static void a2xx_drawctxt_draw_workaround(struct adreno_device *adreno_dev,
struct adreno_context *context)
{
struct kgsl_device *device = &adreno_dev->dev;
unsigned int cmd[11];
unsigned int *cmds = &cmd[0];
if (adreno_is_a225(adreno_dev)) {
adreno_dev->gpudev->ctx_switches_since_last_draw++;
if (adreno_dev->gpudev->ctx_switches_since_last_draw >
ADRENO_NUM_CTX_SWITCH_ALLOWED_BEFORE_DRAW)
adreno_dev->gpudev->ctx_switches_since_last_draw = 0;
else
return;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = (0x4 << 16) | (REG_PA_SU_SC_MODE_CNTL - 0x2000);
*cmds++ = 0;
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 5);
*cmds++ = 0;
*cmds++ = 1<<14;
*cmds++ = 0;
*cmds++ = device->mmu.setstate_memory.gpuaddr;
*cmds++ = 0;
*cmds++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmds++ = 0x00000000;
} else {
*cmds++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
*cmds++ = adreno_encode_istore_size(adreno_dev)
| adreno_dev->pix_shader_start;
}
adreno_ringbuffer_issuecmds(device, context, KGSL_CMD_FLAGS_PMODE,
&cmd[0], cmds - cmd);
}
int adreno_ringbuffer_start(struct adreno_ringbuffer *rb, unsigned int init_ram)
{
int status;
/*cp_rb_cntl_u cp_rb_cntl; */
union reg_cp_rb_cntl cp_rb_cntl;
unsigned int *cmds, rb_cntl;
struct kgsl_device *device = rb->device;
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
uint cmds_gpu;
if (rb->flags & KGSL_FLAGS_STARTED)
return 0;
if (init_ram) {
rb->timestamp = 0;
GSL_RB_INIT_TIMESTAMP(rb);
}
kgsl_sharedmem_set(&rb->memptrs_desc, 0, 0,
sizeof(struct kgsl_rbmemptrs));
kgsl_sharedmem_set(&rb->buffer_desc, 0, 0xAA,
(rb->sizedwords << 2));
adreno_regwrite(device, REG_CP_RB_WPTR_BASE,
(rb->memptrs_desc.gpuaddr
+ GSL_RB_MEMPTRS_WPTRPOLL_OFFSET));
/* setup WPTR delay */
adreno_regwrite(device, REG_CP_RB_WPTR_DELAY, 0 /*0x70000010 */);
/*setup REG_CP_RB_CNTL */
adreno_regread(device, REG_CP_RB_CNTL, &rb_cntl);
cp_rb_cntl.val = rb_cntl;
/*
* The size of the ringbuffer in the hardware is the log2
* representation of the size in quadwords (sizedwords / 2)
*/
cp_rb_cntl.f.rb_bufsz = ilog2(rb->sizedwords >> 1);
/*
* Specify the quadwords to read before updating mem RPTR.
* Like above, pass the log2 representation of the blocksize
* in quadwords.
*/
cp_rb_cntl.f.rb_blksz = ilog2(KGSL_RB_BLKSIZE >> 3);
cp_rb_cntl.f.rb_poll_en = GSL_RB_CNTL_POLL_EN; /* WPTR polling */
/* mem RPTR writebacks */
cp_rb_cntl.f.rb_no_update = GSL_RB_CNTL_NO_UPDATE;
adreno_regwrite(device, REG_CP_RB_CNTL, cp_rb_cntl.val);
adreno_regwrite(device, REG_CP_RB_BASE, rb->buffer_desc.gpuaddr);
adreno_regwrite(device, REG_CP_RB_RPTR_ADDR,
rb->memptrs_desc.gpuaddr +
GSL_RB_MEMPTRS_RPTR_OFFSET);
/* explicitly clear all cp interrupts */
adreno_regwrite(device, REG_CP_INT_ACK, 0xFFFFFFFF);
/* setup scratch/timestamp */
adreno_regwrite(device, REG_SCRATCH_ADDR,
device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(soptimestamp));
adreno_regwrite(device, REG_SCRATCH_UMSK,
GSL_RB_MEMPTRS_SCRATCH_MASK);
/* update the eoptimestamp field with the last retired timestamp */
kgsl_sharedmem_writel(&device->memstore,
KGSL_DEVICE_MEMSTORE_OFFSET(eoptimestamp),
rb->timestamp);
/* load the CP ucode */
status = adreno_ringbuffer_load_pm4_ucode(device);
if (status != 0)
return status;
/* load the prefetch parser ucode */
status = adreno_ringbuffer_load_pfp_ucode(device);
if (status != 0)
return status;
adreno_regwrite(device, REG_CP_QUEUE_THRESHOLDS, 0x000C0804);
rb->rptr = 0;
rb->wptr = 0;
/* clear ME_HALT to start micro engine */
adreno_regwrite(device, REG_CP_ME_CNTL, 0);
/* ME_INIT */
cmds = adreno_ringbuffer_allocspace(rb, 19);
cmds_gpu = rb->buffer_desc.gpuaddr + sizeof(uint)*(rb->wptr-19);
GSL_RB_WRITE(cmds, cmds_gpu, CP_HDR_ME_INIT);
/* All fields present (bits 9:0) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x000003ff);
/* Disable/Enable Real-Time Stream processing (present but ignored) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* Enable (2D <-> 3D) implicit synchronization (present but ignored) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_RB_SURFACE_INFO));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_PA_SC_WINDOW_OFFSET));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_VGT_MAX_VTX_INDX));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_SQ_PROGRAM_CNTL));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_RB_DEPTHCONTROL));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_PA_SU_POINT_SIZE));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_PA_SC_LINE_CNTL));
