[Qemu-devel] [PATCH 09/14] lm32: add Milkymist TMU2 support

[Michael Walle]

This patch adds support for Milkymist's texture mapping unit. For fast
computation this model needs hardware accelerated 3D graphics support
(OpenGL). There is no graphical output, all computations belong to internal
framebuffers only.

Signed-off-by: Michael Walle <address@hidden>
---
 Makefile.target     |    1 +
 hw/milkymist-tmu2.c |  481 +++++++++++++++++++++++++++++++++++++++++++++++++++
 trace-events        |    6 +
 3 files changed, 488 insertions(+), 0 deletions(-)
 create mode 100644 hw/milkymist-tmu2.c

diff --git a/Makefile.target b/Makefile.target
index 1c558a6..0aae6d0 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -263,6 +263,7 @@ obj-lm32-y += milkymist-minimac.o
 obj-lm32-y += milkymist-pfpu.o
 obj-lm32-y += milkymist-softusb.o
 obj-lm32-y += milkymist-sysctl.o
+obj-lm32-$(CONFIG_OPENGL) += milkymist-tmu2.o
 
 obj-mips-y = mips_r4k.o mips_jazz.o mips_malta.o mips_mipssim.o
 obj-mips-y += mips_addr.o mips_timer.o mips_int.o
diff --git a/hw/milkymist-tmu2.c b/hw/milkymist-tmu2.c
new file mode 100644
index 0000000..9cebe31
--- /dev/null
+++ b/hw/milkymist-tmu2.c
@@ -0,0 +1,481 @@
+/*
+ *  QEMU model of the Milkymist texture mapping unit.
+ *
+ *  Copyright (c) 2010 Michael Walle <address@hidden>
+ *  Copyright (c) 2010 Sebastien Bourdeauducq
+ *                       <address@hidden>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, see <http://www.gnu.org/licenses/>.
+ *
+ *
+ * Specification available at:
+ *   http://www.milkymist.org/socdoc/tmu2.pdf
+ *
+ */
+
+#include "hw.h"
+#include "sysbus.h"
+#include "trace.h"
+#include "qemu-error.h"
+
+#include <X11/Xlib.h>
+#include <GL/gl.h>
+#include <GL/glx.h>
+
+enum {
+    R_CTL = 0,
+    R_HMESHLAST,
+    R_VMESHLAST,
+    R_BRIGHTNESS,
+    R_CHROMAKEY,
+    R_VERTICESADDR,
+    R_TEXFBUF,
+    R_TEXHRES,
+    R_TEXVRES,
+    R_TEXHMASK,
+    R_TEXVMASK,
+    R_DSTFBUF,
+    R_DSTHRES,
+    R_DSTVRES,
+    R_DSTHOFFSET,
+    R_DSTVOFFSET,
+    R_DSTSQUAREW,
+    R_DSTSQUAREH,
+    R_ALPHA,
+    R_MAX
+};
+
+enum {
+    CTL_START_BUSY  = (1<<0),
+    CTL_CHROMAKEY   = (1<<1),
+};
+
+enum {
+    MAX_BRIGHTNESS = 63,
+    MAX_ALPHA      = 63,
+};
+
+enum {
+    MESH_MAXSIZE = 128,
+};
+
+struct vertex {
+    int x;
+    int y;
+} __attribute__((packed));
+
+struct MilkymistTMU2State {
+    SysBusDevice busdev;
+    CharDriverState *chr;
+    qemu_irq irq;
+
+    uint32_t regs[R_MAX];
+
+    Display *dpy;
+    GLXFBConfig glx_fb_config;
+    GLXContext glx_context;
+};
+typedef struct MilkymistTMU2State MilkymistTMU2State;
+
+static const int glx_fbconfig_attr[] = {
+    GLX_GREEN_SIZE, 5,
+    GLX_GREEN_SIZE, 6,
+    GLX_BLUE_SIZE, 5,
+    None
+};
+
+static int tmu2_glx_init(MilkymistTMU2State *s)
+{
+    GLXFBConfig *configs;
+    int nelements;
+
+    s->dpy = XOpenDisplay(NULL); /* FIXME: call XCloseDisplay() */
+    if (s->dpy == NULL) {
+        return 1;
