X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fr300%2Fr300_ioctl.c;h=3303078e39abfd455f4592c4347412cb4d240871;hb=97029c1860e8a99448ac157408e71bea3aa2559f;hp=d1cf57959fad2db52dcc510082b4a3f5b860dd07;hpb=2d26d4ac66e8c3e48b73d3e172d0e0d2a2ad31a5;p=mesa.git diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c index d1cf57959fa..3303078e39a 100644 --- a/src/mesa/drivers/dri/r300/r300_ioctl.c +++ b/src/mesa/drivers/dri/r300/r300_ioctl.c @@ -44,6 +44,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/imports.h" #include "main/macros.h" #include "main/context.h" +#include "main/simple_list.h" #include "swrast/swrast.h" #include "radeon_common.h" @@ -55,7 +56,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_vertprog.h" #include "radeon_reg.h" #include "r300_emit.h" -#include "r300_fragprog.h" #include "r300_context.h" #include "vblank.h" @@ -66,13 +66,75 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define CLEARBUFFER_DEPTH 0x2 #define CLEARBUFFER_STENCIL 0x4 +#if 1 + +/** + * Fragment program helper macros + */ + +/* Produce unshifted source selectors */ +#define FP_TMP(idx) (idx) +#define FP_CONST(idx) ((idx) | (1 << 5)) + +/* Produce source/dest selector dword */ +#define FP_SELC_MASK_NO 0 +#define FP_SELC_MASK_X 1 +#define FP_SELC_MASK_Y 2 +#define FP_SELC_MASK_XY 3 +#define FP_SELC_MASK_Z 4 +#define FP_SELC_MASK_XZ 5 +#define FP_SELC_MASK_YZ 6 +#define FP_SELC_MASK_XYZ 7 + +#define FP_SELC(destidx,regmask,outmask,src0,src1,src2) \ + (((destidx) << R300_ALU_DSTC_SHIFT) | \ + (FP_SELC_MASK_##regmask << 23) | \ + (FP_SELC_MASK_##outmask << 26) | \ + ((src0) << R300_ALU_SRC0C_SHIFT) | \ + ((src1) << R300_ALU_SRC1C_SHIFT) | \ + ((src2) << R300_ALU_SRC2C_SHIFT)) + +#define FP_SELA_MASK_NO 0 +#define FP_SELA_MASK_W 1 + +#define FP_SELA(destidx,regmask,outmask,src0,src1,src2) \ + (((destidx) << R300_ALU_DSTA_SHIFT) | \ + (FP_SELA_MASK_##regmask << 23) | \ + (FP_SELA_MASK_##outmask << 24) | \ + ((src0) << R300_ALU_SRC0A_SHIFT) | \ + ((src1) << R300_ALU_SRC1A_SHIFT) | \ + ((src2) << R300_ALU_SRC2A_SHIFT)) + +/* Produce unshifted argument selectors */ +#define FP_ARGC(source) R300_ALU_ARGC_##source +#define FP_ARGA(source) R300_ALU_ARGA_##source +#define FP_ABS(arg) ((arg) | (1 << 6)) +#define FP_NEG(arg) ((arg) ^ (1 << 5)) + +/* Produce instruction dword */ +#define FP_INSTRC(opcode,arg0,arg1,arg2) \ + (R300_ALU_OUTC_##opcode | \ + ((arg0) << R300_ALU_ARG0C_SHIFT) | \ + ((arg1) << R300_ALU_ARG1C_SHIFT) | \ + ((arg2) << R300_ALU_ARG2C_SHIFT)) + +#define FP_INSTRA(opcode,arg0,arg1,arg2) \ + (R300_ALU_OUTA_##opcode | \ + ((arg0) << R300_ALU_ARG0A_SHIFT) | \ + ((arg1) << R300_ALU_ARG1A_SHIFT) | \ + ((arg2) << R300_ALU_ARG2A_SHIFT)) + +#endif + +static void r300EmitClearState(GLcontext * ctx); + static void r300ClearBuffer(r300ContextPtr r300, int flags, struct radeon_renderbuffer *rrb, struct radeon_renderbuffer *rrbd) { BATCH_LOCALS(&r300->radeon); GLcontext *ctx = r300->radeon.glCtx; - __DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); GLuint cbpitch = 0; r300ContextPtr rmesa = r300; @@ -107,18 +169,21 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, } #if 1 if (flags & (CLEARBUFFER_DEPTH | CLEARBUFFER_STENCIL)) { - assert(rrbd != 0); - cbpitch = (rrbd->pitch / rrbd->cpp); + uint32_t zbpitch = (rrbd->pitch / rrbd->cpp); if (rrbd->bo->flags & RADEON_BO_FLAGS_MACRO_TILE){ - cbpitch |= R300_DEPTHMACROTILE_ENABLE; + zbpitch |= R300_DEPTHMACROTILE_ENABLE; } if (rrbd->bo->flags & RADEON_BO_FLAGS_MICRO_TILE){ - cbpitch |= R300_DEPTHMICROTILE_TILED; + zbpitch |= R300_DEPTHMICROTILE_TILED; } BEGIN_BATCH_NO_AUTOSTATE(6); OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1); OUT_BATCH_RELOC(0, rrbd->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); - OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, cbpitch); + OUT_BATCH_REGSEQ(R300_ZB_DEPTHPITCH, 1); + if (!r300->radeon.radeonScreen->kernel_mm) + OUT_BATCH(zbpitch); + else + OUT_BATCH_RELOC(zbpitch, rrbd->bo, zbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); END_BATCH(); } #endif @@ -193,7 +258,7 