GSL_RB_WRITE(cmds, cmds_gpu,
SUBBLOCK_OFFSET(REG_PA_SU_POLY_OFFSET_FRONT_SCALE));
/* Instruction memory size: */
GSL_RB_WRITE(cmds, cmds_gpu,
(adreno_encode_istore_size(adreno_dev)
| adreno_dev->pix_shader_start));
/* Maximum Contexts */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000001);
/* Write Confirm Interval and The CP will wait the
* wait_interval * 16 clocks between polling */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* NQ and External Memory Swap */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* Protected mode error checking */
GSL_RB_WRITE(cmds, cmds_gpu, GSL_RB_PROTECTED_MODE_CONTROL);
/* Disable header dumping and Header dump address */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* Header dump size */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
adreno_ringbuffer_submit(rb);
/* idle device to validate ME INIT */
status = adreno_idle(device);
if (status == 0)
rb->flags |= KGSL_FLAGS_STARTED;
return status;
}
/*copy colour, depth, & stencil buffers from system memory to graphics memory*/
static unsigned int *build_sys2gmem_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
struct gmem_shadow_t *shadow)
{
unsigned int *cmds = shadow->gmem_restore_commands;
unsigned int *start = cmds;
/* Store TP0_CHICKEN register */
*cmds++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmds++ = REG_TP0_CHICKEN;
*cmds++ = tmp_ctx.chicken_restore;
*cmds++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmds++ = 0;
/* Set TP0_CHICKEN to zero */
*cmds++ = cp_type0_packet(REG_TP0_CHICKEN, 1);
*cmds++ = 0x00000000;
/* Set PA_SC_AA_CONFIG to 0 */
*cmds++ = cp_type0_packet(REG_PA_SC_AA_CONFIG, 1);
*cmds++ = 0x00000000;
/* shader constants */
/* vertex buffer constants */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 7);
*cmds++ = (0x1 << 16) | (9 * 6);
/* valid(?) vtx constant flag & addr */
*cmds++ = shadow->quad_vertices.gpuaddr | 0x3;
/* limit = 12 dwords */
*cmds++ = 0x00000030;
/* valid(?) vtx constant flag & addr */
*cmds++ = shadow->quad_texcoords.gpuaddr | 0x3;
/* limit = 8 dwords */
*cmds++ = 0x00000020;
*cmds++ = 0;
*cmds++ = 0;
/* Invalidate L2 cache to make sure vertices are updated */
*cmds++ = cp_type0_packet(REG_TC_CNTL_STATUS, 1);
*cmds++ = 0x1;
cmds = program_shader(cmds, 0, sys2gmem_vtx_pgm, SYS2GMEM_VTX_PGM_LEN);
/* Repartition shaders */
*cmds++ = cp_type0_packet(REG_SQ_INST_STORE_MANAGMENT, 1);
*cmds++ = 0x180;
/* Invalidate Vertex & Pixel instruction code address and sizes */
*cmds++ = cp_type3_packet(CP_INVALIDATE_STATE, 1);
*cmds++ = 0x00000300; /* 0x100 = Vertex, 0x200 = Pixel */
*cmds++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
*cmds++ = adreno_encode_istore_size(adreno_dev)
| adreno_dev->pix_shader_start;
/* Load the patched fragment shader stream */
cmds =
program_shader(cmds, 1, sys2gmem_frag_pgm, SYS2GMEM_FRAG_PGM_LEN);
/* SQ_PROGRAM_CNTL / SQ_CONTEXT_MISC */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_PROGRAM_CNTL);
*cmds++ = 0x10030002;
*cmds++ = 0x00000008;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_AA_MASK);
*cmds++ = 0x0000ffff; /* REG_PA_SC_AA_MASK */
if (!adreno_is_a22x(adreno_dev)) {
/* PA_SC_VIZ_QUERY */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_VIZ_QUERY);
*cmds++ = 0x0; /*REG_PA_SC_VIZ_QUERY */
}
/* RB_COLORCONTROL */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLORCONTROL);
*cmds++ = 0x00000c20;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 4);
*cmds++ = CP_REG(REG_VGT_MAX_VTX_INDX);
*cmds++ = 0x00ffffff; /* mmVGT_MAX_VTX_INDX */
*cmds++ = 0x0; /* mmVGT_MIN_VTX_INDX */
*cmds++ = 0x00000000; /* mmVGT_INDX_OFFSET */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_VGT_VERTEX_REUSE_BLOCK_CNTL);
*cmds++ = 0x00000002; /* mmVGT_VERTEX_REUSE_BLOCK_CNTL */
*cmds++ = 0x00000002; /* mmVGT_OUT_DEALLOC_CNTL */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_SQ_INTERPOLATOR_CNTL);