+    }
+
+    configs = glXChooseFBConfig(s->dpy, 0, glx_fbconfig_attr, &nelements);
+    if (configs == NULL) {
+        return 1;
+    }
+
+    s->glx_fb_config = *configs;
+    XFree(configs);
+
+    /* FIXME: call glXDestroyContext() */
+    s->glx_context = glXCreateNewContext(s->dpy, s->glx_fb_config,
+            GLX_RGBA_TYPE, NULL, 1);
+    if (s->glx_context == NULL) {
+        return 1;
+    }
+
+    return 0;
+}
+
+static void tmu2_gl_map(struct vertex *mesh, int texhres, int texvres,
+        int hmeshlast, int vmeshlast, int ho, int vo, int sw, int sh)
+{
+    int x, y;
+    int x0, y0, x1, y1;
+    int u0, v0, u1, v1, u2, v2, u3, v3;
+    double xscale = 1.0 / ((double)(64 * texhres));
+    double yscale = 1.0 / ((double)(64 * texvres));
+
+    glLoadIdentity();
+    glTranslatef(ho, vo, 0);
+    glEnable(GL_TEXTURE_2D);
+    glBegin(GL_QUADS);
+
+    for (y = 0; y < vmeshlast; y++) {
+        y0 = y * sh;
+        y1 = y0 + sh;
+        for (x = 0; x < hmeshlast; x++) {
+            x0 = x * sw;
+            x1 = x0 + sw;
+
+            u0 = be32_to_cpu(mesh[MESH_MAXSIZE * y + x].x);
+            v0 = be32_to_cpu(mesh[MESH_MAXSIZE * y + x].y);
+            u1 = be32_to_cpu(mesh[MESH_MAXSIZE * y + x + 1].x);
+            v1 = be32_to_cpu(mesh[MESH_MAXSIZE * y + x + 1].y);
+            u2 = be32_to_cpu(mesh[MESH_MAXSIZE * (y + 1) + x + 1].x);
+            v2 = be32_to_cpu(mesh[MESH_MAXSIZE * (y + 1) + x + 1].y);
+            u3 = be32_to_cpu(mesh[MESH_MAXSIZE * (y + 1) + x].x);
+            v3 = be32_to_cpu(mesh[MESH_MAXSIZE * (y + 1) + x].y);
+
+            glTexCoord2d(((double)u0) * xscale, ((double)v0) * yscale);
+            glVertex3i(x0, y0, 0);
+            glTexCoord2d(((double)u1) * xscale, ((double)v1) * yscale);
+            glVertex3i(x1, y0, 0);
+            glTexCoord2d(((double)u2) * xscale, ((double)v2) * yscale);
+            glVertex3i(x1, y1, 0);
+            glTexCoord2d(((double)u3) * xscale, ((double)v3) * yscale);
+            glVertex3i(x0, y1, 0);
+        }
+    }
+
+    glEnd();
+}
+
+static void tmu2_start(MilkymistTMU2State *s)
+{
+    int pbuffer_attrib[6] = {
+        GLX_PBUFFER_WIDTH,
+        0,
+        GLX_PBUFFER_HEIGHT,
+        0,
+        GLX_PRESERVED_CONTENTS,
+        True
+    };
+
+    GLXPbuffer pbuffer;
+    GLuint texture;
+    void *fb;
+    target_phys_addr_t fb_len;
+    void *mesh;
+    target_phys_addr_t mesh_len;
+    float m;
+
+    trace_milkymist_tmu2_start();
+
+    /* Create and set up a suitable OpenGL context */
+    pbuffer_attrib[1] = s->regs[R_DSTHRES];
+    pbuffer_attrib[3] = s->regs[R_DSTVRES];
+    pbuffer = glXCreatePbuffer(s->dpy, s->glx_fb_config, pbuffer_attrib);
+    glXMakeContextCurrent(s->dpy, pbuffer, pbuffer, s->glx_context);
+
+    /* Fixup endianness. TODO: would it work on BE hosts? */
+    glPixelStorei(GL_UNPACK_SWAP_BYTES, 1);
+    glPixelStorei(GL_PACK_SWAP_BYTES, 1);
+
+    /* Row alignment */