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, OUT_BATCH_FLOAT32(ctx->Color.ClearColor[2]); OUT_BATCH_FLOAT32(ctx->Color.ClearColor[3]); } - + r300EmitCacheFlush(rmesa); cp_wait(&r300->radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN); @@ -206,14 +271,13 @@ static void r300EmitClearState(GLcontext * ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); BATCH_LOCALS(&r300->radeon); - __DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); int i; - int has_tcl = 1; + int has_tcl; int is_r500 = 0; GLuint vap_cntl; - if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) - has_tcl = 0; + has_tcl = r300->options.hw_tcl_enabled; if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) is_r500 = 1; @@ -441,9 +505,17 @@ static void r300EmitClearState(GLcontext * ctx) R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 | R500_ALU_RGBA_A_SWIZ_0; - + r500fp.cmd[7] = 0; - emit_r500fp(ctx, &r500fp); + if (r300->radeon.radeonScreen->kernel_mm) { + emit_r500fp(ctx, &r500fp); + } else { + int dwords = r500fp.check(ctx,&r500fp); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(r500fp.cmd, dwords); + END_BATCH(); + } + } BEGIN_BATCH(2); @@ -487,6 +559,7 @@ static void r300EmitClearState(GLcontext * ctx) struct radeon_state_atom vpu; uint32_t _cmd[10]; R300_STATECHANGE(r300, pvs); + R300_STATECHANGE(r300, vap_flush); R300_STATECHANGE(r300, vpi); BEGIN_BATCH(4); @@ -507,12 +580,12 @@ static void r300EmitClearState(GLcontext * ctx) 0, 0xf, PVS_DST_REG_OUT); vpu.cmd[2] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, - PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[3] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, - PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[4] = 0x0; vpu.cmd[5] = PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 1, 0xf, @@ -520,90 +593,95 @@ static void r300EmitClearState(GLcontext * ctx) vpu.cmd[6] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, - - VSF_FLAG_NONE); + NEGATE_NONE); vpu.cmd[7] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, - PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[8] = 0x0; - r300->vap_flush_needed = GL_TRUE; - emit_vpu(ctx, &vpu); + if (r300->radeon.radeonScreen->kernel_mm) { + int dwords = r300->hw.vap_flush.check(ctx,&r300->hw.vap_flush); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(r300->hw.vap_flush.cmd, dwords); + END_BATCH(); + emit_vpu(ctx, &vpu); + } else { + int dwords = vpu.check(ctx,&vpu); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(vpu.cmd, dwords); + END_BATCH(); + } + } } -/** - * Buffer clear - */ -static void r300Clear(GLcontext * ctx, GLbitfield mask) +static int r300KernelClear(GLcontext *ctx, GLuint flags) { r300ContextPtr r300 = R300_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); struct radeon_framebuffer *rfb = dPriv->driverPrivate; struct radeon_renderbuffer *rrb; struct radeon_renderbuffer *rrbd; - int flags = 0; - int bits = 0; - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "r300Clear\n"); + int bits = 0, ret; - { - LOCK_HARDWARE(&r300->radeon); - UNLOCK_HARDWARE(&r300->radeon); - if (dPriv->numClipRects == 0) - return; - } - - /* Flush swtcl vertices if necessary, because we will change hardware - * state during clear. See also the state-related comment in - * r300EmitClearState. - */ - R300_NEWPRIM(r300); + /* Make sure it fits there. */ + radeon_cs_space_reset_bos(r300->radeon.cmdbuf.cs); - if (mask & BUFFER_BIT_FRONT_LEFT) { - flags |= BUFFER_BIT_FRONT_LEFT; - mask &= ~BUFFER_BIT_FRONT_LEFT; + if (flags & BUFFER_BIT_COLOR0) { + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_COLOR0); + radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, + rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM); } - if (mask & BUFFER_BIT_BACK_LEFT) { - flags |= BUFFER_BIT_BACK_LEFT; - mask &= ~BUFFER_BIT_BACK_LEFT; + if (flags & BUFFER_BIT_FRONT_LEFT) { + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_FRONT_LEFT); + radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, + rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM); } - if (mask & BUFFER_BIT_DEPTH) { - bits |= CLEARBUFFER_DEPTH; - mask &= ~BUFFER_BIT_DEPTH; + if (flags & BUFFER_BIT_BACK_LEFT) { + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_BACK_LEFT); + radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, + rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM); } - if ((mask & BUFFER_BIT_STENCIL) && r300->radeon.state.stencil.hwBuffer) { - bits |= CLEARBUFFER_STENCIL; - mask &= ~BUFFER_BIT_STENCIL; + rrbd = radeon_get_renderbuffer(&rfb->base, BUFFER_DEPTH); + if (rrbd) { + radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, + rrbd->bo, 0, RADEON_GEM_DOMAIN_VRAM); } - if (mask) { - if (RADEON_DEBUG & DEBUG_FALLBACKS) - fprintf(stderr, "%s: swrast clear, mask: %x\n", - __FUNCTION__, mask); - _swrast_Clear(ctx, mask); - } + ret = radeon_cs_space_check(r300->radeon.cmdbuf.cs); + if (ret) + return -1; - /* Make sure it fits there. */ rcommonEnsureCmdBufSpace(&r300->radeon, 421 * 3, __FUNCTION__); if (flags || bits) r300EmitClearState(ctx); - rrbd = (void *)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer; + + rrbd = radeon_get_renderbuffer(&rfb->base, BUFFER_DEPTH); + if (rrbd && (flags & BUFFER_BIT_DEPTH)) + bits |= CLEARBUFFER_DEPTH; + + if (rrbd && (flags & BUFFER_BIT_STENCIL)) + bits |= CLEARBUFFER_STENCIL; + + if (flags & BUFFER_BIT_COLOR0) { + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_COLOR0); + r300ClearBuffer(r300, CLEARBUFFER_COLOR, rrb, NULL); + bits = 0; + } if (flags & BUFFER_BIT_FRONT_LEFT) { - rrb = (void *)rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer; + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_FRONT_LEFT); r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, rrb, rrbd); bits = 0; } if (flags & BUFFER_BIT_BACK_LEFT) { - rrb = (void *)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer; + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_BACK_LEFT); r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, rrb, rrbd); bits = 0; } @@ -612,6 +690,88 @@ static void r300Clear(GLcontext * ctx, GLbitfield mask) r300ClearBuffer(r300, bits, NULL, rrbd); COMMIT_BATCH(); + return 0; +} + +/** + * Buffer clear + */ +static void r300Clear(GLcontext * ctx, GLbitfield mask) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); + const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask); + GLbitfield swrast_mask = 0, tri_mask = 0; + int i, ret; + struct gl_framebuffer *fb = ctx->DrawBuffer; + + if (RADEON_DEBUG & DEBUG_IOCTL) + fprintf(stderr, "r300Clear\n"); + + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + LOCK_HARDWARE(&r300->radeon); + UNLOCK_HARDWARE(&r300->radeon); + if (dPriv->numClipRects == 0) + return; + } + + /* Flush swtcl vertices if necessary, because we will change hardware + * state during clear. See also the state-related comment in + * r300EmitClearState. + */ + R300_NEWPRIM(r300); + + if (colorMask == ~0) + tri_mask |= (mask & BUFFER_BITS_COLOR); + else + tri_mask |= (mask & (BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_BACK_LEFT)); + + + /* HW stencil */ + if (mask & BUFFER_BIT_STENCIL) { + tri_mask |= BUFFER_BIT_STENCIL; + } + + /* HW depth */ + if (mask & BUFFER_BIT_DEPTH) { + tri_mask |= BUFFER_BIT_DEPTH; + } + + /* If we're doing a tri pass for depth/stencil, include a likely color + * buffer with it. + */ + + for (i = 0; i < BUFFER_COUNT; i++) { + GLuint bufBit = 1 << i; + if ((tri_mask) & bufBit) { + if (!fb->Attachment[i].Renderbuffer->ClassID) { + tri_mask &= ~bufBit; + swrast_mask |= bufBit; + } + } + } + + /* SW fallback clearing */ + swrast_mask = mask & ~tri_mask; + + ret = 0; + if (tri_mask) { + if (r300->radeon.radeonScreen->kernel_mm) + radeonUserClear(ctx, tri_mask); + else { + /* if kernel clear fails due to size restraints fallback */ + ret = r300KernelClear(ctx, tri_mask); + if (ret < 0) + swrast_mask |= tri_mask; + } + } + + if (swrast_mask) { + if (RADEON_DEBUG & DEBUG_FALLBACKS) + fprintf(stderr, "%s: swrast clear, mask: %x\n", + __FUNCTION__, swrast_mask); + _swrast_Clear(ctx, swrast_mask); + } } void r300InitIoctlFuncs(struct dd_function_table *functions)