*cmds++ = 0xffffffff; /* mmSQ_INTERPOLATOR_CNTL */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_AA_CONFIG);
*cmds++ = 0x00000000; /* REG_PA_SC_AA_CONFIG */
/* set REG_PA_SU_SC_MODE_CNTL
* Front_ptype = draw triangles
* Back_ptype = draw triangles
* Provoking vertex = last
*/
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SU_SC_MODE_CNTL);
*cmds++ = 0x00080240;
/* texture constants */
*cmds++ =
cp_type3_packet(CP_SET_CONSTANT, (SYS2GMEM_TEX_CONST_LEN + 1));
*cmds++ = (0x1 << 16) | (0 * 6);
memcpy(cmds, sys2gmem_tex_const, SYS2GMEM_TEX_CONST_LEN << 2);
cmds[0] |= (shadow->pitch >> 5) << 22;
cmds[1] |=
shadow->gmemshadow.gpuaddr | surface_format_table[shadow->format];
cmds[2] |= (shadow->width - 1) | (shadow->height - 1) << 13;
cmds += SYS2GMEM_TEX_CONST_LEN;
/* program surface info */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_RB_SURFACE_INFO);
*cmds++ = shadow->gmem_pitch; /* pitch, MSAA = 1 */
/* RB_COLOR_INFO Endian=none, Linear, Format=RGBA8888, Swap=0,
* Base=gmem_base
*/
*cmds++ =
(shadow->
format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT) | tmp_ctx.gmem_base;
/* RB_DEPTHCONTROL */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_DEPTHCONTROL);
if (adreno_is_a22x(adreno_dev))
*cmds++ = 8; /* disable Z */
else
*cmds++ = 0; /* disable Z */
/* Use maximum scissor values -- quad vertices already
* have the correct bounds */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_SCREEN_SCISSOR_TL);
*cmds++ = (0 << 16) | 0;
*cmds++ = ((0x1fff) << 16) | 0x1fff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_WINDOW_SCISSOR_TL);
*cmds++ = (unsigned int)((1U << 31) | (0 << 16) | 0);
*cmds++ = ((0x1fff) << 16) | 0x1fff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_VTE_CNTL);
/* disable X/Y/Z transforms, X/Y/Z are premultiplied by W */
*cmds++ = 0x00000b00;
/*load the viewport so that z scale = clear depth and z offset = 0.0f */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_CL_VPORT_ZSCALE);
*cmds++ = 0xbf800000;
*cmds++ = 0x0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_MASK);
*cmds++ = 0x0000000f; /* R = G = B = 1:enabled */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_DEST_MASK);
*cmds++ = 0xffffffff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_WRAPPING_0);
*cmds++ = 0x00000000;
*cmds++ = 0x00000000;
/* load the stencil ref value
* $AAM - do this later
*/
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_MODECONTROL);
/* draw pixels with color and depth/stencil component */
*cmds++ = 0x4;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_CLIP_CNTL);
*cmds++ = 0x00010000;
if (adreno_is_a22x(adreno_dev)) {
*cmds++ = cp_type3_packet(CP_SET_DRAW_INIT_FLAGS, 1);
*cmds++ = 0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_A220_RB_LRZ_VSC_CONTROL);
*cmds++ = 0x0000000;
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 3);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, SrcSel=AutoIndex, VisCullMode=Ignore*/
*cmds++ = 0x00004088;
*cmds++ = 3; /* NumIndices=3 */
} else {
/* queue the draw packet */
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 2);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, NumIndices=3, SrcSel=AutoIndex */
*cmds++ = 0x00030088;
}
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, shadow->gmem_restore, start, cmds);
return cmds;
}
/*copy colour, depth, & stencil buffers from graphics memory to system memory*/
static unsigned int *build_gmem2sys_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
struct gmem_shadow_t *shadow)
{
unsigned int *cmds = shadow->gmem_save_commands;
unsigned int *start = cmds;
/* Calculate the new offset based on the adjusted base */
unsigned int bytesperpixel = format2bytesperpixel[shadow->format];
unsigned int addr = shadow->gmemshadow.gpuaddr;
unsigned int offset = (addr - (addr & 0xfffff000)) / bytesperpixel;
/* Store TP0_CHICKEN register */
*cmds++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmds++ = REG_TP0_CHICKEN;
*cmds++ = tmp_ctx.chicken_restore;
*cmds++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmds++ = 0;
/* Set TP0_CHICKEN to zero */
*cmds++ = cp_type0_packet(REG_TP0_CHICKEN, 1);
*cmds++ = 0x00000000;
/* Set PA_SC_AA_CONFIG to 0 */