+    glPixelStorei(GL_UNPACK_ALIGNMENT, 2);
+    glPixelStorei(GL_PACK_ALIGNMENT, 2);
+
+    /* Read the QEMU source framebuffer into an OpenGL texture */
+    glGenTextures(1, &texture);
+    glBindTexture(GL_TEXTURE_2D, texture);
+    fb_len = 2*s->regs[R_TEXHRES]*s->regs[R_TEXVRES];
+    fb = cpu_physical_memory_map(s->regs[R_TEXFBUF], &fb_len, 0);
+    if (fb == NULL) {
+        glDeleteTextures(1, &texture);
+        glXMakeContextCurrent(s->dpy, None, None, NULL);
+        glXDestroyPbuffer(s->dpy, pbuffer);
+        return;
+    }
+    glTexImage2D(GL_TEXTURE_2D, 0, 3, s->regs[R_TEXHRES], s->regs[R_TEXVRES],
+            0, GL_RGB, GL_UNSIGNED_SHORT_5_6_5, fb);
+    cpu_physical_memory_unmap(fb, fb_len, 0, fb_len);
+
+    /* Set up texturing options */
+    /* WARNING:
+     * Many cases of TMU2 masking are not supported by OpenGL.
+     * We only implement the most common ones:
+     *  - full bilinear filtering vs. nearest texel
+     *  - texture clamping vs. texture wrapping
+     */
+    if ((s->regs[R_TEXHMASK] & 0x3f) > 0x20) {
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+    } else {
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+    }
+    if ((s->regs[R_TEXHMASK] >> 6) & s->regs[R_TEXHRES]) {
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP);
+    } else {
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
+    }
+    if ((s->regs[R_TEXVMASK] >> 6) & s->regs[R_TEXVRES]) {
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP);
+    } else {
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
+    }
+
+    /* Translucency and decay */
+    glEnable(GL_BLEND);
+    glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
+    m = (float)(s->regs[R_BRIGHTNESS] + 1) / 64.0f;
+    glColor4f(m, m, m, (float)(s->regs[R_ALPHA] + 1) / 64.0f);
+
+    /* Read the QEMU dest. framebuffer into the OpenGL framebuffer */
+    fb_len = 2 * s->regs[R_DSTHRES] * s->regs[R_DSTVRES];
+    fb = cpu_physical_memory_map(s->regs[R_DSTFBUF], &fb_len, 0);
+    if (fb == NULL) {
+        glDeleteTextures(1, &texture);
+        glXMakeContextCurrent(s->dpy, None, None, NULL);
+        glXDestroyPbuffer(s->dpy, pbuffer);
+        return;
+    }
+
+    glDrawPixels(s->regs[R_DSTHRES], s->regs[R_DSTVRES], GL_RGB,
+            GL_UNSIGNED_SHORT_5_6_5, fb);
+    cpu_physical_memory_unmap(fb, fb_len, 0, fb_len);
+    glViewport(0, 0, s->regs[R_DSTHRES], s->regs[R_DSTVRES]);
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    glOrtho(0.0, s->regs[R_DSTHRES], 0.0, s->regs[R_DSTVRES], -1.0, 1.0);
+    glMatrixMode(GL_MODELVIEW);
+
+    /* Map the texture */
+    mesh_len = MESH_MAXSIZE*MESH_MAXSIZE*sizeof(struct vertex);
+    mesh = cpu_physical_memory_map(s->regs[R_VERTICESADDR], &mesh_len, 0);
+    if (mesh == NULL) {
+        glDeleteTextures(1, &texture);
+        glXMakeContextCurrent(s->dpy, None, None, NULL);
+        glXDestroyPbuffer(s->dpy, pbuffer);
+        return;
+    }
+
