 0); - /* tmp = func(src0) */ brw_MOV(p, brw_message_reg(2), src0); brw_math(p, - tmp, + dst, func, (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 2, brw_null_reg(), BRW_MATH_DATA_VECTOR, BRW_MATH_PRECISION_FULL); - - /*tmp.dw1.bits.swizzle = SWIZZLE_XXXX;*/ - - /* replicate tmp value across enabled dest channels */ - for (i = 0; i < 4; i++) { - if (inst->DstReg.WriteMask & (1 << i)) { - dst = get_dst_reg(c, inst, i); - brw_MOV(p, dst, tmp); - } - } - - release_tmps(c, mark); } static void emit_rcp(struct brw_wm_compile *c, @@ -1192,24 +1202,6 @@ static void emit_arl(struct brw_wm_compile *c, brw_set_saturate(p, 0); } -static void emit_sub(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, src1, dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - for (i = 0 ; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - src1 = get_src_reg_imm(c, inst, 1, i); - brw_ADD(p, dst, src0, negate(src1)); - } - } - brw_set_saturate(p, 0); -} static void emit_mul(struct brw_wm_compile *c, const struct prog_instruction *inst) @@ -1321,7 +1313,15 @@ static void emit_pow(struct brw_wm_compile *c, { struct brw_compile *p = &c->func; struct brw_reg dst, src0, src1; - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + + dst = get_dst_reg(c, inst, dst_chan); src0 = get_src_reg_imm(c, inst, 0, 0); src1 = get_src_reg_imm(c, inst, 1, 0); @@ -2828,18 +2828,12 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) case WM_FRONTFACING: emit_frontfacing(c, inst); break; - case OPCODE_ABS: - emit_abs(c, inst); - break; case OPCODE_ADD: emit_add(c, inst); break; case OPCODE_ARL: emit_arl(c, inst); break; - case OPCODE_SUB: - emit_sub(c, inst); - break; case OPCODE_FRC: emit_frc(c, inst); break; @@ -3007,7 +3001,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) loop_depth--; inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); - /* patch all the BREAK/CONT instructions from last BEGINLOOP */ + /* patch all the BREAK/CONT instructions from last BGNLOOP */ while (inst0 > loop_inst[loop_depth]) { inst0--; if (inst0->header.opcode == BRW_OPCODE_BREAK) { @@ -3032,8 +3026,14 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) brw_set_predicate_control(p, BRW_PREDICATE_NONE); } post_wm_emit(c); -} + if (INTEL_DEBUG & DEBUG_WM) { + _mesa_printf("wm-native:\n"); + for (i = 0; i < p->nr_insn; i++) + brw_disasm(stderr, &p->store[i]); + _mesa_printf("\n"); + } +} /** * Do GPU code generation for shaders that use GLSL features such as diff --git a/src/mesa/drivers/dri/i965/brw_wm_iz.c b/src/mesa/drivers/dri/i965/brw_wm_iz.c index 8fd067abe7..5e399ac62a 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_iz.c +++ b/src/mesa/drivers/dri/i965/brw_wm_iz.c @@ -122,6 +122,7 @@ const struct { */ void brw_wm_lookup_iz( GLuint line_aa, GLuint lookup, + GLboolean ps_uses_depth, struct brw_wm_prog_key *key ) { GLuint reg = 2; @@ -131,7 +132,7 @@ void brw_wm_lookup_iz( GLuint line_aa, if (lookup & IZ_PS_COMPUTES_DEPTH_BIT) key->computes_depth = 1; - if (wm_iz_table[lookup].sd_present) { + if (wm_iz_table[lookup].sd_present || ps_uses_depth) { key->source_depth_reg = reg; reg += 2; } diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass0.c b/src/mesa/drivers/dri/i965/brw_wm_pass0.c index 92142764f5..6279258339 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass0.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass0.c @@ -257,34 +257,6 @@ static void pass0_set_dst( struct brw_wm_compile *c, } -static void pass0_set_dst_scalar( struct brw_wm_compile *c, - struct brw_wm_instruction *out, - const struct prog_instruction *inst, - GLuint writemask ) -{ - if (writemask) { - const struct prog_dst_register *dst = &inst->DstReg; - GLuint i; - - /* Compute only the first (X) value: - */ - out->writemask = WRITEMASK_X; - out->dst[0] = get_value(c); - - /* Update our tracking register file for all the components in - * writemask: - */ - for (i = 0; i < 4; i++) { - if (writemask & (1<<i)) { - pass0_set_fpreg_value(c, dst->File, dst->Index, i, out->dst[0]); - } - } - } - else - out->writemask = 0; -} - - static const struct brw_wm_ref *get_fp_src_reg_ref( struct brw_wm_compile *c, struct prog_src_register src, GLuint i ) @@ -363,10 +335,7 @@ translate_insn(struct brw_wm_compile *c, /* Dst: */ - if (brw_wm_is_scalar_result(out->opcode)) - pass0_set_dst_scalar(c, out, inst, writemask); - else - pass0_set_dst(c, out, inst, writemask); + pass0_set_dst(c, out, inst, writemask); } diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index 3fc18ff1f3..dff466587a 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -103,6 +103,10 @@ struct wm_sampler_key { GLenum minfilter, magfilter; GLenum comparemode, comparefunc; dri_bo *sdc_bo; + + /** If target is cubemap, take context setting. + */ + GLboolean seamless_cube_map; } sampler[BRW_MAX_TEX_UNIT]; }; @@ -169,30 +173,33 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key, } } - if (key->tex_target == GL_TEXTURE_CUBE_MAP && - (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) { - /* If we're using anything but nearest sampling for a cube map, we - * need to set this wrap mode to avoid GPU lock-ups. - */ - sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE; - sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; - sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; - } - else if (key->tex_target == GL_TEXTURE_1D) { + sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); + sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); + sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t); + + /* Cube-maps on 965 and later must use the same wrap mode for all 3 + * coordinate dimensions. Futher, only CUBE and CLAMP are valid. + */ + if (key->tex_target == GL_TEXTURE_CUBE_MAP) { + if (key->seamless_cube_map && + (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) { + sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE; + sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; + } else { + sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP; + sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP; + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP; + } + } else if (key->tex_target == GL_TEXTURE_1D) { /* There's a bug in 1D texture sampling - it actually pays * attention to the wrap_t value, though it should not. * Override the wrap_t value here to GL_REPEAT to keep * any nonexistent border pixels from floating in. */ - sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); - sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP; } - else { - sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); - sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); - sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t); - } + /* Set shadow function: */ @@ -249,6 +256,9 @@ brw_wm_sampler_populate_key(struct brw_context *brw, entry->tex_target = texObj->Target; + entry->seamless_cube_map = (texObj->Target == GL_TEXTURE_CUBE_MAP) + ? ctx->Texture.CubeMapSeamless : GL_FALSE; + entry->wrap_r = texObj->WrapR; entry->wrap_s = texObj->WrapS; entry->wrap_t = texObj->WrapT; |