*cmds++ = cp_type0_packet(REG_PA_SC_AA_CONFIG, 1);
*cmds++ = 0x00000000;
/* program shader */
/* load shader vtx constants ... 5 dwords */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 4);
*cmds++ = (0x1 << 16) | SHADER_CONST_ADDR;
*cmds++ = 0;
/* valid(?) vtx constant flag & addr */
*cmds++ = shadow->quad_vertices.gpuaddr | 0x3;
/* limit = 12 dwords */
*cmds++ = 0x00000030;
/* Invalidate L2 cache to make sure vertices are updated */
*cmds++ = cp_type0_packet(REG_TC_CNTL_STATUS, 1);
*cmds++ = 0x1;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 4);
*cmds++ = CP_REG(REG_VGT_MAX_VTX_INDX);
*cmds++ = 0x00ffffff; /* REG_VGT_MAX_VTX_INDX */
*cmds++ = 0x0; /* REG_VGT_MIN_VTX_INDX */
*cmds++ = 0x00000000; /* REG_VGT_INDX_OFFSET */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_AA_MASK);
*cmds++ = 0x0000ffff; /* REG_PA_SC_AA_MASK */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLORCONTROL);
*cmds++ = 0x00000c20;
/* Repartition shaders */
*cmds++ = cp_type0_packet(REG_SQ_INST_STORE_MANAGMENT, 1);
*cmds++ = 0x180;
/* Invalidate Vertex & Pixel instruction code address and sizes */
*cmds++ = cp_type3_packet(CP_INVALIDATE_STATE, 1);
*cmds++ = 0x00003F00;
*cmds++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
*cmds++ = adreno_encode_istore_size(adreno_dev)
| adreno_dev->pix_shader_start;
/* load the patched vertex shader stream */
cmds = program_shader(cmds, 0, gmem2sys_vtx_pgm, GMEM2SYS_VTX_PGM_LEN);
/* Load the patched fragment shader stream */
cmds =
program_shader(cmds, 1, gmem2sys_frag_pgm, GMEM2SYS_FRAG_PGM_LEN);
/* SQ_PROGRAM_CNTL / SQ_CONTEXT_MISC */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_PROGRAM_CNTL);
if (adreno_is_a22x(adreno_dev))
*cmds++ = 0x10018001;
else
*cmds++ = 0x10010001;
*cmds++ = 0x00000008;
/* resolve */
/* PA_CL_VTE_CNTL */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_VTE_CNTL);
/* disable X/Y/Z transforms, X/Y/Z are premultiplied by W */
*cmds++ = 0x00000b00;
/* program surface info */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_RB_SURFACE_INFO);
*cmds++ = shadow->gmem_pitch; /* pitch, MSAA = 1 */
/* RB_COLOR_INFO Endian=none, Linear, Format=RGBA8888, Swap=0,
* Base=gmem_base
*/
/* gmem base assumed 4K aligned. */
BUG_ON(tmp_ctx.gmem_base & 0xFFF);
*cmds++ =
(shadow->
format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT) | tmp_ctx.gmem_base;
/* disable Z */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_DEPTHCONTROL);
if (adreno_is_a22x(adreno_dev))
*cmds++ = 0x08;
else
*cmds++ = 0;
/* set REG_PA_SU_SC_MODE_CNTL
* Front_ptype = draw triangles
* Back_ptype = draw triangles
* Provoking vertex = last
*/
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SU_SC_MODE_CNTL);
*cmds++ = 0x00080240;
/* Use maximum scissor values -- quad vertices already have the
* correct bounds */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_SCREEN_SCISSOR_TL);
*cmds++ = (0 << 16) | 0;
*cmds++ = (0x1fff << 16) | (0x1fff);
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_WINDOW_SCISSOR_TL);
*cmds++ = (unsigned int)((1U << 31) | (0 << 16) | 0);
*cmds++ = (0x1fff << 16) | (0x1fff);
/* load the viewport so that z scale = clear depth and
* z offset = 0.0f
*/
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_CL_VPORT_ZSCALE);
*cmds++ = 0xbf800000; /* -1.0f */
*cmds++ = 0x0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_MASK);
*cmds++ = 0x0000000f; /* R = G = B = 1:enabled */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_DEST_MASK);
*cmds++ = 0xffffffff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_WRAPPING_0);
*cmds++ = 0x00000000;
*cmds++ = 0x00000000;
/* load the stencil ref value
* $AAM - do this later
*/
/* load the COPY state */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 6);
*cmds++ = CP_REG(REG_RB_COPY_CONTROL);
*cmds++ = 0; /* RB_COPY_CONTROL */
*cmds++ = addr & 0xfffff000; /* RB_COPY_DEST_BASE */
*cmds++ = shadow->pitch >> 5; /* RB_COPY_DEST_PITCH */
/* Endian=none, Linear, Format=RGBA8888,Swap=0,!Dither,
* MaskWrite:R=G=B=A=1
*/
*cmds++ = 0x0003c008 |
(shadow->format << RB_COPY_DEST_INFO__COPY_DEST_FORMAT__SHIFT);
/* Make sure we stay in offsetx field. */
BUG_ON(offset & 0xfffff000);