+    tmu2_gl_map((struct vertex *)mesh,
+        s->regs[R_TEXHRES], s->regs[R_TEXVRES],
+        s->regs[R_HMESHLAST], s->regs[R_VMESHLAST],
+        s->regs[R_DSTHOFFSET], s->regs[R_DSTVOFFSET],
+        s->regs[R_DSTSQUAREW], s->regs[R_DSTSQUAREH]);
+    cpu_physical_memory_unmap(mesh, mesh_len, 0, mesh_len);
+
+    /* Write back the OpenGL framebuffer to the QEMU framebuffer */
+    fb_len = 2 * s->regs[R_DSTHRES] * s->regs[R_DSTVRES];
+    fb = cpu_physical_memory_map(s->regs[R_DSTFBUF], &fb_len, 1);
+    if (fb == NULL) {
+        glDeleteTextures(1, &texture);
+        glXMakeContextCurrent(s->dpy, None, None, NULL);
+        glXDestroyPbuffer(s->dpy, pbuffer);
+        return;
+    }
+
+    glReadPixels(0, 0, s->regs[R_DSTHRES], s->regs[R_DSTVRES], GL_RGB,
+            GL_UNSIGNED_SHORT_5_6_5, fb);
+    cpu_physical_memory_unmap(fb, fb_len, 1, fb_len);
+
+    /* Free OpenGL allocs */
+    glDeleteTextures(1, &texture);
+    glXMakeContextCurrent(s->dpy, None, None, NULL);
+    glXDestroyPbuffer(s->dpy, pbuffer);
+
+    s->regs[R_CTL] &= ~CTL_START_BUSY;
+
+    trace_milkymist_tmu2_pulse_irq();
+    qemu_irq_pulse(s->irq);
+}
+
+static uint32_t tmu2_read(void *opaque, target_phys_addr_t addr)
+{
+    MilkymistTMU2State *s = opaque;
+    uint32_t r = 0;
+
+    addr >>= 2;
+    switch (addr) {
+    case R_CTL:
+    case R_HMESHLAST:
+    case R_VMESHLAST:
+    case R_BRIGHTNESS:
+    case R_CHROMAKEY:
+    case R_VERTICESADDR:
+    case R_TEXFBUF:
+    case R_TEXHRES:
+    case R_TEXVRES:
+    case R_TEXHMASK:
+    case R_TEXVMASK:
+    case R_DSTFBUF:
+    case R_DSTHRES:
+    case R_DSTVRES:
+    case R_DSTHOFFSET:
+    case R_DSTVOFFSET:
+    case R_DSTSQUAREW:
+    case R_DSTSQUAREH:
+    case R_ALPHA:
+        r = s->regs[addr];
+        break;
+
+    default:
+        error_report("milkymist_tmu2: read access to unknown register 0x"
+                TARGET_FMT_plx, addr << 2);
+        break;
+    }
+
+    trace_milkymist_tmu2_memory_read(addr << 2, r);
+
+    return r;
+}
+
+static void tmu2_check_registers(MilkymistTMU2State *s)
+{
+    if (s->regs[R_BRIGHTNESS] > MAX_BRIGHTNESS) {
+        error_report("milkymist_tmu2: max brightness is %d\n", MAX_BRIGHTNESS);
+    }
+
+    if (s->regs[R_ALPHA] > MAX_ALPHA) {
+        error_report("milkymist_tmu2: max alpha is %d\n", MAX_ALPHA);
+    }
+
+    if (s->regs[R_VERTICESADDR] & 0x07) {
+        error_report("milkymist_tmu2: vertex mesh address has to be 64-bit "
+                "aligned\n");
+    }
+
+    if (s->regs[R_TEXFBUF] & 0x01) {
+        error_report("milkymist_tmu2: texture buffer address has to be "
+                "16-bit aligned\n");
+    }
+}
+
+static void tmu2_write(void *opaque, target_phys_addr_t addr, uint32_t value)
+{
+    MilkymistTMU2State *s = opaque;
+
+    trace_milkymist_tmu2_memory_write(addr, value);
+
+    addr >>= 2;
+    switch (addr) {
+    case R_CTL:
+        s->regs[addr] = value;
+        if (value & CTL_START_BUSY) {