*cmds++ = offset;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_MODECONTROL);
*cmds++ = 0x6; /* EDRAM copy */
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_CLIP_CNTL);
*cmds++ = 0x00010000;
if (adreno_is_a22x(adreno_dev)) {
*cmds++ = cp_type3_packet(CP_SET_DRAW_INIT_FLAGS, 1);
*cmds++ = 0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_A220_RB_LRZ_VSC_CONTROL);
*cmds++ = 0x0000000;
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 3);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, SrcSel=AutoIndex, VisCullMode=Ignore*/
*cmds++ = 0x00004088;
*cmds++ = 3; /* NumIndices=3 */
} else {
/* queue the draw packet */
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 2);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, NumIndices=3, SrcSel=AutoIndex */
*cmds++ = 0x00030088;
}
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, shadow->gmem_save, start, cmds);
return cmds;
}
static void
build_shader_save_restore_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt)
{
unsigned int *cmd = tmp_ctx.cmd;
unsigned int *save, *restore, *fixup;
unsigned int *startSizeVtx, *startSizePix, *startSizeShared;
unsigned int *partition1;
unsigned int *shaderBases, *partition2;
/* compute vertex, pixel and shared instruction shadow GPU addresses */
tmp_ctx.shader_vertex = drawctxt->gpustate.gpuaddr + SHADER_OFFSET;
tmp_ctx.shader_pixel = tmp_ctx.shader_vertex
+ _shader_shadow_size(adreno_dev);
tmp_ctx.shader_shared = tmp_ctx.shader_pixel
+ _shader_shadow_size(adreno_dev);
/* restore shader partitioning and instructions */
restore = cmd; /* start address */
/* Invalidate Vertex & Pixel instruction code address and sizes */
*cmd++ = cp_type3_packet(CP_INVALIDATE_STATE, 1);
*cmd++ = 0x00000300; /* 0x100 = Vertex, 0x200 = Pixel */
/* Restore previous shader vertex & pixel instruction bases. */
*cmd++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
shaderBases = cmd++; /* TBD #5: shader bases (from fixup) */
/* write the shader partition information to a scratch register */
*cmd++ = cp_type0_packet(REG_SQ_INST_STORE_MANAGMENT, 1);
partition1 = cmd++; /* TBD #4a: partition info (from save) */
/* load vertex shader instructions from the shadow. */
*cmd++ = cp_type3_packet(CP_IM_LOAD, 2);
*cmd++ = tmp_ctx.shader_vertex + 0x0; /* 0x0 = Vertex */
startSizeVtx = cmd++; /* TBD #1: start/size (from save) */
/* load pixel shader instructions from the shadow. */
*cmd++ = cp_type3_packet(CP_IM_LOAD, 2);
*cmd++ = tmp_ctx.shader_pixel + 0x1; /* 0x1 = Pixel */
startSizePix = cmd++; /* TBD #2: start/size (from save) */
/* load shared shader instructions from the shadow. */
*cmd++ = cp_type3_packet(CP_IM_LOAD, 2);
*cmd++ = tmp_ctx.shader_shared + 0x2; /* 0x2 = Shared */
startSizeShared = cmd++; /* TBD #3: start/size (from save) */
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, drawctxt->shader_restore, restore, cmd);
/*
* fixup SET_SHADER_BASES data
*
* since self-modifying PM4 code is being used here, a seperate
* command buffer is used for this fixup operation, to ensure the
* commands are not read by the PM4 engine before the data fields
* have been written.
*/
fixup = cmd; /* start address */
/* write the shader partition information to a scratch register */
*cmd++ = cp_type0_packet(REG_SCRATCH_REG2, 1);
partition2 = cmd++; /* TBD #4b: partition info (from save) */
/* mask off unused bits, then OR with shader instruction memory size */
*cmd++ = cp_type3_packet(CP_REG_RMW, 3);
*cmd++ = REG_SCRATCH_REG2;
/* AND off invalid bits. */
*cmd++ = 0x0FFF0FFF;
/* OR in instruction memory size. */
*cmd++ = adreno_encode_istore_size(adreno_dev);
/* write the computed value to the SET_SHADER_BASES data field */
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SCRATCH_REG2;
/* TBD #5: shader bases (to restore) */
*cmd++ = virt2gpu(shaderBases, &drawctxt->gpustate);
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, drawctxt->shader_fixup, fixup, cmd);
/* save shader partitioning and instructions */
save = cmd; /* start address */
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
/* fetch the SQ_INST_STORE_MANAGMENT register value,
* store the value in the data fields of the SET_CONSTANT commands
* above.
*/
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SQ_INST_STORE_MANAGMENT;
/* TBD #4a: partition info (to restore) */
*cmd++ = virt2gpu(partition1, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SQ_INST_STORE_MANAGMENT;
/* TBD #4b: partition info (to fixup) */
*cmd++ = virt2gpu(partition2, &drawctxt->gpustate);
/* store the vertex shader instructions */
*cmd++ = cp_type3_packet(CP_IM_STORE, 2);
*cmd++ = tmp_ctx.shader_vertex + 0x0; /* 0x0 = Vertex */
/* TBD #1: start/size (to restore) */
*cmd++ = virt2gpu(startSizeVtx, &drawctxt->gpustate);
/* store the pixel shader instructions */
*cmd++ = cp_type3_packet(CP_IM_STORE, 2);
*cmd++ = tmp_ctx.shader_pixel + 0x1; /* 0x1 = Pixel */
/* TBD #2: start/size (to restore) */
*cmd++ = virt2gpu(startSizePix, &drawctxt->gpustate);
/* store the shared shader instructions if vertex base is nonzero */
*cmd++ = cp_type3_packet(CP_IM_STORE, 2);
*cmd++ = tmp_ctx.shader_shared + 0x2; /* 0x2 = Shared */
/* TBD #3: start/size (to restore) */
*cmd++ = virt2gpu(startSizeShared, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, drawctxt->shader_save, save, cmd);
tmp_ctx.cmd = cmd;
}
static unsigned int *build_sys2gmem_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
struct gmem_shadow_t *shadow)
{
unsigned int *cmds = shadow->gmem_restore_commands;
unsigned int *start = cmds;
if (!(drawctxt->flags & CTXT_FLAGS_PREAMBLE)) {
*cmds++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmds++ = REG_TP0_CHICKEN;
*cmds++ = tmp_ctx.chicken_restore;
*cmds++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmds++ = 0;
}
*cmds++ = cp_type0_packet(REG_TP0_CHICKEN, 1);
*cmds++ = 0x00000000;
*cmds++ = cp_type0_packet(REG_PA_SC_AA_CONFIG, 1);
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 7);
*cmds++ = (0x1 << 16) | (9 * 6);
*cmds++ = shadow->quad_vertices.gpuaddr | 0x3;
*cmds++ = 0x00000030;
*cmds++ = shadow->quad_texcoords.gpuaddr | 0x3;
*cmds++ = 0x00000020;
*cmds++ = 0;
*cmds++ = 0;
*cmds++ = cp_type0_packet(REG_TC_CNTL_STATUS, 1);
*cmds++ = 0x1;
cmds = program_shader(cmds, 0, sys2gmem_vtx_pgm, SYS2GMEM_VTX_PGM_LEN);
*cmds++ = cp_type0_packet(REG_SQ_INST_STORE_MANAGMENT, 1);
*cmds++ = adreno_dev->pix_shader_start;
*cmds++ = cp_type3_packet(CP_INVALIDATE_STATE, 1);
*cmds++ = 0x00000300;
*cmds++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
*cmds++ = adreno_encode_istore_size(adreno_dev)
| adreno_dev->pix_shader_start;
cmds =
program_shader(cmds, 1, sys2gmem_frag_pgm, SYS2GMEM_FRAG_PGM_LEN);
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_PROGRAM_CNTL);
*cmds++ = 0x10030002;
*cmds++ = 0x00000008;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_AA_MASK);
*cmds++ = 0x0000ffff;
if (!adreno_is_a22x(adreno_dev)) {
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_VIZ_QUERY);
*cmds++ = 0x0;
}
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLORCONTROL);
*cmds++ = 0x00000c20;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 4);
*cmds++ = CP_REG(REG_VGT_MAX_VTX_INDX);
*cmds++ = 0x00ffffff;
*cmds++ = 0x0;
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_VGT_VERTEX_REUSE_BLOCK_CNTL);
*cmds++ = 0x00000002;
*cmds++ = 0x00000002;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_SQ_INTERPOLATOR_CNTL);
*cmds++ = 0xffffffff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_AA_CONFIG);
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SU_SC_MODE_CNTL);
*cmds++ = 0x00080240;
*cmds++ =
cp_type3_packet(CP_SET_CONSTANT, (SYS2GMEM_TEX_CONST_LEN + 1));
*cmds++ = (0x1 << 16) | (0 * 6);
memcpy(cmds, sys2gmem_tex_const, SYS2GMEM_TEX_CONST_LEN << 2);
cmds[0] |= (shadow->pitch >> 5) << 22;
cmds[1] |=
shadow->gmemshadow.gpuaddr | surface_format_table[shadow->format];
cmds[2] |= (shadow->width - 1) | (shadow->height - 1) << 13;
cmds += SYS2GMEM_TEX_CONST_LEN;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_RB_SURFACE_INFO);
*cmds++ = shadow->gmem_pitch;
*cmds++ =
(shadow->
format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT) | tmp_ctx.gmem_base;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_DEPTHCONTROL);