+            tmu2_start(s);
+        }
+        break;
+    case R_BRIGHTNESS:
+    case R_HMESHLAST:
+    case R_VMESHLAST:
+    case R_CHROMAKEY:
+    case R_VERTICESADDR:
+    case R_TEXFBUF:
+    case R_TEXHRES:
+    case R_TEXVRES:
+    case R_TEXHMASK:
+    case R_TEXVMASK:
+    case R_DSTFBUF:
+    case R_DSTHRES:
+    case R_DSTVRES:
+    case R_DSTHOFFSET:
+    case R_DSTVOFFSET:
+    case R_DSTSQUAREW:
+    case R_DSTSQUAREH:
+    case R_ALPHA:
+        s->regs[addr] = value;
+        break;
+
+    default:
+        error_report("milkymist_tmu2: write access to unknown register 0x"
+                TARGET_FMT_plx, addr << 2);
+        break;
+    }
+
+    tmu2_check_registers(s);
+}
+
+static CPUReadMemoryFunc * const tmu2_read_fn[] = {
+    NULL,
+    NULL,
+    &tmu2_read,
+};
+
+static CPUWriteMemoryFunc * const tmu2_write_fn[] = {
+    NULL,
+    NULL,
+    &tmu2_write,
+};
+
+static void milkymist_tmu2_reset(DeviceState *d)
+{
+    MilkymistTMU2State *s = container_of(d, MilkymistTMU2State, busdev.qdev);
+    int i;
+
+    for (i = 0; i < R_MAX; i++) {
+        s->regs[i] = 0;
+    }
+}
+
+static int milkymist_tmu2_init(SysBusDevice *dev)
+{
+    MilkymistTMU2State *s = FROM_SYSBUS(typeof(*s), dev);
+    int tmu2_regs;
+
+    if (tmu2_glx_init(s)) {
+        return 1;
+    }
+
+    sysbus_init_irq(dev, &s->irq);
+
+    tmu2_regs = cpu_register_io_memory(tmu2_read_fn, tmu2_write_fn, s,
+            DEVICE_NATIVE_ENDIAN);
+    sysbus_init_mmio(dev, R_MAX * 4, tmu2_regs);
+
+    return 0;
+}
+
+static const VMStateDescription vmstate_milkymist_tmu2 = {
+    .name = "milkymist-tmu2",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .minimum_version_id_old = 1,
+    .fields      = (VMStateField[]) {
+        VMSTATE_UINT32_ARRAY(regs, MilkymistTMU2State, R_MAX),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static SysBusDeviceInfo milkymist_tmu2_info = {
+    .init = milkymist_tmu2_init,
+    .qdev.name  = "milkymist-tmu2",
+    .qdev.size  = sizeof(MilkymistTMU2State),
+    .qdev.vmsd  = &vmstate_milkymist_tmu2,
+    .qdev.reset = milkymist_tmu2_reset,
+};
+
+static void milkymist_tmu2_register(void)
+{
+    sysbus_register_withprop(&milkymist_tmu2_info);
+}
+
+device_init(milkymist_tmu2_register)
diff --git a/trace-events b/trace-events
index 02ea2e8..85b6b21 100644
--- a/trace-events
+++ b/trace-events
@@ -340,3 +340,9 @@ disable milkymist_sysctl_start_timer1(void) "Start timer1"
 disable milkymist_sysctl_stop_timer1(void) "Stop timer1"
 disable milkymist_sysctl_pulse_irq_timer0(void) "Pulse IRQ Timer0"
 disable milkymist_sysctl_pulse_irq_timer1(void) "Pulse IRQ Timer1"
+
+# hw/milkymist-tmu2.c
+disable milkymist_tmu2_memory_read(uint32_t addr, uint32_t value) "addr %08x 
value %08x"
+disable milkymist_tmu2_memory_write(uint32_t addr, uint32_t value) "addr %08x 
value %08x"
+disable milkymist_tmu2_start(void) "Start TMU"
+disable milkymist_tmu2_pulse_irq(void) "Pulse IRQ"
-- 
1.7.2.3