if (adreno_is_a22x(adreno_dev))
*cmds++ = 8;
else
*cmds++ = 0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_SCREEN_SCISSOR_TL);
*cmds++ = (0 << 16) | 0;
*cmds++ = ((0x1fff) << 16) | 0x1fff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_WINDOW_SCISSOR_TL);
*cmds++ = (unsigned int)((1U << 31) | (0 << 16) | 0);
*cmds++ = ((0x1fff) << 16) | 0x1fff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_VTE_CNTL);
*cmds++ = 0x00000b00;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_CL_VPORT_ZSCALE);
*cmds++ = 0xbf800000;
*cmds++ = 0x0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_MASK);
*cmds++ = 0x0000000f;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_DEST_MASK);
*cmds++ = 0xffffffff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_WRAPPING_0);
*cmds++ = 0x00000000;
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_MODECONTROL);
*cmds++ = 0x4;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_CLIP_CNTL);
*cmds++ = 0x00010000;
if (adreno_is_a22x(adreno_dev)) {
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_A220_RB_LRZ_VSC_CONTROL);
*cmds++ = 0x0000000;
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 3);
*cmds++ = 0;
*cmds++ = 0x00004088;
*cmds++ = 3;
} else {
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 2);
*cmds++ = 0;
*cmds++ = 0x00030088;
}
create_ib1(drawctxt, shadow->gmem_restore, start, cmds);
return cmds;
}
static unsigned int *build_gmem2sys_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
struct gmem_shadow_t *shadow)
{
unsigned int *cmds = shadow->gmem_save_commands;
unsigned int *start = cmds;
unsigned int bytesperpixel = format2bytesperpixel[shadow->format];
unsigned int addr = shadow->gmemshadow.gpuaddr;
unsigned int offset = (addr - (addr & 0xfffff000)) / bytesperpixel;
if (!(drawctxt->flags & CTXT_FLAGS_PREAMBLE)) {
*cmds++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmds++ = REG_TP0_CHICKEN;
*cmds++ = tmp_ctx.chicken_restore;
*cmds++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmds++ = 0;
}
*cmds++ = cp_type0_packet(REG_TP0_CHICKEN, 1);
*cmds++ = 0x00000000;
*cmds++ = cp_type0_packet(REG_PA_SC_AA_CONFIG, 1);
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 4);
*cmds++ = (0x1 << 16) | SHADER_CONST_ADDR;
*cmds++ = 0;
*cmds++ = shadow->quad_vertices.gpuaddr | 0x3;
*cmds++ = 0x00000030;
*cmds++ = cp_type0_packet(REG_TC_CNTL_STATUS, 1);
*cmds++ = 0x1;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 4);
*cmds++ = CP_REG(REG_VGT_MAX_VTX_INDX);
*cmds++ = 0x00ffffff;
*cmds++ = 0x0;
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SC_AA_MASK);
*cmds++ = 0x0000ffff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLORCONTROL);
*cmds++ = 0x00000c20;
*cmds++ = cp_type0_packet(REG_SQ_INST_STORE_MANAGMENT, 1);
*cmds++ = adreno_dev->pix_shader_start;
*cmds++ = cp_type3_packet(CP_INVALIDATE_STATE, 1);
*cmds++ = 0x00003F00;
*cmds++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
*cmds++ = adreno_encode_istore_size(adreno_dev)
| adreno_dev->pix_shader_start;
cmds = program_shader(cmds, 0, gmem2sys_vtx_pgm, GMEM2SYS_VTX_PGM_LEN);
cmds =
program_shader(cmds, 1, gmem2sys_frag_pgm, GMEM2SYS_FRAG_PGM_LEN);
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_PROGRAM_CNTL);
if (adreno_is_a22x(adreno_dev))
*cmds++ = 0x10018001;
else
*cmds++ = 0x10010001;
*cmds++ = 0x00000008;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_VTE_CNTL);
*cmds++ = 0x00000b00;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_RB_SURFACE_INFO);
*cmds++ = shadow->gmem_pitch;
BUG_ON(tmp_ctx.gmem_base & 0xFFF);
*cmds++ =
(shadow->
format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT) | tmp_ctx.gmem_base;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_DEPTHCONTROL);
if (adreno_is_a22x(adreno_dev))
*cmds++ = 0x08;
else
*cmds++ = 0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_SU_SC_MODE_CNTL);
*cmds++ = 0x00080240;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_SCREEN_SCISSOR_TL);
*cmds++ = (0 << 16) | 0;
*cmds++ = (0x1fff << 16) | (0x1fff);
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_SC_WINDOW_SCISSOR_TL);
*cmds++ = (unsigned int)((1U << 31) | (0 << 16) | 0);
*cmds++ = (0x1fff << 16) | (0x1fff);
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_PA_CL_VPORT_ZSCALE);
*cmds++ = 0xbf800000;
*cmds++ = 0x0;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_MASK);
*cmds++ = 0x0000000f;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_COLOR_DEST_MASK);
*cmds++ = 0xffffffff;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 3);
*cmds++ = CP_REG(REG_SQ_WRAPPING_0);
*cmds++ = 0x00000000;
*cmds++ = 0x00000000;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 6);
*cmds++ = CP_REG(REG_RB_COPY_CONTROL);
*cmds++ = 0;
*cmds++ = addr & 0xfffff000;
*cmds++ = shadow->pitch >> 5;
*cmds++ = 0x0003c008 |
(shadow->format << RB_COPY_DEST_INFO__COPY_DEST_FORMAT__SHIFT);
BUG_ON(offset & 0xfffff000);
*cmds++ = offset;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_RB_MODECONTROL);
*cmds++ = 0x6;
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_PA_CL_CLIP_CNTL);
*cmds++ = 0x00010000;
if (adreno_is_a22x(adreno_dev)) {
*cmds++ = cp_type3_packet(CP_SET_CONSTANT, 2);
*cmds++ = CP_REG(REG_A220_RB_LRZ_VSC_CONTROL);
*cmds++ = 0x0000000;
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 3);
*cmds++ = 0;
*cmds++ = 0x00004088;
*cmds++ = 3;
} else {
*cmds++ = cp_type3_packet(CP_DRAW_INDX, 2);
*cmds++ = 0;
*cmds++ = 0x00030088;
}
create_ib1(drawctxt, shadow->gmem_save, start, cmds);
return cmds;
}
static void
build_shader_save_restore_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt)
{
unsigned int *cmd = tmp_ctx.cmd;
unsigned int *save, *restore, *fixup;
unsigned int *startSizeVtx, *startSizePix, *startSizeShared;
unsigned int *partition1;
unsigned int *shaderBases, *partition2;
tmp_ctx.shader_vertex = drawctxt->gpustate.gpuaddr + SHADER_OFFSET;
tmp_ctx.shader_pixel = tmp_ctx.shader_vertex
+ _shader_shadow_size(adreno_dev);
tmp_ctx.shader_shared = tmp_ctx.shader_pixel
+ _shader_shadow_size(adreno_dev);
restore = cmd;
*cmd++ = cp_type3_packet(CP_INVALIDATE_STATE, 1);
*cmd++ = 0x00000300;
*cmd++ = cp_type3_packet(CP_SET_SHADER_BASES, 1);
shaderBases = cmd++;
*cmd++ = cp_type0_packet(REG_SQ_INST_STORE_MANAGMENT, 1);
partition1 = cmd++;
*cmd++ = cp_type3_packet(CP_IM_LOAD, 2);
*cmd++ = tmp_ctx.shader_vertex + 0x0;
startSizeVtx = cmd++;
*cmd++ = cp_type3_packet(CP_IM_LOAD, 2);
*cmd++ = tmp_ctx.shader_pixel + 0x1;
startSizePix = cmd++;
*cmd++ = cp_type3_packet(CP_IM_LOAD, 2);
*cmd++ = tmp_ctx.shader_shared + 0x2;
startSizeShared = cmd++;
create_ib1(drawctxt, drawctxt->shader_restore, restore, cmd);
/*
* fixup SET_SHADER_BASES data
*
* since self-modifying PM4 code is being used here, a seperate
* command buffer is used for this fixup operation, to ensure the
* commands are not read by the PM4 engine before the data fields
* have been written.
*/
fixup = cmd;
*cmd++ = cp_type0_packet(REG_SCRATCH_REG2, 1);
partition2 = cmd++;
*cmd++ = cp_type3_packet(CP_REG_RMW, 3);
*cmd++ = REG_SCRATCH_REG2;
*cmd++ = 0x0FFF0FFF;
*cmd++ = adreno_encode_istore_size(adreno_dev);
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SCRATCH_REG2;
*cmd++ = virt2gpu(shaderBases, &drawctxt->gpustate);
create_ib1(drawctxt, drawctxt->shader_fixup, fixup, cmd);
save = cmd;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SQ_INST_STORE_MANAGMENT;
*cmd++ = virt2gpu(partition1, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SQ_INST_STORE_MANAGMENT;
*cmd++ = virt2gpu(partition2, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_IM_STORE, 2);
*cmd++ = tmp_ctx.shader_vertex + 0x0;
*cmd++ = virt2gpu(startSizeVtx, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_IM_STORE, 2);
*cmd++ = tmp_ctx.shader_pixel + 0x1;
*cmd++ = virt2gpu(startSizePix, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_IM_STORE, 2);
*cmd++ = tmp_ctx.shader_shared + 0x2;
*cmd++ = virt2gpu(startSizeShared, &drawctxt->gpustate);
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
create_ib1(drawctxt, drawctxt->shader_save, save, cmd);
tmp_ctx.cmd = cmd;
}
